path: root/nova/compute/manager.py

# Copyright 2010 United States Government as represented by the
# Administrator of the National Aeronautics and Space Administration.
# Copyright 2011 Justin Santa Barbara
# All Rights Reserved.
#
#    Licensed under the Apache License, Version 2.0 (the "License"); you may
#    not use this file except in compliance with the License. You may obtain
#    a copy of the License at
#
#         http://www.apache.org/licenses/LICENSE-2.0
#
#    Unless required by applicable law or agreed to in writing, software
#    distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
#    WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
#    License for the specific language governing permissions and limitations
#    under the License.

"""Handles all processes relating to instances (guest vms).

The :py:class:`ComputeManager` class is a :py:class:`nova.manager.Manager` that
handles RPC calls relating to creating instances.  It is responsible for
building a disk image, launching it via the underlying virtualization driver,
responding to calls to check its state, attaching persistent storage, and
terminating it.

"""

import base64
import binascii
import contextlib
import copy
import functools
import inspect
import math
import sys
import time
import traceback
import typing as ty

from cinderclient import exceptions as cinder_exception
from cursive import exception as cursive_exception
import eventlet.event
from eventlet import greenthread
import eventlet.semaphore
import eventlet.timeout
import futurist
from keystoneauth1 import exceptions as keystone_exception
import os_traits
from oslo_log import log as logging
import oslo_messaging as messaging
from oslo_serialization import jsonutils
from oslo_service import loopingcall
from oslo_service import periodic_task
from oslo_utils import excutils
from oslo_utils import strutils
from oslo_utils import timeutils
from oslo_utils import units

from nova.accelerator import cyborg
from nova import block_device
from nova.compute import api as compute
from nova.compute import build_results
from nova.compute import claims
from nova.compute import power_state
from nova.compute import resource_tracker
from nova.compute import rpcapi as compute_rpcapi
from nova.compute import task_states
from nova.compute import utils as compute_utils
from nova.compute.utils import wrap_instance_event
from nova.compute import vm_states
from nova import conductor
import nova.conf
import nova.context
from nova import exception
from nova import exception_wrapper
from nova.i18n import _
from nova.image import glance
from nova import manager
from nova.network import model as network_model
from nova.network import neutron
from nova import objects
from nova.objects import base as obj_base
from nova.objects import external_event as external_event_obj
from nova.objects import fields
from nova.objects import instance as obj_instance
from nova.objects import migrate_data as migrate_data_obj
from nova.pci import request as pci_req_module
from nova.pci import whitelist
from nova import safe_utils
from nova.scheduler.client import query
from nova.scheduler.client import report
from nova.scheduler import utils as scheduler_utils
from nova import utils
from nova.virt import block_device as driver_block_device
from nova.virt import configdrive
from nova.virt import driver
from nova.virt import event as virtevent
from nova.virt import hardware
from nova.virt import storage_users
from nova.virt import virtapi
from nova.volume import cinder

CONF = nova.conf.CONF

LOG = logging.getLogger(__name__)

wrap_exception = functools.partial(
    exception_wrapper.wrap_exception, service='compute', binary='nova-compute')


@contextlib.contextmanager
def errors_out_migration_ctxt(migration):
    """Context manager to error out migration on failure."""

    try:
        yield
    except Exception:
        with excutils.save_and_reraise_exception():
            if migration:
                # We may have been passed None for our migration if we're
                # receiving from an older client. The migration will be
                # errored via the legacy path.
                migration.status = 'error'
                try:
                    migration.save()
                except Exception:
                    LOG.debug(
                        'Error setting migration status for instance %s.',
                        migration.instance_uuid, exc_info=True)


@utils.expects_func_args('migration')
def errors_out_migration(function):
    """Decorator to error out migration on failure."""

    @functools.wraps(function)
    def decorated_function(self, context, *args, **kwargs):
        wrapped_func = safe_utils.get_wrapped_function(function)
        keyed_args = inspect.getcallargs(wrapped_func, self, context,
                                         *args, **kwargs)
        migration = keyed_args['migration']
        with errors_out_migration_ctxt(migration):
            return function(self, context, *args, **kwargs)

    return decorated_function


@utils.expects_func_args('instance')
def reverts_task_state(function):
    """Decorator to revert task_state on failure."""

    @functools.wraps(function)
    def decorated_function(self, context, *args, **kwargs):
        try:
            return function(self, context, *args, **kwargs)
        except exception.UnexpectedTaskStateError as e:
            # Note(maoy): unexpected task state means the current
            # task is preempted. Do not clear task state in this
            # case.
            with excutils.save_and_reraise_exception():
                LOG.info("Task possibly preempted: %s",
                         e.format_message())
        except Exception:
            with excutils.save_and_reraise_exception():
                wrapped_func = safe_utils.get_wrapped_function(function)
                keyed_args = inspect.getcallargs(wrapped_func, self, context,
                                                 *args, **kwargs)
                # NOTE(mriedem): 'instance' must be in keyed_args because we
                # have utils.expects_func_args('instance') decorating this
                # method.
                instance = keyed_args['instance']
                original_task_state = instance.task_state
                try:
                    self._instance_update(context, instance, task_state=None)
                    LOG.info("Successfully reverted task state from %s on "
                             "failure for instance.",
                             original_task_state, instance=instance)
                except exception.InstanceNotFound:
                    # We might delete an instance that failed to build shortly
                    # after it errored out this is an expected case and we
                    # should not trace on it.
                    pass
                except Exception as e:
                    LOG.warning("Failed to revert task state for instance. "
                                "Error: %s", e, instance=instance)

    return decorated_function


@utils.expects_func_args('instance')
def wrap_instance_fault(function):
    """Wraps a method to catch exceptions related to instances.

    This decorator wraps a method to catch any exceptions having to do with
    an instance that may get thrown. It then logs an instance fault in the db.
    """

    @functools.wraps(function)
    def decorated_function(self, context, *args, **kwargs):
        try:
            return function(self, context, *args, **kwargs)
        except exception.InstanceNotFound:
            raise
        except Exception as e:
            # NOTE(gtt): If argument 'instance' is in args rather than kwargs,
            # we will get a KeyError exception which will cover up the real
            # exception. So, we update kwargs with the values from args first.
            # then, we can get 'instance' from kwargs easily.
            kwargs.update(dict(zip(function.__code__.co_varnames[2:], args)))

            with excutils.save_and_reraise_exception():
                compute_utils.add_instance_fault_from_exc(context,
                        kwargs['instance'], e, sys.exc_info())

    return decorated_function


@utils.expects_func_args('image_id', 'instance')
def delete_image_on_error(function):
    """Used for snapshot related method to ensure the image created in
    compute.api is deleted when an error occurs.
    """

    @functools.wraps(function)
    def decorated_function(self, context, image_id, instance,
                           *args, **kwargs):
        try:
            return function(self, context, image_id, instance,
                            *args, **kwargs)
        except Exception:
            with excutils.save_and_reraise_exception():
                compute_utils.delete_image(
                    context, instance, self.image_api, image_id,
                    log_exc_info=True)

    return decorated_function


# Each collection of events is a dict of eventlet Events keyed by a tuple of
# event name and associated tag
_InstanceEvents = ty.Dict[ty.Tuple[str, str], eventlet.event.Event]


class InstanceEvents(object):
    def __init__(self):
        self._events: ty.Optional[ty.Dict[str, _InstanceEvents]] = {}

    @staticmethod
    def _lock_name(instance) -> str:
        return '%s-%s' % (instance.uuid, 'events')

    def prepare_for_instance_event(
        self,
        instance: 'objects.Instance',
        name: str,
        tag: str,
    ) -> eventlet.event.Event:
        """Prepare to receive an event for an instance.

        This will register an event for the given instance that we will
        wait on later. This should be called before initiating whatever
        action will trigger the event. The resulting eventlet.event.Event
        object should be wait()'d on to ensure completion.

        :param instance: the instance for which the event will be generated
        :param name: the name of the event we're expecting
        :param tag: the tag associated with the event we're expecting
        :returns: an event object that should be wait()'d on
        """
        @utils.synchronized(self._lock_name(instance))
        def _create_or_get_event():
            if self._events is None:
                # NOTE(danms): We really should have a more specific error
                # here, but this is what we use for our default error case
                raise exception.NovaException(
                    'In shutdown, no new events can be scheduled')

            instance_events = self._events.setdefault(instance.uuid, {})
            return instance_events.setdefault((name, tag),
                                              eventlet.event.Event())
        LOG.debug('Preparing to wait for external event %(name)s-%(tag)s',
                  {'name': name, 'tag': tag}, instance=instance)
        return _create_or_get_event()

    def pop_instance_event(self, instance, event):
        """Remove a pending event from the wait list.

        This will remove a pending event from the wait list so that it
        can be used to signal the waiters to wake up.

        :param instance: the instance for which the event was generated
        :param event: the nova.objects.external_event.InstanceExternalEvent
                      that describes the event
        :returns: the eventlet.event.Event object on which the waiters
                  are blocked
        """
        no_events_sentinel = object()
        no_matching_event_sentinel = object()

        @utils.synchronized(self._lock_name(instance))
        def _pop_event():
            if self._events is None:
                LOG.debug('Unexpected attempt to pop events during shutdown',
                          instance=instance)
                return no_events_sentinel
            events = self._events.get(instance.uuid)
            if not events:
                return no_events_sentinel
            _event = events.pop((event.name, event.tag), None)
            if not events:
                del self._events[instance.uuid]
            if _event is None:
                return no_matching_event_sentinel
            return _event

        result = _pop_event()
        if result is no_events_sentinel:
            LOG.debug('No waiting events found dispatching %(event)s',
                      {'event': event.key},
                      instance=instance)
            return None
        elif result is no_matching_event_sentinel:
            LOG.debug(
                'No event matching %(event)s in %(events)s',
                {
                    'event': event.key,
                    # mypy can't identify the none check in _pop_event
                    'events': self._events.get(  # type: ignore
                        instance.uuid, {}).keys(),
                },
                instance=instance,
            )
            return None
        else:
            return result

    def clear_events_for_instance(self, instance):
        """Remove all pending events for an instance.

        This will remove all events currently pending for an instance
        and return them (indexed by event name).

        :param instance: the instance for which events should be purged
        :returns: a dictionary of {event_name: eventlet.event.Event}
        """
        @utils.synchronized(self._lock_name(instance))
        def _clear_events():
            if self._events is None:
                LOG.debug('Unexpected attempt to clear events during shutdown',
                          instance=instance)
                return dict()
            # NOTE(danms): We have historically returned the raw internal
            # format here, which is {event.key: [events, ...])} so just
            # trivially convert it here.
            return {'%s-%s' % k: e
                    for k, e in self._events.pop(instance.uuid, {}).items()}
        return _clear_events()

    def cancel_all_events(self):
        if self._events is None:
            LOG.debug('Unexpected attempt to cancel events during shutdown.')
            return
        our_events = self._events
        # NOTE(danms): Block new events
        self._events = None

        for instance_uuid, events in our_events.items():
            for (name, tag), eventlet_event in events.items():
                LOG.debug('Canceling in-flight event %(name)s-%(tag)s for '
                          'instance %(instance_uuid)s',
                          {'name': name,
                           'tag': tag,
                           'instance_uuid': instance_uuid})
                event = objects.InstanceExternalEvent(
                    instance_uuid=instance_uuid,
                    name=name, status='failed',
                    tag=tag, data={})
                eventlet_event.send(event)


class ComputeVirtAPI(virtapi.VirtAPI):
    def __init__(self, compute):
        super(ComputeVirtAPI, self).__init__()
        self._compute = compute
        self.reportclient = compute.reportclient

        class ExitEarly(Exception):
            def __init__(self, events):
                super(Exception, self).__init__()
                self.events = events

        self._exit_early_exc = ExitEarly

    def exit_wait_early(self, events):
        """Exit a wait_for_instance_event() immediately and avoid
        waiting for some events.

        :param: events: A list of (name, tag) tuples for events that we should
                        skip waiting for during a wait_for_instance_event().
        """
        raise self._exit_early_exc(events=events)

    def _default_error_callback(self, event_name, instance):
        raise exception.NovaException(_('Instance event failed'))

    class _InstanceEvent:
        EXPECTED = "expected"
        WAITING = "waiting"
        RECEIVED = "received"
        RECEIVED_EARLY = "received early"
        TIMED_OUT = "timed out"
        RECEIVED_NOT_PROCESSED = "received but not processed"

        def __init__(self, name: str, event: eventlet.event.Event) -> None:
            self.name = name
            self.event = event
            self.status = self.EXPECTED
            self.wait_time = None

        def mark_as_received_early(self) -> None:
            self.status = self.RECEIVED_EARLY

        def is_received_early(self) -> bool:
            return self.status == self.RECEIVED_EARLY

        def _update_status_no_wait(self):
            if self.status == self.EXPECTED and self.event.ready():
                self.status = self.RECEIVED_NOT_PROCESSED

        def wait(self) -> 'objects.InstanceExternalEvent':
            self.status = self.WAITING
            try:
                with timeutils.StopWatch() as sw:
                    instance_event = self.event.wait()
            except eventlet.timeout.Timeout:
                self.status = self.TIMED_OUT
                self.wait_time = sw.elapsed()

                raise

            self.status = self.RECEIVED
            self.wait_time = sw.elapsed()
            return instance_event

        def __str__(self) -> str:
            self._update_status_no_wait()
            if self.status == self.EXPECTED:
                return f"{self.name}: expected but not received"
            if self.status == self.RECEIVED:
                return (
                    f"{self.name}: received after waiting "
                    f"{self.wait_time:.2f} seconds")
            if self.status == self.TIMED_OUT:
                return (
                    f"{self.name}: timed out after "
                    f"{self.wait_time:.2f} seconds")
            return f"{self.name}: {self.status}"

    @staticmethod
    def _wait_for_instance_events(
        instance: 'objects.Instance',
        events: dict,
        error_callback: ty.Callable,
    ) -> None:
        for event_name, event in events.items():
            if event.is_received_early():
                continue
            else:
                actual_event = event.wait()
                if actual_event.status == 'completed':
                    continue
            # If we get here, we have an event that was not completed,
            # nor skipped via exit_wait_early(). Decide whether to
            # keep waiting by calling the error_callback() hook.
            decision = error_callback(event_name, instance)
            if decision is False:
                break

    @contextlib.contextmanager
    def wait_for_instance_event(self, instance, event_names, deadline=300,
                                error_callback=None):
        """Plan to wait for some events, run some code, then wait.

        This context manager will first create plans to wait for the
        provided event_names, yield, and then wait for all the scheduled
        events to complete.

        Note that this uses an eventlet.timeout.Timeout to bound the
        operation, so callers should be prepared to catch that
        failure and handle that situation appropriately.

        If the event is not received by the specified timeout deadline,
        eventlet.timeout.Timeout is raised.

        If the event is received but did not have a 'completed'
        status, a NovaException is raised.  If an error_callback is
        provided, instead of raising an exception as detailed above
        for the failure case, the callback will be called with the
        event_name and instance, and can return True to continue
        waiting for the rest of the events, False to stop processing,
        or raise an exception which will bubble up to the waiter.

        If the inner code wishes to abort waiting for one or more
        events because it knows some state to be finished or condition
        to be satisfied, it can use VirtAPI.exit_wait_early() with a
        list of event (name,tag) items to avoid waiting for those
        events upon context exit. Note that exit_wait_early() exits
        the context immediately and should be used to signal that all
        work has been completed and provide the unified list of events
        that need not be waited for. Waiting for the remaining events
        will begin immediately upon early exit as if the context was
        exited normally.

        :param instance: The instance for which an event is expected
        :param event_names: A list of event names. Each element is a
                            tuple of strings to indicate (name, tag),
                            where name is required, but tag may be None.
        :param deadline: Maximum number of seconds we should wait for all
                         of the specified events to arrive.
        :param error_callback: A function to be called if an event arrives

        """

        if error_callback is None:
            error_callback = self._default_error_callback
        events = {}
        for event_name in event_names:
            name, tag = event_name
            event_name = objects.InstanceExternalEvent.make_key(name, tag)
            try:
                event = (
                    self._compute.instance_events.prepare_for_instance_event(
                        instance, name, tag))
                events[event_name] = self._InstanceEvent(event_name, event)
            except exception.NovaException:
                error_callback(event_name, instance)
                # NOTE(danms): Don't wait for any of the events. They
                # should all be canceled and fired immediately below,
                # but don't stick around if not.
                deadline = 0
        try:
            yield
        except self._exit_early_exc as e:
            early_events = set([objects.InstanceExternalEvent.make_key(n, t)
                                for n, t in e.events])

            # If there are expected events that received early, mark them,
            # so they won't be waited for later
            for early_event_name in early_events:
                if early_event_name in events:
                    events[early_event_name].mark_as_received_early()

        sw = timeutils.StopWatch()
        sw.start()
        try:
            with eventlet.timeout.Timeout(deadline):
                self._wait_for_instance_events(
                    instance, events, error_callback)
        except eventlet.timeout.Timeout:
            LOG.warning(
                'Timeout waiting for %(events)s for instance with '
                'vm_state %(vm_state)s and task_state %(task_state)s. '
                'Event states are: %(event_states)s',
                {
                    'events': list(events.keys()),
                    'vm_state': instance.vm_state,
                    'task_state': instance.task_state,
                    'event_states':
                        ', '.join([str(event) for event in events.values()]),
                },
                instance=instance)

            raise

        LOG.debug('Instance event wait completed in %i seconds for %s',
                  sw.elapsed(),
                  ','.join(x[0] for x in event_names),
                  instance=instance)

    def update_compute_provider_status(self, context, rp_uuid, enabled):
        """Used to add/remove the COMPUTE_STATUS_DISABLED trait on the provider

        :param context: nova auth RequestContext
        :param rp_uuid: UUID of a compute node resource provider in Placement
        :param enabled: True if the node is enabled in which case the trait
            would be removed, False if the node is disabled in which case
            the trait would be added.
        :raises: ResourceProviderTraitRetrievalFailed
        :raises: ResourceProviderUpdateConflict
        :raises: ResourceProviderUpdateFailed
        :raises: TraitRetrievalFailed
        :raises: keystoneauth1.exceptions.ClientException
        """
        trait_name = os_traits.COMPUTE_STATUS_DISABLED
        # Get the current traits (and generation) for the provider.
        # TODO(mriedem): Leverage the ProviderTree cache in get_provider_traits
        trait_info = self.reportclient.get_provider_traits(context, rp_uuid)
        # If the host is enabled, remove the trait (if set), else add
        # the trait if it doesn't already exist.
        original_traits = trait_info.traits
        new_traits = None
        if enabled and trait_name in original_traits:
            new_traits = original_traits - {trait_name}
            LOG.debug('Removing trait %s from compute node resource '
                      'provider %s in placement.', trait_name, rp_uuid)
        elif not enabled and trait_name not in original_traits:
            new_traits = original_traits | {trait_name}
            LOG.debug('Adding trait %s to compute node resource '
                      'provider %s in placement.', trait_name, rp_uuid)

        if new_traits is not None:
            self.reportclient.set_traits_for_provider(
                context, rp_uuid, new_traits, generation=trait_info.generation)


class ComputeManager(manager.Manager):
    """Manages the running instances from creation to destruction."""

    target = messaging.Target(version='6.1')

    def __init__(self, compute_driver=None, *args, **kwargs):
        """Load configuration options and connect to the hypervisor."""
        # We want the ComputeManager, ResourceTracker and ComputeVirtAPI all
        # using the same instance of SchedulerReportClient which has the
        # ProviderTree cache for this compute service.
        # NOTE(danms): We do not use the global placement client
        # singleton here, because the above-mentioned stack of objects
        # maintain local state in the client. Thus, keeping our own
        # private object for that stack avoids any potential conflict
        # with other users in our process outside of the above.
        self.reportclient = report.SchedulerReportClient()
        self.virtapi = ComputeVirtAPI(self)
        self.network_api = neutron.API()
        self.volume_api = cinder.API()
        self.image_api = glance.API()
        self._last_bw_usage_poll = 0.0
        self.compute_api = compute.API()
        self.compute_rpcapi = compute_rpcapi.ComputeAPI()
        self.compute_task_api = conductor.ComputeTaskAPI()
        self.query_client = query.SchedulerQueryClient()
        self.instance_events = InstanceEvents()
        self._sync_power_pool = eventlet.GreenPool(
            size=CONF.sync_power_state_pool_size)
        self._syncs_in_progress = {}
        self.send_instance_updates = (
            CONF.filter_scheduler.track_instance_changes)
        if CONF.max_concurrent_builds != 0:
            self._build_semaphore = eventlet.semaphore.Semaphore(
                CONF.max_concurrent_builds)
        else:
            self._build_semaphore = compute_utils.UnlimitedSemaphore()
        if CONF.max_concurrent_snapshots > 0:
            self._snapshot_semaphore = eventlet.semaphore.Semaphore(
                CONF.max_concurrent_snapshots)
        else:
            self._snapshot_semaphore = compute_utils.UnlimitedSemaphore()
        if CONF.max_concurrent_live_migrations > 0:
            self._live_migration_executor = futurist.GreenThreadPoolExecutor(
                max_workers=CONF.max_concurrent_live_migrations)
        else:
            # CONF.max_concurrent_live_migrations is 0 (unlimited)
            self._live_migration_executor = futurist.GreenThreadPoolExecutor()
        # This is a dict, keyed by instance uuid, to a two-item tuple of
        # migration object and Future for the queued live migration.
        self._waiting_live_migrations = {}

        super(ComputeManager, self).__init__(service_name="compute",
                                             *args, **kwargs)

        # TODO(sbauza): Remove this call once we delete the V5Proxy class
        self.additional_endpoints.append(_ComputeV5Proxy(self))

        # NOTE(russellb) Load the driver last.  It may call back into the
        # compute manager via the virtapi, so we want it to be fully
        # initialized before that happens.
        self.driver = driver.load_compute_driver(self.virtapi, compute_driver)
        self.rt = resource_tracker.ResourceTracker(
            self.host, self.driver, reportclient=self.reportclient)

    def reset(self):
        LOG.info('Reloading compute RPC API')
        compute_rpcapi.reset_globals()
        self.compute_rpcapi = compute_rpcapi.ComputeAPI()
        self.reportclient.clear_provider_cache()

    def _update_resource_tracker(self, context, instance):
        """Let the resource tracker know that an instance has changed state."""

        if instance.host == self.host:
            self.rt.update_usage(context, instance, instance.node)

    def _instance_update(self, context, instance, **kwargs):
        """Update an instance in the database using kwargs as value."""

        for k, v in kwargs.items():
            setattr(instance, k, v)
        instance.save()
        self._update_resource_tracker(context, instance)

    def _nil_out_instance_obj_host_and_node(self, instance):
        # NOTE(jwcroppe): We don't do instance.save() here for performance
        # reasons; a call to this is expected to be immediately followed by
        # another call that does instance.save(), thus avoiding two writes
        # to the database layer.
        instance.host = None
        instance.node = None
        # ResourceTracker._set_instance_host_and_node also sets launched_on
        # to the same value as host and is really only ever used by legacy
        # nova-network code, but we should also null it out to avoid confusion
        # if there is an instance in the database with no host set but
        # launched_on is set. Note that we do not care about using launched_on
        # as some kind of debug helper if diagnosing a build failure, that is
        # what instance action events are for.
        instance.launched_on = None
        # If the instance is not on a host, it's not in an aggregate and
        # therefore is not in an availability zone.
        instance.availability_zone = None

    def _set_instance_obj_error_state(self, instance, clean_task_state=False):
        try:
            instance.vm_state = vm_states.ERROR
            if clean_task_state:
                instance.task_state = None
            instance.save()
        except exception.InstanceNotFound:
            LOG.debug('Instance has been destroyed from under us while '
                      'trying to set it to ERROR', instance=instance)

    def _get_instances_on_driver(self, context, filters=None):
        """Return a list of instance records for the instances found
        on the hypervisor which satisfy the specified filters. If filters=None
        return a list of instance records for all the instances found on the
        hypervisor.
        """
        if not filters:
            filters = {}
        try:
            driver_uuids = self.driver.list_instance_uuids()
            if len(driver_uuids) == 0:
                # Short circuit, don't waste a DB call
                return objects.InstanceList()
            filters['uuid'] = driver_uuids
            local_instances = objects.InstanceList.get_by_filters(
                context, filters, use_slave=True)
            return local_instances
        except NotImplementedError:
            pass

        # The driver doesn't support uuids listing, so we'll have
        # to brute force.
        driver_instances = self.driver.list_instances()
        # NOTE(mjozefcz): In this case we need to apply host filter.
        # Without this all instance data would be fetched from db.
        filters['host'] = self.host
        instances = objects.InstanceList.get_by_filters(context, filters,
                                                        use_slave=True)
        name_map = {instance.name: instance for instance in instances}
        local_instances = []
        for driver_instance in driver_instances:
            instance = name_map.get(driver_instance)
            if not instance:
                continue
            local_instances.append(instance)
        return local_instances

    def _destroy_evacuated_instances(self, context, node_cache):
        """Destroys evacuated instances.

        While nova-compute was down, the instances running on it could be
        evacuated to another host. This method looks for evacuation migration
        records where this is the source host and which were either started
        (accepted), in-progress (pre-migrating) or migrated (done). From those
        migration records, local instances reported by the hypervisor are
        compared to the instances for the migration records and those local
        guests are destroyed, along with instance allocation records in
        Placement for this node.
        Then allocations are removed from Placement for every instance that is
        evacuated from this host regardless if the instance is reported by the
        hypervisor or not.

        :param context: The request context
        :param node_cache: A dict of ComputeNode objects keyed by the UUID of
            the compute node
        :return: A dict keyed by instance uuid mapped to Migration objects
            for instances that were migrated away from this host
        """
        filters = {
            'source_compute': self.host,
            # NOTE(mriedem): Migration records that have been accepted are
            # included in case the source node comes back up while instances
            # are being evacuated to another host. We don't want the same
            # instance being reported from multiple hosts.
            # NOTE(lyarwood): pre-migrating is also included here as the
            # source compute can come back online shortly after the RT
            # claims on the destination that in-turn moves the migration to
            # pre-migrating. If the evacuate fails on the destination host,
            # the user can rebuild the instance (in ERROR state) on the source
            # host.
            'status': ['accepted', 'pre-migrating', 'done'],
            'migration_type': fields.MigrationType.EVACUATION,
        }
        with utils.temporary_mutation(context, read_deleted='yes'):
            evacuations = objects.MigrationList.get_by_filters(context,
                                                               filters)
        if not evacuations:
            return {}
        evacuations = {mig.instance_uuid: mig for mig in evacuations}

        # TODO(mriedem): We could optimize by pre-loading the joined fields
        # we know we'll use, like info_cache and flavor.
        local_instances = self._get_instances_on_driver(context)
        evacuated_local_instances = {inst.uuid: inst
                                     for inst in local_instances
                                     if inst.uuid in evacuations}

        for instance in evacuated_local_instances.values():
            LOG.info('Destroying instance as it has been evacuated from '
                     'this host but still exists in the hypervisor',
                     instance=instance)
            try:
                network_info = self.network_api.get_instance_nw_info(
                    context, instance)
                bdi = self._get_instance_block_device_info(context,
                                                           instance)
                destroy_disks = not (self._is_instance_storage_shared(
                    context, instance))
            except exception.InstanceNotFound:
                network_info = network_model.NetworkInfo()
                bdi = {}
                LOG.info('Instance has been marked deleted already, '
                         'removing it from the hypervisor.',
                         instance=instance)
                # always destroy disks if the instance was deleted
                destroy_disks = True
            self.driver.destroy(context, instance,
                                network_info,
                                bdi, destroy_disks)

        hostname_to_cn_uuid = {
            cn.hypervisor_hostname: cn.uuid
            for cn in node_cache.values()}

        for instance_uuid, migration in evacuations.items():
            try:
                if instance_uuid in evacuated_local_instances:
                    # Avoid the db call if we already have the instance loaded
                    # above
                    instance = evacuated_local_instances[instance_uuid]
                else:
                    instance = objects.Instance.get_by_uuid(
                        context, instance_uuid)
            except exception.InstanceNotFound:
                # The instance already deleted so we expect that every
                # allocation of that instance has already been cleaned up
                continue

            LOG.info('Cleaning up allocations of the instance as it has been '
                     'evacuated from this host',
                     instance=instance)
            if migration.source_node not in hostname_to_cn_uuid:
                LOG.error("Failed to clean allocation of evacuated "
                          "instance as the source node %s is not found",
                          migration.source_node, instance=instance)
                continue
            cn_uuid = hostname_to_cn_uuid[migration.source_node]

            # If the instance was deleted in the interim, assume its
            # allocations were properly cleaned up (either by its hosting
            # compute service or the API).
            if (not instance.deleted and
                    not self.reportclient.
                        remove_provider_tree_from_instance_allocation(
                            context, instance.uuid, cn_uuid)):
                LOG.error("Failed to clean allocation of evacuated instance "
                          "on the source node %s",
                          cn_uuid, instance=instance)

            migration.status = 'completed'
            migration.save()
        return evacuations

    def _is_instance_storage_shared(self, context, instance, host=None):
        shared_storage = True
        data = None
        try:
            data = self.driver.check_instance_shared_storage_local(context,
                                                       instance)
            if data:
                shared_storage = (self.compute_rpcapi.
                                  check_instance_shared_storage(context,
                                  data, instance=instance, host=host))
        except NotImplementedError:
            LOG.debug('Hypervisor driver does not support '
                      'instance shared storage check, '
                      'assuming it\'s not on shared storage',
                      instance=instance)
            shared_storage = False
        except Exception:
            LOG.exception('Failed to check if instance shared',
                          instance=instance)
        finally:
            if data:
                self.driver.check_instance_shared_storage_cleanup(context,
                                                                  data)
        return shared_storage

    def _complete_partial_deletion(self, context, instance):
        """Complete deletion for instances in DELETED status but not marked as
        deleted in the DB
        """
        instance.destroy()
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)
        self._complete_deletion(context,
                                instance)
        self._notify_about_instance_usage(context, instance, "delete.end")
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.DELETE,
                phase=fields.NotificationPhase.END, bdms=bdms)

    def _complete_deletion(self, context, instance):
        self._update_resource_tracker(context, instance)

        # If we're configured to do deferred deletes, don't force deletion of
        # allocations if there's a conflict.
        force = False if CONF.reclaim_instance_interval > 0 else True

        self.reportclient.delete_allocation_for_instance(context,
                                                         instance.uuid,
                                                         force=force)

        self._clean_instance_console_tokens(context, instance)
        self._delete_scheduler_instance_info(context, instance.uuid)

    def _validate_pinning_configuration(self, instances):
        if not self.driver.capabilities.get('supports_pcpus', False):
            return

        for instance in instances:
            # ignore deleted instances
            if instance.deleted:
                continue

            # if this is an unpinned instance and the host only has
            # 'cpu_dedicated_set' configured, we need to tell the operator to
            # correct their configuration
            if not instance.numa_topology or (
                instance.numa_topology.cpu_policy in (
                    None, fields.CPUAllocationPolicy.SHARED
                )
            ):
                # we don't need to check 'vcpu_pin_set' since it can't coexist
                # alongside 'cpu_dedicated_set'
                if (CONF.compute.cpu_dedicated_set and
                        not CONF.compute.cpu_shared_set):
                    msg = _("This host has unpinned instances but has no CPUs "
                            "set aside for this purpose; configure '[compute] "
                            "cpu_shared_set' instead of, or in addition to, "
                            "'[compute] cpu_dedicated_set'")
                    raise exception.InvalidConfiguration(msg)

                continue

            # ditto for pinned instances if only 'cpu_shared_set' is configured
            if (CONF.compute.cpu_shared_set and
                    not CONF.compute.cpu_dedicated_set and
                    not CONF.vcpu_pin_set):
                msg = _("This host has pinned instances but has no CPUs "
                        "set aside for this purpose; configure '[compute] "
                        "cpu_dedicated_set' instead of, or in addition to, "
                        "'[compute] cpu_shared_set'.")
                raise exception.InvalidConfiguration(msg)

            # if this is a mixed instance with both pinned and unpinned CPUs,
            # the host must have both 'cpu_dedicated_set' and 'cpu_shared_set'
            # configured. check if 'cpu_shared_set' is set.
            if (instance.numa_topology.cpu_policy ==
                    fields.CPUAllocationPolicy.MIXED and
                    not CONF.compute.cpu_shared_set):
                msg = _("This host has mixed instance requesting both pinned "
                        "and unpinned CPUs but hasn't set aside unpinned CPUs "
                        "for this purpose; Configure "
                        "'[compute] cpu_shared_set'.")
                raise exception.InvalidConfiguration(msg)

            # for mixed instance check if 'cpu_dedicated_set' is set.
            if (instance.numa_topology.cpu_policy ==
                    fields.CPUAllocationPolicy.MIXED and
                    not CONF.compute.cpu_dedicated_set):
                msg = _("This host has mixed instance requesting both pinned "
                        "and unpinned CPUs but hasn't set aside pinned CPUs "
                        "for this purpose; Configure "
                        "'[compute] cpu_dedicated_set'")
                raise exception.InvalidConfiguration(msg)

            # also check to make sure the operator hasn't accidentally
            # dropped some cores that instances are currently using
            available_dedicated_cpus = (hardware.get_vcpu_pin_set() or
                                        hardware.get_cpu_dedicated_set())
            pinned_cpus = instance.numa_topology.cpu_pinning
            if available_dedicated_cpus and (
                    pinned_cpus - available_dedicated_cpus):
                # we can't raise an exception because of bug #1289064,
                # which meant we didn't recalculate CPU pinning information
                # when we live migrated a pinned instance
                LOG.warning(
                    "Instance is pinned to host CPUs %(cpus)s "
                    "but one or more of these CPUs are not included in "
                    "either '[compute] cpu_dedicated_set' or "
                    "'vcpu_pin_set'; you should update these "
                    "configuration options to include the missing CPUs "
                    "or rebuild or cold migrate this instance.",
                    {'cpus': list(pinned_cpus)},
                    instance=instance)

    def _validate_vtpm_configuration(self, instances):
        if self.driver.capabilities.get('supports_vtpm', False):
            return

        for instance in instances:
            if instance.deleted:
                continue

            # NOTE(stephenfin): We don't have an attribute on the instance to
            # check for this, so we need to inspect the flavor/image metadata
            if hardware.get_vtpm_constraint(
                instance.flavor, instance.image_meta,
            ):
                msg = _(
                    'This host has instances with the vTPM feature enabled, '
                    'but the host is not correctly configured; enable '
                    'vTPM support.'
                )
                raise exception.InvalidConfiguration(msg)

    def _reset_live_migration(self, context, instance):
        migration = None
        try:
            migration = objects.Migration.get_by_instance_and_status(
                                      context, instance.uuid, 'running')
            if migration:
                self.live_migration_abort(context, instance, migration.id)
        except Exception:
            LOG.exception('Failed to abort live-migration',
                          instance=instance)
        finally:
            if migration:
                self._set_migration_status(migration, 'error')
            LOG.info('Instance found in migrating state during '
                     'startup. Resetting task_state',
                     instance=instance)
            instance.task_state = None
            instance.save(expected_task_state=[task_states.MIGRATING])

    def _init_instance(self, context, instance):
        """Initialize this instance during service init."""

        # NOTE(danms): If the instance appears to not be owned by this
        # host, it may have been evacuated away, but skipped by the
        # evacuation cleanup code due to configuration. Thus, if that
        # is a possibility, don't touch the instance in any way, but
        # log the concern. This will help avoid potential issues on
        # startup due to misconfiguration.
        if instance.host != self.host:
            LOG.warning('Instance %(uuid)s appears to not be owned '
                        'by this host, but by %(host)s. Startup '
                        'processing is being skipped.',
                        {'uuid': instance.uuid,
                         'host': instance.host})
            return

        # Instances that are shut down, or in an error state can not be
        # initialized and are not attempted to be recovered. The exception
        # to this are instances that are in RESIZE_MIGRATING or DELETING,
        # which are dealt with further down.
        if (instance.vm_state == vm_states.SOFT_DELETED or
            (instance.vm_state == vm_states.ERROR and
            instance.task_state not in
            (task_states.RESIZE_MIGRATING, task_states.DELETING))):
            LOG.debug("Instance is in %s state.",
                      instance.vm_state, instance=instance)
            return

        if instance.vm_state == vm_states.DELETED:
            try:
                self._complete_partial_deletion(context, instance)
            except Exception:
                # we don't want that an exception blocks the init_host
                LOG.exception('Failed to complete a deletion',
                              instance=instance)
            return

        if (instance.vm_state == vm_states.BUILDING or
            instance.task_state in [task_states.SCHEDULING,
                                    task_states.BLOCK_DEVICE_MAPPING,
                                    task_states.NETWORKING,
                                    task_states.SPAWNING]):
            # NOTE(dave-mcnally) compute stopped before instance was fully
            # spawned so set to ERROR state. This is safe to do as the state
            # may be set by the api but the host is not so if we get here the
            # instance has already been scheduled to this particular host.
            LOG.debug("Instance failed to spawn correctly, "
                      "setting to ERROR state", instance=instance)
            self._set_instance_obj_error_state(instance, clean_task_state=True)
            return

        if (instance.vm_state in [vm_states.ACTIVE, vm_states.STOPPED] and
            instance.task_state in [task_states.REBUILDING,
                                    task_states.REBUILD_BLOCK_DEVICE_MAPPING,
                                    task_states.REBUILD_SPAWNING]):
            # NOTE(jichenjc) compute stopped before instance was fully
            # spawned so set to ERROR state. This is consistent to BUILD
            LOG.debug("Instance failed to rebuild correctly, "
                      "setting to ERROR state", instance=instance)
            self._set_instance_obj_error_state(instance, clean_task_state=True)
            return

        if (instance.vm_state != vm_states.ERROR and
            instance.task_state in [task_states.IMAGE_SNAPSHOT_PENDING,
                                    task_states.IMAGE_PENDING_UPLOAD,
                                    task_states.IMAGE_UPLOADING,
                                    task_states.IMAGE_SNAPSHOT]):
            LOG.debug("Instance in transitional state %s at start-up "
                      "clearing task state",
                      instance.task_state, instance=instance)
            instance.task_state = None
            instance.save()

        if (instance.vm_state != vm_states.ERROR and
            instance.task_state in [task_states.RESIZE_PREP]):
            LOG.debug("Instance in transitional state %s at start-up "
                      "clearing task state",
                      instance['task_state'], instance=instance)
            instance.task_state = None
            instance.save()

        if instance.task_state == task_states.DELETING:
            try:
                LOG.info('Service started deleting the instance during '
                         'the previous run, but did not finish. Restarting'
                         ' the deletion now.', instance=instance)
                instance.obj_load_attr('metadata')
                instance.obj_load_attr('system_metadata')
                bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                        context, instance.uuid)
                self._delete_instance(context, instance, bdms)
            except Exception:
                # we don't want that an exception blocks the init_host
                LOG.exception('Failed to complete a deletion',
                              instance=instance)
                self._set_instance_obj_error_state(instance)
            return

        current_power_state = self._get_power_state(instance)
        try_reboot, reboot_type = self._retry_reboot(
            instance, current_power_state)

        if try_reboot:
            LOG.debug("Instance in transitional state (%(task_state)s) at "
                      "start-up and power state is (%(power_state)s), "
                      "triggering reboot",
                      {'task_state': instance.task_state,
                       'power_state': current_power_state},
                      instance=instance)

            # NOTE(mikal): if the instance was doing a soft reboot that got as
            # far as shutting down the instance but not as far as starting it
            # again, then we've just become a hard reboot. That means the
            # task state for the instance needs to change so that we're in one
            # of the expected task states for a hard reboot.
            if (instance.task_state in task_states.soft_reboot_states and
                reboot_type == 'HARD'):
                instance.task_state = task_states.REBOOT_PENDING_HARD
                instance.save()

            self.reboot_instance(context, instance, block_device_info=None,
                                 reboot_type=reboot_type)
            return

        elif (current_power_state == power_state.RUNNING and
              instance.task_state in [task_states.REBOOT_STARTED,
                                      task_states.REBOOT_STARTED_HARD,
                                      task_states.PAUSING,
                                      task_states.UNPAUSING]):
            LOG.warning("Instance in transitional state "
                        "(%(task_state)s) at start-up and power state "
                        "is (%(power_state)s), clearing task state",
                        {'task_state': instance.task_state,
                         'power_state': current_power_state},
                        instance=instance)
            instance.task_state = None
            instance.vm_state = vm_states.ACTIVE
            instance.save()
        elif (current_power_state == power_state.PAUSED and
              instance.task_state == task_states.UNPAUSING):
            LOG.warning("Instance in transitional state "
                        "(%(task_state)s) at start-up and power state "
                        "is (%(power_state)s), clearing task state "
                        "and unpausing the instance",
                        {'task_state': instance.task_state,
                         'power_state': current_power_state},
                        instance=instance)
            try:
                self.unpause_instance(context, instance)
            except NotImplementedError:
                # Some virt driver didn't support pause and unpause
                pass
            except Exception:
                LOG.exception('Failed to unpause instance', instance=instance)
            return

        if instance.task_state == task_states.POWERING_OFF:
            try:
                LOG.debug("Instance in transitional state %s at start-up "
                          "retrying stop request",
                          instance.task_state, instance=instance)
                self.stop_instance(context, instance, True)
            except Exception:
                # we don't want that an exception blocks the init_host
                LOG.exception('Failed to stop instance', instance=instance)
            return

        if instance.task_state == task_states.POWERING_ON:
            try:
                LOG.debug("Instance in transitional state %s at start-up "
                          "retrying start request",
                          instance.task_state, instance=instance)
                self.start_instance(context, instance)
            except Exception:
                # we don't want that an exception blocks the init_host
                LOG.exception('Failed to start instance', instance=instance)
            return

        net_info = instance.get_network_info()
        try:
            self.driver.plug_vifs(instance, net_info)
        except NotImplementedError as e:
            LOG.debug(e, instance=instance)
        except exception.VirtualInterfacePlugException:
            # NOTE(mriedem): If we get here, it could be because the vif_type
            # in the cache is "binding_failed" or "unbound".
            # The periodic task _heal_instance_info_cache checks for this
            # condition. It should fix this by binding the ports again when
            # it gets to this instance.
            LOG.exception('Virtual interface plugging failed for instance. '
                          'The port binding:host_id may need to be manually '
                          'updated.', instance=instance)
            self._set_instance_obj_error_state(instance)
            return
        except exception.PciDeviceNotFoundById:
            # This is bug 1981813 where the bound port vnic_type has changed
            # from direct to macvtap. Nova does not support that and it
            # already printed an ERROR when the change is detected during
            # _heal_instance_info_cache. Now we print an ERROR again and skip
            # plugging the vifs but let the service startup continue to init
            # the other instances
            LOG.exception(
                'Virtual interface plugging failed for instance. Probably the '
                'vnic_type of the bound port has been changed. Nova does not '
                'support such change.',
                instance=instance
            )
            return

        if instance.task_state == task_states.RESIZE_MIGRATING:
            # We crashed during resize/migration, so roll back for safety
            try:
                # NOTE(mriedem): check old_vm_state for STOPPED here, if it's
                # not in system_metadata we default to True for backwards
                # compatibility
                power_on = (instance.system_metadata.get('old_vm_state') !=
                            vm_states.STOPPED)

                block_dev_info = self._get_instance_block_device_info(context,
                                                                      instance)

                migration = objects.Migration.get_by_id_and_instance(
                    context, instance.migration_context.migration_id,
                    instance.uuid)
                self.driver.finish_revert_migration(context, instance,
                    net_info, migration, block_dev_info, power_on)

            except Exception:
                LOG.exception('Failed to revert crashed migration',
                              instance=instance)
            finally:
                LOG.info('Instance found in migrating state during '
                         'startup. Resetting task_state',
                         instance=instance)
                instance.task_state = None
                instance.save()
        if instance.task_state == task_states.MIGRATING:
            # Live migration did not complete, but instance is on this
            # host. Abort ongoing migration if still running and reset state.
            self._reset_live_migration(context, instance)

        db_state = instance.power_state
        drv_state = self._get_power_state(instance)
        expect_running = (db_state == power_state.RUNNING and
                          drv_state != db_state)

        LOG.debug('Current state is %(drv_state)s, state in DB is '
                  '%(db_state)s.',
                  {'drv_state': drv_state, 'db_state': db_state},
                  instance=instance)

        if expect_running and CONF.resume_guests_state_on_host_boot:
            self._resume_guests_state(context, instance, net_info)

    def _resume_guests_state(self, context, instance, net_info):
        LOG.info('Rebooting instance after nova-compute restart.',
                 instance=instance)
        block_device_info = \
            self._get_instance_block_device_info(context, instance)

        try:
            self.driver.resume_state_on_host_boot(
                context, instance, net_info, block_device_info)
        except NotImplementedError:
            LOG.warning('Hypervisor driver does not support '
                        'resume guests', instance=instance)
        except Exception:
            # NOTE(vish): The instance failed to resume, so we set the
            #             instance to error and attempt to continue.
            LOG.warning('Failed to resume instance',
                        instance=instance)
            self._set_instance_obj_error_state(instance)

    def _retry_reboot(self, instance, current_power_state):
        current_task_state = instance.task_state
        retry_reboot = False
        reboot_type = compute_utils.get_reboot_type(current_task_state,
                                                    current_power_state)

        pending_soft = (
            current_task_state == task_states.REBOOT_PENDING and
            instance.vm_state in vm_states.ALLOW_SOFT_REBOOT)
        pending_hard = (
            current_task_state == task_states.REBOOT_PENDING_HARD and
            instance.vm_state in vm_states.ALLOW_HARD_REBOOT)
        started_not_running = (current_task_state in
                               [task_states.REBOOT_STARTED,
                                task_states.REBOOT_STARTED_HARD] and
                               current_power_state != power_state.RUNNING)

        if pending_soft or pending_hard or started_not_running:
            retry_reboot = True

        return retry_reboot, reboot_type

    def handle_lifecycle_event(self, event):
        LOG.info("VM %(state)s (Lifecycle Event)",
                 {'state': event.get_name()},
                 instance_uuid=event.get_instance_uuid())
        context = nova.context.get_admin_context(read_deleted='yes')
        vm_power_state = None
        event_transition = event.get_transition()
        if event_transition == virtevent.EVENT_LIFECYCLE_STOPPED:
            vm_power_state = power_state.SHUTDOWN
        elif event_transition == virtevent.EVENT_LIFECYCLE_STARTED:
            vm_power_state = power_state.RUNNING
        elif event_transition in (
                virtevent.EVENT_LIFECYCLE_PAUSED,
                virtevent.EVENT_LIFECYCLE_POSTCOPY_STARTED,
                virtevent.EVENT_LIFECYCLE_MIGRATION_COMPLETED):
            vm_power_state = power_state.PAUSED
        elif event_transition == virtevent.EVENT_LIFECYCLE_RESUMED:
            vm_power_state = power_state.RUNNING
        elif event_transition == virtevent.EVENT_LIFECYCLE_SUSPENDED:
            vm_power_state = power_state.SUSPENDED
        else:
            LOG.warning("Unexpected lifecycle event: %d", event_transition)

        migrate_finish_statuses = {
            # This happens on the source node and indicates live migration
            # entered post-copy mode.
            virtevent.EVENT_LIFECYCLE_POSTCOPY_STARTED: 'running (post-copy)',
            # Suspended for offline migration.
            virtevent.EVENT_LIFECYCLE_MIGRATION_COMPLETED: 'running'
        }

        expected_attrs = []
        if event_transition in migrate_finish_statuses:
            # Join on info_cache since that's needed in migrate_instance_start.
            expected_attrs.append('info_cache')
        instance = objects.Instance.get_by_uuid(context,
                                                event.get_instance_uuid(),
                                                expected_attrs=expected_attrs)

        # Note(lpetrut): The event may be delayed, thus not reflecting
        # the current instance power state. In that case, ignore the event.
        current_power_state = self._get_power_state(instance)
        if current_power_state == vm_power_state:
            LOG.debug('Synchronizing instance power state after lifecycle '
                      'event "%(event)s"; current vm_state: %(vm_state)s, '
                      'current task_state: %(task_state)s, current DB '
                      'power_state: %(db_power_state)s, VM power_state: '
                      '%(vm_power_state)s',
                      {'event': event.get_name(),
                       'vm_state': instance.vm_state,
                       'task_state': instance.task_state,
                       'db_power_state': instance.power_state,
                       'vm_power_state': vm_power_state},
                      instance_uuid=instance.uuid)
            self._sync_instance_power_state(context,
                                            instance,
                                            vm_power_state)

        # The following checks are for live migration. We want to activate
        # the port binding for the destination host before the live migration
        # is resumed on the destination host in order to reduce network
        # downtime. Otherwise the ports are bound to the destination host
        # in post_live_migration_at_destination.
        # TODO(danms): Explore options for using a different live migration
        # specific callback for this instead of piggy-backing on the
        # handle_lifecycle_event callback.
        if (instance.task_state == task_states.MIGRATING and
                event_transition in migrate_finish_statuses):
            status = migrate_finish_statuses[event_transition]
            try:
                migration = objects.Migration.get_by_instance_and_status(
                            context, instance.uuid, status)
                LOG.debug('Binding ports to destination host: %s',
                          migration.dest_compute, instance=instance)
                # For neutron, migrate_instance_start will activate the
                # destination host port bindings, if there are any created by
                # conductor before live migration started.
                self.network_api.migrate_instance_start(
                    context, instance, migration)
            except exception.MigrationNotFoundByStatus:
                LOG.warning("Unable to find migration record with status "
                            "'%s' for instance. Port binding will happen in "
                            "post live migration.", status, instance=instance)

    def handle_events(self, event):
        if isinstance(event, virtevent.LifecycleEvent):
            try:
                self.handle_lifecycle_event(event)
            except exception.InstanceNotFound:
                LOG.debug("Event %s arrived for non-existent instance. The "
                          "instance was probably deleted.", event)
        else:
            LOG.debug("Ignoring event %s", event)

    def init_virt_events(self):
        if CONF.workarounds.handle_virt_lifecycle_events:
            self.driver.register_event_listener(self.handle_events)
        else:
            # NOTE(mriedem): If the _sync_power_states periodic task is
            # disabled we should emit a warning in the logs.
            if CONF.sync_power_state_interval < 0:
                LOG.warning('Instance lifecycle events from the compute '
                            'driver have been disabled. Note that lifecycle '
                            'changes to an instance outside of the compute '
                            'service will not be synchronized '
                            'automatically since the _sync_power_states '
                            'periodic task is also disabled.')
            else:
                LOG.info('Instance lifecycle events from the compute '
                         'driver have been disabled. Note that lifecycle '
                         'changes to an instance outside of the compute '
                         'service will only be synchronized by the '
                         '_sync_power_states periodic task.')

    def _get_nodes(self, context):
        """Queried the ComputeNode objects from the DB that are reported by the
        hypervisor.

        :param context: the request context
        :return: a dict of ComputeNode objects keyed by the UUID of the given
            node.
        """
        nodes_by_uuid = {}
        try:
            node_names = self.driver.get_available_nodes()
        except exception.VirtDriverNotReady:
            LOG.warning(
                "Virt driver is not ready. If this is the first time this "
                "service is starting on this host, then you can ignore this "
                "warning.")
            return {}

        for node_name in node_names:
            try:
                node = objects.ComputeNode.get_by_host_and_nodename(
                    context, self.host, node_name)
                nodes_by_uuid[node.uuid] = node
            except exception.ComputeHostNotFound:
                LOG.warning(
                    "Compute node %s not found in the database. If this is "
                    "the first time this service is starting on this host, "
                    "then you can ignore this warning.", node_name)
        return nodes_by_uuid

    def init_host(self):
        """Initialization for a standalone compute service."""

        if CONF.pci.device_spec:
            # Simply loading the PCI passthrough spec will do a bunch of
            # validation that would otherwise wait until the PciDevTracker is
            # constructed when updating available resources for the compute
            # node(s) in the resource tracker, effectively killing that task.
            # So load up the spec when starting the compute service to
            # flush any invalid configuration early, so we can kill the service
            # if the configuration is wrong.
            whitelist.Whitelist(CONF.pci.device_spec)

        nova.conf.neutron.register_dynamic_opts(CONF)
        # Even if only libvirt uses them, make it available for all drivers
        nova.conf.devices.register_dynamic_opts(CONF)

        # Override the number of concurrent disk operations allowed if the
        # user has specified a limit.
        if CONF.compute.max_concurrent_disk_ops != 0:
            compute_utils.disk_ops_semaphore = \
                eventlet.semaphore.BoundedSemaphore(
                    CONF.compute.max_concurrent_disk_ops)

        if CONF.compute.max_disk_devices_to_attach == 0:
            msg = _('[compute]max_disk_devices_to_attach has been set to 0, '
                    'which will prevent instances from being able to boot. '
                    'Set -1 for unlimited or set >= 1 to limit the maximum '
                    'number of disk devices.')
            raise exception.InvalidConfiguration(msg)

        self.driver.init_host(host=self.host)
        context = nova.context.get_admin_context()
        instances = objects.InstanceList.get_by_host(
            context, self.host,
            expected_attrs=['info_cache', 'metadata', 'numa_topology'])

        self.init_virt_events()

        self._validate_pinning_configuration(instances)
        self._validate_vtpm_configuration(instances)

        # NOTE(gibi): At this point the compute_nodes of the resource tracker
        # has not been populated yet so we cannot rely on the resource tracker
        # here.
        # NOTE(gibi): If ironic and vcenter virt driver slow start time
        # becomes problematic here then we should consider adding a config
        # option or a driver flag to tell us if we should thread
        # _destroy_evacuated_instances and
        # _error_out_instances_whose_build_was_interrupted out in the
        # background on startup
        nodes_by_uuid = self._get_nodes(context)

        try:
            # checking that instance was not already evacuated to other host
            evacuated_instances = self._destroy_evacuated_instances(
                context, nodes_by_uuid)

            # Initialise instances on the host that are not evacuating
            for instance in instances:
                if instance.uuid not in evacuated_instances:
                    self._init_instance(context, instance)

            # NOTE(gibi): collect all the instance uuids that is in some way
            # was already handled above. Either by init_instance or by
            # _destroy_evacuated_instances. This way we can limit the scope of
            # the _error_out_instances_whose_build_was_interrupted call to look
            # only for instances that have allocations on this node and not
            # handled by the above calls.
            already_handled = {instance.uuid for instance in instances}.union(
                evacuated_instances)
            self._error_out_instances_whose_build_was_interrupted(
                context, already_handled, nodes_by_uuid.keys())

        finally:
            if instances:
                # We only send the instance info to the scheduler on startup
                # if there is anything to send, otherwise this host might
                # not be mapped yet in a cell and the scheduler may have
                # issues dealing with the information. Later changes to
                # instances on this host will update the scheduler, or the
                # _sync_scheduler_instance_info periodic task will.
                self._update_scheduler_instance_info(context, instances)

    def _error_out_instances_whose_build_was_interrupted(
            self, context, already_handled_instances, node_uuids):
        """If there are instances in BUILDING state that are not
        assigned to this host but have allocations in placement towards
        this compute that means the nova-compute service was
        restarted while those instances waited for the resource claim
        to finish and the _set_instance_host_and_node() to update the
        instance.host field. We need to push them to ERROR state here to
        prevent keeping them in BUILDING state forever.

        :param context: The request context
        :param already_handled_instances: The set of instance UUIDs that the
            host initialization process already handled in some way.
        :param node_uuids: The list of compute node uuids handled by this
            service
        """

        # Strategy:
        # 1) Get the allocations from placement for our compute node(s)
        # 2) Remove the already handled instances from the consumer list;
        #    they are either already initialized or need to be skipped.
        # 3) Check which remaining consumer is an instance in BUILDING state
        #    and push it to ERROR state.

        LOG.info(
            "Looking for unclaimed instances stuck in BUILDING status for "
            "nodes managed by this host")
        for cn_uuid in node_uuids:
            try:
                f = self.reportclient.get_allocations_for_resource_provider
                allocations = f(context, cn_uuid).allocations
            except (exception.ResourceProviderAllocationRetrievalFailed,
                    keystone_exception.ClientException) as e:
                LOG.error(
                    "Could not retrieve compute node resource provider %s and "
                    "therefore unable to error out any instances stuck in "
                    "BUILDING state. Error: %s", cn_uuid, str(e))
                continue

            not_handled_consumers = (set(allocations) -
                                     already_handled_instances)

            if not not_handled_consumers:
                continue

            filters = {
                'vm_state': vm_states.BUILDING,
                'uuid': not_handled_consumers
            }

            instances = objects.InstanceList.get_by_filters(
                context, filters, expected_attrs=[])

            for instance in instances:
                LOG.debug(
                    "Instance spawn was interrupted before instance_claim, "
                    "setting instance to ERROR state", instance=instance)
                self._set_instance_obj_error_state(
                    instance, clean_task_state=True)

    def cleanup_host(self):
        self.driver.register_event_listener(None)
        self.instance_events.cancel_all_events()
        self.driver.cleanup_host(host=self.host)
        self._cleanup_live_migrations_in_pool()

    def _cleanup_live_migrations_in_pool(self):
        # Shutdown the pool so we don't get new requests.
        self._live_migration_executor.shutdown(wait=False)
        # For any queued migrations, cancel the migration and update
        # its status.
        for migration, future in self._waiting_live_migrations.values():
            # If we got here before the Future was submitted then we need
            # to move on since there isn't anything we can do.
            if future is None:
                continue
            if future.cancel():
                self._set_migration_status(migration, 'cancelled')
                LOG.info('Successfully cancelled queued live migration.',
                         instance_uuid=migration.instance_uuid)
            else:
                LOG.warning('Unable to cancel live migration.',
                            instance_uuid=migration.instance_uuid)
        self._waiting_live_migrations.clear()

    def pre_start_hook(self):
        """After the service is initialized, but before we fully bring
        the service up by listening on RPC queues, make sure to update
        our available resources (and indirectly our available nodes).
        """
        self.update_available_resource(nova.context.get_admin_context(),
                                       startup=True)

    def _get_power_state(self, instance):
        """Retrieve the power state for the given instance."""
        LOG.debug('Checking state', instance=instance)
        try:
            return self.driver.get_info(instance, use_cache=False).state
        except exception.InstanceNotFound:
            return power_state.NOSTATE

    def _await_block_device_map_created(self, context, vol_id):
        # TODO(yamahata): creating volume simultaneously
        #                 reduces creation time?
        # TODO(yamahata): eliminate dumb polling
        start = time.time()
        retries = CONF.block_device_allocate_retries
        # (1) if the configured value is 0, one attempt should be made
        # (2) if the configured value is > 0, then the total number attempts
        #      is (retries + 1)
        attempts = 1
        if retries >= 1:
            attempts = retries + 1
        for attempt in range(1, attempts + 1):
            volume = self.volume_api.get(context, vol_id)
            volume_status = volume['status']
            if volume_status not in ['creating', 'downloading']:
                if volume_status == 'available':
                    return attempt
                LOG.warning("Volume id: %(vol_id)s finished being "
                            "created but its status is %(vol_status)s.",
                            {'vol_id': vol_id,
                             'vol_status': volume_status})
                break
            greenthread.sleep(CONF.block_device_allocate_retries_interval)
        raise exception.VolumeNotCreated(volume_id=vol_id,
                                         seconds=int(time.time() - start),
                                         attempts=attempt,
                                         volume_status=volume_status)

    def _decode_files(self, injected_files):
        """Base64 decode the list of files to inject."""
        if not injected_files:
            return []

        def _decode(f):
            path, contents = f
            # Py3 raises binascii.Error instead of TypeError as in Py27
            try:
                decoded = base64.b64decode(contents)
                return path, decoded
            except (TypeError, binascii.Error):
                raise exception.Base64Exception(path=path)

        return [_decode(f) for f in injected_files]

    def _validate_instance_group_policy(self, context, instance,
                                        scheduler_hints=None):

        if CONF.workarounds.disable_group_policy_check_upcall:
            return

        # NOTE(russellb) Instance group policy is enforced by the scheduler.
        # However, there is a race condition with the enforcement of
        # the policy.  Since more than one instance may be scheduled at the
        # same time, it's possible that more than one instance with an
        # anti-affinity policy may end up here.  It's also possible that
        # multiple instances with an affinity policy could end up on different
        # hosts.  This is a validation step to make sure that starting the
        # instance here doesn't violate the policy.
        if scheduler_hints is not None:
            # only go through here if scheduler_hints is provided,
            # even if it is empty.
            group_hint = scheduler_hints.get('group')
            if not group_hint:
                return
            else:
                # The RequestSpec stores scheduler_hints as key=list pairs
                # so we need to check the type on the value and pull the
                # single entry out. The API request schema validates that
                # the 'group' hint is a single value.
                if isinstance(group_hint, list):
                    group_hint = group_hint[0]
                try:
                    group = objects.InstanceGroup.get_by_hint(
                        context, group_hint
                    )
                except exception.InstanceGroupNotFound:
                    return
        else:
            # TODO(ganso): a call to DB can be saved by adding request_spec
            # to rpcapi payload of live_migration, pre_live_migration and
            # check_can_live_migrate_destination
            try:
                group = objects.InstanceGroup.get_by_instance_uuid(
                    context, instance.uuid
                )
            except exception.InstanceGroupNotFound:
                return

        @utils.synchronized(group['uuid'])
        def _do_validation(context, instance, group):
            if group.policy and 'anti-affinity' == group.policy:

                # instances on host
                instances_uuids = objects.InstanceList.get_uuids_by_host(
                    context, self.host)
                ins_on_host = set(instances_uuids)

                # instance param is just for logging, the nodename obtained is
                # not actually related to the instance at all
                nodename = self._get_nodename(instance)

                # instances being migrated to host
                migrations = (
                    objects.MigrationList.get_in_progress_by_host_and_node(
                        context, self.host, nodename))
                migration_vm_uuids = {mig.instance_uuid for mig in migrations}

                total_instances = migration_vm_uuids | ins_on_host

                # refresh group to get updated members within locked block
                group = objects.InstanceGroup.get_by_uuid(context,
                                                          group['uuid'])
                members = set(group.members)
                # Determine the set of instance group members on this host
                # which are not the instance in question. This is used to
                # determine how many other members from the same anti-affinity
                # group can be on this host.
                members_on_host = (total_instances & members -
                                   set([instance.uuid]))
                rules = group.rules
                if rules and 'max_server_per_host' in rules:
                    max_server = rules['max_server_per_host']
                else:
                    max_server = 1
                if len(members_on_host) >= max_server:
                    msg = _("Anti-affinity instance group policy "
                            "was violated.")
                    raise exception.RescheduledException(
                            instance_uuid=instance.uuid,
                            reason=msg)

            # NOTE(ganso): The check for affinity below does not work and it
            # can easily be violated because the lock happens in different
            # compute hosts.
            # The only fix seems to be a DB lock to perform the check whenever
            # setting the host field to an instance.
            elif group.policy and 'affinity' == group.policy:
                group_hosts = group.get_hosts(exclude=[instance.uuid])
                if group_hosts and self.host not in group_hosts:
                    msg = _("Affinity instance group policy was violated.")
                    raise exception.RescheduledException(
                            instance_uuid=instance.uuid,
                            reason=msg)

        _do_validation(context, instance, group)

    def _log_original_error(self, exc_info, instance_uuid):
        LOG.error('Error: %s', exc_info[1], instance_uuid=instance_uuid,
                  exc_info=exc_info)

    @periodic_task.periodic_task
    def _check_instance_build_time(self, context):
        """Ensure that instances are not stuck in build."""
        timeout = CONF.instance_build_timeout
        if timeout == 0:
            return

        filters = {'vm_state': vm_states.BUILDING,
                   'host': self.host}

        building_insts = objects.InstanceList.get_by_filters(context,
                           filters, expected_attrs=[], use_slave=True)

        for instance in building_insts:
            if timeutils.is_older_than(instance.created_at, timeout):
                self._set_instance_obj_error_state(instance)
                LOG.warning("Instance build timed out. Set to error "
                            "state.", instance=instance)

    def _check_instance_exists(self, instance):
        """Ensure an instance with the same name is not already present."""
        if self.driver.instance_exists(instance):
            raise exception.InstanceExists(name=instance.name)

    def _allocate_network_async(self, context, instance, requested_networks,
                                security_groups, resource_provider_mapping,
                                network_arqs):
        """Method used to allocate networks in the background.

        Broken out for testing.
        """
        # First check to see if we're specifically not supposed to allocate
        # networks because if so, we can exit early.
        if requested_networks and requested_networks.no_allocate:
            LOG.debug("Not allocating networking since 'none' was specified.",
                      instance=instance)
            return network_model.NetworkInfo([])

        LOG.debug("Allocating IP information in the background.",
                  instance=instance)
        retries = CONF.network_allocate_retries
        attempts = retries + 1
        retry_time = 1
        bind_host_id = self.driver.network_binding_host_id(context, instance)
        for attempt in range(1, attempts + 1):
            try:
                nwinfo = self.network_api.allocate_for_instance(
                        context, instance,
                        requested_networks=requested_networks,
                        security_groups=security_groups,
                        bind_host_id=bind_host_id,
                        resource_provider_mapping=resource_provider_mapping,
                        network_arqs=network_arqs)
                LOG.debug('Instance network_info: |%s|', nwinfo,
                          instance=instance)
                instance.system_metadata['network_allocated'] = 'True'
                # NOTE(JoshNang) do not save the instance here, as it can cause
                # races. The caller shares a reference to instance and waits
                # for this async greenthread to finish before calling
                # instance.save().
                return nwinfo
            except Exception as e:
                log_info = {'attempt': attempt,
                            'attempts': attempts}
                if attempt == attempts:
                    LOG.exception('Instance failed network setup '
                                  'after %(attempts)d attempt(s)',
                                  log_info)
                    raise e
                LOG.warning('Instance failed network setup '
                            '(attempt %(attempt)d of %(attempts)d)',
                            log_info, instance=instance)
                time.sleep(retry_time)
                retry_time *= 2
                if retry_time > 30:
                    retry_time = 30
        # Not reached.

    def _build_networks_for_instance(self, context, instance,
            requested_networks, security_groups, resource_provider_mapping,
            network_arqs):

        # If we're here from a reschedule the network may already be allocated.
        if strutils.bool_from_string(
                instance.system_metadata.get('network_allocated', 'False')):
            # NOTE(alex_xu): The network_allocated is True means the network
            # resource already allocated at previous scheduling, and the
            # network setup is cleanup at previous. After rescheduling, the
            # network resource need setup on the new host.
            self.network_api.setup_instance_network_on_host(
                context, instance, instance.host)
            return self.network_api.get_instance_nw_info(context, instance)

        network_info = self._allocate_network(context, instance,
                requested_networks, security_groups,
                resource_provider_mapping,
                network_arqs)

        return network_info

    def _allocate_network(self, context, instance, requested_networks,
                          security_groups, resource_provider_mapping,
                          network_arqs):
        """Start network allocation asynchronously.  Return an instance
        of NetworkInfoAsyncWrapper that can be used to retrieve the
        allocated networks when the operation has finished.
        """
        # NOTE(comstud): Since we're allocating networks asynchronously,
        # this task state has little meaning, as we won't be in this
        # state for very long.
        instance.vm_state = vm_states.BUILDING
        instance.task_state = task_states.NETWORKING
        instance.save(expected_task_state=[None])

        return network_model.NetworkInfoAsyncWrapper(
                self._allocate_network_async, context, instance,
                requested_networks, security_groups, resource_provider_mapping,
                network_arqs)

    def _default_root_device_name(self, instance, image_meta, root_bdm):
        """Gets a default root device name from the driver.

        :param nova.objects.Instance instance:
            The instance for which to get the root device name.
        :param nova.objects.ImageMeta image_meta:
            The metadata of the image of the instance.
        :param nova.objects.BlockDeviceMapping root_bdm:
            The description of the root device.
        :returns: str -- The default root device name.
        :raises: InternalError, TooManyDiskDevices
        """
        try:
            return self.driver.default_root_device_name(instance,
                                                        image_meta,
                                                        root_bdm)
        except NotImplementedError:
            return compute_utils.get_next_device_name(instance, [])

    def _default_device_names_for_instance(self, instance,
                                           root_device_name,
                                           *block_device_lists):
        """Default the missing device names in the BDM from the driver.

        :param nova.objects.Instance instance:
            The instance for which to get default device names.
        :param str root_device_name: The root device name.
        :param list block_device_lists: List of block device mappings.
        :returns: None
        :raises: InternalError, TooManyDiskDevices
        """
        try:
            self.driver.default_device_names_for_instance(instance,
                                                          root_device_name,
                                                          *block_device_lists)
        except NotImplementedError:
            compute_utils.default_device_names_for_instance(
                instance, root_device_name, *block_device_lists)

    def _get_device_name_for_instance(self, instance, bdms, block_device_obj):
        """Get the next device name from the driver, based on the BDM.

        :param nova.objects.Instance instance:
            The instance whose volume is requesting a device name.
        :param nova.objects.BlockDeviceMappingList bdms:
            The block device mappings for the instance.
        :param nova.objects.BlockDeviceMapping block_device_obj:
            A block device mapping containing info about the requested block
            device.
        :returns: The next device name.
        :raises: InternalError, TooManyDiskDevices
        """
        # NOTE(ndipanov): Copy obj to avoid changing the original
        block_device_obj = block_device_obj.obj_clone()
        try:
            return self.driver.get_device_name_for_instance(
                instance, bdms, block_device_obj)
        except NotImplementedError:
            return compute_utils.get_device_name_for_instance(
                instance, bdms, block_device_obj.get("device_name"))

    def _default_block_device_names(self, instance, image_meta, block_devices):
        """Verify that all the devices have the device_name set. If not,
        provide a default name.

        It also ensures that there is a root_device_name and is set to the
        first block device in the boot sequence (boot_index=0).
        """
        root_bdm = block_device.get_root_bdm(block_devices)
        if not root_bdm:
            return

        # Get the root_device_name from the root BDM or the instance
        root_device_name = None
        update_root_bdm = False

        if root_bdm.device_name:
            root_device_name = root_bdm.device_name
            instance.root_device_name = root_device_name
        elif instance.root_device_name:
            root_device_name = instance.root_device_name
            root_bdm.device_name = root_device_name
            update_root_bdm = True
        else:
            root_device_name = self._default_root_device_name(instance,
                                                              image_meta,
                                                              root_bdm)

            instance.root_device_name = root_device_name
            root_bdm.device_name = root_device_name
            update_root_bdm = True

        if update_root_bdm:
            root_bdm.save()

        ephemerals = []
        swap = []
        block_device_mapping = []
        image = []

        for device in block_devices:
            if block_device.new_format_is_ephemeral(device):
                ephemerals.append(device)

            if block_device.new_format_is_swap(device):
                swap.append(device)

            if driver_block_device.is_block_device_mapping(device):
                block_device_mapping.append(device)

            if driver_block_device.is_local_image(device):
                image.append(device)

        self._default_device_names_for_instance(instance,
                                                root_device_name,
                                                image,
                                                ephemerals,
                                                swap,
                                                block_device_mapping)

    def _add_missing_dev_names(self, bdms, instance):
        for bdm in bdms:
            if bdm.device_name is not None:
                continue

            device_name = self._get_device_name_for_instance(instance,
                                                             bdms, bdm)
            values = {'device_name': device_name}
            bdm.update(values)
            bdm.save()

    def _prep_block_device(self, context, instance, bdms):
        """Set up the block device for an instance with error logging."""
        try:
            self._add_missing_dev_names(bdms, instance)
            block_device_info = driver.get_block_device_info(instance, bdms)
            mapping = driver.block_device_info_get_mapping(block_device_info)
            driver_block_device.attach_block_devices(
                mapping, context, instance, self.volume_api, self.driver,
                wait_func=self._await_block_device_map_created)

            return block_device_info

        except exception.OverQuota as e:
            LOG.warning('Failed to create block device for instance due'
                        ' to exceeding volume related resource quota.'
                        ' Error: %s', e.message, instance=instance)
            raise

        except Exception as ex:
            LOG.exception('Instance failed block device setup',
                          instance=instance)
            # InvalidBDM will eventually result in a BuildAbortException when
            # booting from volume, and will be recorded as an instance fault.
            # Maintain the original exception message which most likely has
            # useful details which the standard InvalidBDM error message lacks.
            raise exception.InvalidBDM(str(ex))

    def _update_instance_after_spawn(self, instance,
                                     vm_state=vm_states.ACTIVE):
        instance.power_state = self._get_power_state(instance)
        instance.vm_state = vm_state
        instance.task_state = None
        # NOTE(sean-k-mooney): configdrive.update_instance checks
        # instance.launched_at to determine if it is the first or
        # subsequent spawn of an instance. We need to call update_instance
        # first before setting instance.launched_at or instance.config_drive
        # will never be set to true based on the value of force_config_drive.
        # As a result the config drive will be lost on a hard reboot of the
        # instance even when force_config_drive=true. see bug #1835822.
        configdrive.update_instance(instance)
        instance.launched_at = timeutils.utcnow()

    def _update_scheduler_instance_info(self, context, instance):
        """Sends an InstanceList with created or updated Instance objects to
        the Scheduler client.

        In the case of init_host, the value passed will already be an
        InstanceList. Other calls will send individual Instance objects that
        have been created or resized. In this case, we create an InstanceList
        object containing that Instance.
        """
        if not self.send_instance_updates:
            return
        if isinstance(instance, obj_instance.Instance):
            instance = objects.InstanceList(objects=[instance])
        context = context.elevated()
        self.query_client.update_instance_info(context, self.host,
                                               instance)

    def _delete_scheduler_instance_info(self, context, instance_uuid):
        """Sends the uuid of the deleted Instance to the Scheduler client."""
        if not self.send_instance_updates:
            return
        context = context.elevated()
        self.query_client.delete_instance_info(context, self.host,
                                               instance_uuid)

    @periodic_task.periodic_task(spacing=CONF.scheduler_instance_sync_interval)
    def _sync_scheduler_instance_info(self, context):
        if not self.send_instance_updates:
            return
        context = context.elevated()
        instances = objects.InstanceList.get_by_host(context, self.host,
                                                     expected_attrs=[],
                                                     use_slave=True)
        uuids = [instance.uuid for instance in instances]
        self.query_client.sync_instance_info(context, self.host, uuids)

    def _notify_about_instance_usage(self, context, instance, event_suffix,
                                     network_info=None, extra_usage_info=None,
                                     fault=None):
        compute_utils.notify_about_instance_usage(
            self.notifier, context, instance, event_suffix,
            network_info=network_info,
            extra_usage_info=extra_usage_info, fault=fault)

    def _deallocate_network(self, context, instance,
                            requested_networks=None):
        # If we were told not to allocate networks let's save ourselves
        # the trouble of calling the network API.
        if requested_networks and requested_networks.no_allocate:
            LOG.debug("Skipping network deallocation for instance since "
                      "networking was not requested.", instance=instance)
            return

        LOG.debug('Deallocating network for instance', instance=instance)
        with timeutils.StopWatch() as timer:
            self.network_api.deallocate_for_instance(
                context, instance, requested_networks=requested_networks)
        # nova-network does an rpc call so we're OK tracking time spent here
        LOG.info('Took %0.2f seconds to deallocate network for instance.',
                 timer.elapsed(), instance=instance)

    def _get_instance_block_device_info(self, context, instance,
                                        refresh_conn_info=False,
                                        bdms=None):
        """Transform block devices to the driver block_device format."""

        if bdms is None:
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid)
        block_device_info = driver.get_block_device_info(instance, bdms)

        if not refresh_conn_info:
            # if the block_device_mapping has no value in connection_info
            # (returned as None), don't include in the mapping
            block_device_info['block_device_mapping'] = [
                bdm for bdm in driver.block_device_info_get_mapping(
                                    block_device_info)
                if bdm.get('connection_info')]
        else:
            driver_block_device.refresh_conn_infos(
                driver.block_device_info_get_mapping(block_device_info),
                context, instance, self.volume_api, self.driver)

        return block_device_info

    def _build_failed(self, node):
        if CONF.compute.consecutive_build_service_disable_threshold:
            # NOTE(danms): Update our counter, but wait for the next
            # update_available_resource() periodic to flush it to the DB
            self.rt.build_failed(node)

    def _build_succeeded(self, node):
        self.rt.build_succeeded(node)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_fault
    def build_and_run_instance(self, context, instance, image, request_spec,
                     filter_properties, accel_uuids, admin_password=None,
                     injected_files=None, requested_networks=None,
                     security_groups=None, block_device_mapping=None,
                     node=None, limits=None, host_list=None):

        @utils.synchronized(instance.uuid)
        def _locked_do_build_and_run_instance(*args, **kwargs):
            # NOTE(danms): We grab the semaphore with the instance uuid
            # locked because we could wait in line to build this instance
            # for a while and we want to make sure that nothing else tries
            # to do anything with this instance while we wait.
            with self._build_semaphore:
                try:
                    result = self._do_build_and_run_instance(*args, **kwargs)
                except Exception:
                    # NOTE(mriedem): This should really only happen if
                    # _decode_files in _do_build_and_run_instance fails, and
                    # that's before a guest is spawned so it's OK to remove
                    # allocations for the instance for this node from Placement
                    # below as there is no guest consuming resources anyway.
                    # The _decode_files case could be handled more specifically
                    # but that's left for another day.
                    result = build_results.FAILED
                    raise
                finally:
                    if result == build_results.FAILED:
                        # Remove the allocation records from Placement for the
                        # instance if the build failed. The instance.host is
                        # likely set to None in _do_build_and_run_instance
                        # which means if the user deletes the instance, it
                        # will be deleted in the API, not the compute service.
                        # Setting the instance.host to None in
                        # _do_build_and_run_instance means that the
                        # ResourceTracker will no longer consider this instance
                        # to be claiming resources against it, so we want to
                        # reflect that same thing in Placement.  No need to
                        # call this for a reschedule, as the allocations will
                        # have already been removed in
                        # self._do_build_and_run_instance().
                        self.reportclient.delete_allocation_for_instance(
                            context, instance.uuid, force=True)

                    if result in (build_results.FAILED,
                                  build_results.RESCHEDULED):
                        self._build_failed(node)
                    else:
                        self._build_succeeded(node)

        # NOTE(danms): We spawn here to return the RPC worker thread back to
        # the pool. Since what follows could take a really long time, we don't
        # want to tie up RPC workers.
        utils.spawn_n(_locked_do_build_and_run_instance,
                      context, instance, image, request_spec,
                      filter_properties, admin_password, injected_files,
                      requested_networks, security_groups,
                      block_device_mapping, node, limits, host_list,
                      accel_uuids)

    def _check_device_tagging(self, requested_networks, block_device_mapping):
        tagging_requested = False
        if requested_networks:
            for net in requested_networks:
                if 'tag' in net and net.tag is not None:
                    tagging_requested = True
                    break
        if block_device_mapping and not tagging_requested:
            for bdm in block_device_mapping:
                if 'tag' in bdm and bdm.tag is not None:
                    tagging_requested = True
                    break
        if (tagging_requested and
                not self.driver.capabilities.get('supports_device_tagging',
                                                 False)):
            raise exception.BuildAbortException('Attempt to boot guest with '
                                                'tagged devices on host that '
                                                'does not support tagging.')

    def _check_trusted_certs(self, instance):
        if (instance.trusted_certs and
                not self.driver.capabilities.get('supports_trusted_certs',
                                                 False)):
            raise exception.BuildAbortException(
                'Trusted image certificates provided on host that does not '
                'support certificate validation.')

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def _do_build_and_run_instance(self, context, instance, image,
            request_spec, filter_properties, admin_password, injected_files,
            requested_networks, security_groups, block_device_mapping,
            node=None, limits=None, host_list=None, accel_uuids=None):

        try:
            LOG.debug('Starting instance...', instance=instance)
            instance.vm_state = vm_states.BUILDING
            instance.task_state = None
            instance.save(expected_task_state=
                    (task_states.SCHEDULING, None))
        except exception.InstanceNotFound:
            msg = 'Instance disappeared before build.'
            LOG.debug(msg, instance=instance)
            return build_results.FAILED
        except exception.UnexpectedTaskStateError as e:
            LOG.debug(e.format_message(), instance=instance)
            return build_results.FAILED

        # b64 decode the files to inject:
        decoded_files = self._decode_files(injected_files)

        if limits is None:
            limits = {}

        if node is None:
            node = self._get_nodename(instance, refresh=True)

        try:
            with timeutils.StopWatch() as timer:
                self._build_and_run_instance(context, instance, image,
                        decoded_files, admin_password, requested_networks,
                        security_groups, block_device_mapping, node, limits,
                        filter_properties, request_spec, accel_uuids)
            LOG.info('Took %0.2f seconds to build instance.',
                     timer.elapsed(), instance=instance)
            return build_results.ACTIVE
        except exception.RescheduledException as e:
            retry = filter_properties.get('retry')
            if not retry:
                # no retry information, do not reschedule.
                LOG.debug("Retry info not present, will not reschedule",
                    instance=instance)
                self._cleanup_allocated_networks(context, instance,
                    requested_networks)
                compute_utils.add_instance_fault_from_exc(context,
                        instance, e, sys.exc_info(),
                        fault_message=e.kwargs['reason'])
                self._nil_out_instance_obj_host_and_node(instance)
                self._set_instance_obj_error_state(instance,
                                                   clean_task_state=True)
                return build_results.FAILED
            LOG.debug(e.format_message(), instance=instance)
            # This will be used for logging the exception
            retry['exc'] = traceback.format_exception(*sys.exc_info())
            # This will be used for setting the instance fault message
            retry['exc_reason'] = e.kwargs['reason']

            self._cleanup_allocated_networks(context, instance,
                                             requested_networks)

            self._nil_out_instance_obj_host_and_node(instance)
            instance.task_state = task_states.SCHEDULING
            instance.save()
            # The instance will have already claimed resources from this host
            # before this build was attempted. Now that it has failed, we need
            # to unclaim those resources before casting to the conductor, so
            # that if there are alternate hosts available for a retry, it can
            # claim resources on that new host for the instance.
            self.reportclient.delete_allocation_for_instance(
                context, instance.uuid, force=True)

            self.compute_task_api.build_instances(context, [instance],
                    image, filter_properties, admin_password,
                    injected_files, requested_networks, security_groups,
                    block_device_mapping, request_spec=request_spec,
                    host_lists=[host_list])
            return build_results.RESCHEDULED
        except (exception.InstanceNotFound,
                exception.UnexpectedDeletingTaskStateError):
            msg = 'Instance disappeared during build.'
            LOG.debug(msg, instance=instance)
            self._cleanup_allocated_networks(context, instance,
                    requested_networks)
            return build_results.FAILED
        except Exception as e:
            if isinstance(e, exception.BuildAbortException):
                LOG.error(e.format_message(), instance=instance)
            else:
                # Should not reach here.
                LOG.exception('Unexpected build failure, not rescheduling '
                              'build.', instance=instance)
            self._cleanup_allocated_networks(context, instance,
                    requested_networks)
            self._cleanup_volumes(context, instance,
                    block_device_mapping, raise_exc=False)
            compute_utils.add_instance_fault_from_exc(context, instance,
                    e, sys.exc_info())
            self._nil_out_instance_obj_host_and_node(instance)
            self._set_instance_obj_error_state(instance, clean_task_state=True)
            return build_results.FAILED

    @staticmethod
    def _get_scheduler_hints(filter_properties, request_spec=None):
        """Helper method to get scheduler hints.

        This method prefers to get the hints out of the request spec, but that
        might not be provided. Conductor will pass request_spec down to the
        first compute chosen for a build but older computes will not pass
        the request_spec to conductor's build_instances method for a
        a reschedule, so if we're on a host via a retry, request_spec may not
        be provided so we need to fallback to use the filter_properties
        to get scheduler hints.
        """
        hints = {}
        if request_spec is not None and 'scheduler_hints' in request_spec:
            hints = request_spec.scheduler_hints
        if not hints:
            hints = filter_properties.get('scheduler_hints') or {}
        return hints

    @staticmethod
    def _get_request_group_mapping(request_spec):
        """Return request group resource - provider mapping. This is currently
        used for Neutron ports that have resource request due to the port
        having QoS minimum bandwidth policy rule attached.

        :param request_spec: A RequestSpec object or None
        :returns: A dict keyed by RequestGroup requester_id, currently Neutron
        port_id, to resource provider UUID that provides resource for that
        RequestGroup. Or None if the request_spec was None.
        """
        # TODO(sbauza): Remove this conditional once we only support
        # RPC API 6.0
        if request_spec:
            return request_spec.get_request_group_mapping()
        else:
            return None

    def _build_and_run_instance(self, context, instance, image, injected_files,
            admin_password, requested_networks, security_groups,
            block_device_mapping, node, limits, filter_properties,
            request_spec=None, accel_uuids=None):

        image_name = image.get('name')
        self._notify_about_instance_usage(context, instance, 'create.start',
                extra_usage_info={'image_name': image_name})
        compute_utils.notify_about_instance_create(
            context, instance, self.host,
            phase=fields.NotificationPhase.START,
            bdms=block_device_mapping)

        # NOTE(mikal): cache the keystone roles associated with the instance
        # at boot time for later reference
        instance.system_metadata.update(
            {'boot_roles': ','.join(context.roles)})

        self._check_device_tagging(requested_networks, block_device_mapping)
        self._check_trusted_certs(instance)

        provider_mapping = self._get_request_group_mapping(request_spec)

        if provider_mapping:
            try:
                compute_utils\
                    .update_pci_request_spec_with_allocated_interface_name(
                        context, self.reportclient,
                        instance.pci_requests.requests, provider_mapping)
            except (exception.AmbiguousResourceProviderForPCIRequest,
                    exception.UnexpectedResourceProviderNameForPCIRequest
                    ) as e:
                raise exception.BuildAbortException(
                    reason=str(e), instance_uuid=instance.uuid)

        # TODO(Luyao) cut over to get_allocs_for_consumer
        allocs = self.reportclient.get_allocations_for_consumer(
                context, instance.uuid)

        try:
            scheduler_hints = self._get_scheduler_hints(filter_properties,
                                                        request_spec)
            with self.rt.instance_claim(context, instance, node, allocs,
                                        limits):
                # NOTE(russellb) It's important that this validation be done
                # *after* the resource tracker instance claim, as that is where
                # the host is set on the instance.
                self._validate_instance_group_policy(context, instance,
                                                     scheduler_hints)
                image_meta = objects.ImageMeta.from_dict(image)

                with self._build_resources(context, instance,
                        requested_networks, security_groups, image_meta,
                        block_device_mapping, provider_mapping,
                        accel_uuids) as resources:
                    instance.vm_state = vm_states.BUILDING
                    instance.task_state = task_states.SPAWNING
                    # NOTE(JoshNang) This also saves the changes to the
                    # instance from _allocate_network_async, as they aren't
                    # saved in that function to prevent races.
                    instance.save(expected_task_state=
                            task_states.BLOCK_DEVICE_MAPPING)
                    block_device_info = resources['block_device_info']
                    network_info = resources['network_info']
                    accel_info = resources['accel_info']
                    LOG.debug('Start spawning the instance on the hypervisor.',
                              instance=instance)
                    with timeutils.StopWatch() as timer:
                        self.driver.spawn(context, instance, image_meta,
                                          injected_files, admin_password,
                                          allocs, network_info=network_info,
                                          block_device_info=block_device_info,
                                          accel_info=accel_info)
                    LOG.info('Took %0.2f seconds to spawn the instance on '
                             'the hypervisor.', timer.elapsed(),
                             instance=instance)
        except (exception.InstanceNotFound,
                exception.UnexpectedDeletingTaskStateError) as e:
            with excutils.save_and_reraise_exception():
                self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
                compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
        except exception.ComputeResourcesUnavailable as e:
            LOG.debug(e.format_message(), instance=instance)
            self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
            compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
            raise exception.RescheduledException(
                    instance_uuid=instance.uuid, reason=e.format_message())
        except exception.BuildAbortException as e:
            with excutils.save_and_reraise_exception():
                LOG.debug(e.format_message(), instance=instance)
                self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
                compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
        except exception.NoMoreFixedIps as e:
            LOG.warning('No more fixed IP to be allocated',
                        instance=instance)
            self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
            compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
            msg = _('Failed to allocate the network(s) with error %s, '
                    'not rescheduling.') % e.format_message()
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=msg)
        except (exception.ExternalNetworkAttachForbidden,
                exception.VirtualInterfaceCreateException,
                exception.VirtualInterfaceMacAddressException,
                exception.FixedIpInvalidOnHost,
                exception.UnableToAutoAllocateNetwork,
                exception.NetworksWithQoSPolicyNotSupported) as e:
            LOG.exception('Failed to allocate network(s)',
                          instance=instance)
            self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
            compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
            msg = _('Failed to allocate the network(s), not rescheduling.')
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=msg)
        except (exception.FlavorDiskTooSmall,
                exception.FlavorMemoryTooSmall,
                exception.ImageNotActive,
                exception.ImageUnacceptable,
                exception.InvalidDiskInfo,
                exception.InvalidDiskFormat,
                cursive_exception.SignatureVerificationError,
                exception.CertificateValidationFailed,
                exception.VolumeEncryptionNotSupported,
                exception.InvalidInput,
                # TODO(mriedem): We should be validating RequestedVRamTooHigh
                # in the API during server create and rebuild.
                exception.RequestedVRamTooHigh) as e:
            self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
            compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=e.format_message())
        except Exception as e:
            LOG.exception('Failed to build and run instance',
                          instance=instance)
            self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
            compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)
            raise exception.RescheduledException(
                    instance_uuid=instance.uuid, reason=str(e))

        # NOTE(alaski): This is only useful during reschedules, remove it now.
        instance.system_metadata.pop('network_allocated', None)

        # If CONF.default_access_ip_network_name is set, grab the
        # corresponding network and set the access ip values accordingly.
        network_name = CONF.default_access_ip_network_name
        if (network_name and not instance.access_ip_v4 and
                not instance.access_ip_v6):
            # Note that when there are multiple ips to choose from, an
            # arbitrary one will be chosen.
            for vif in network_info:
                if vif['network']['label'] == network_name:
                    for ip in vif.fixed_ips():
                        if not instance.access_ip_v4 and ip['version'] == 4:
                            instance.access_ip_v4 = ip['address']
                        if not instance.access_ip_v6 and ip['version'] == 6:
                            instance.access_ip_v6 = ip['address']
                    break

        self._update_instance_after_spawn(instance)

        try:
            instance.save(expected_task_state=task_states.SPAWNING)
        except (exception.InstanceNotFound,
                exception.UnexpectedDeletingTaskStateError) as e:
            with excutils.save_and_reraise_exception():
                self._notify_about_instance_usage(context, instance,
                    'create.error', fault=e)
                compute_utils.notify_about_instance_create(
                    context, instance, self.host,
                    phase=fields.NotificationPhase.ERROR, exception=e,
                    bdms=block_device_mapping)

        self._update_scheduler_instance_info(context, instance)
        self._notify_about_instance_usage(context, instance, 'create.end',
                extra_usage_info={'message': _('Success')},
                network_info=network_info)
        compute_utils.notify_about_instance_create(context, instance,
                self.host, phase=fields.NotificationPhase.END,
                bdms=block_device_mapping)

    def _build_resources_cleanup(self, instance, network_info):
        # Make sure the async call finishes
        if network_info is not None:
            network_info.wait(do_raise=False)
            self.driver.clean_networks_preparation(instance,
                                                   network_info)
        self.driver.failed_spawn_cleanup(instance)

    @contextlib.contextmanager
    def _build_resources(self, context, instance, requested_networks,
                         security_groups, image_meta, block_device_mapping,
                         resource_provider_mapping, accel_uuids):
        resources = {}
        network_info = None
        spec_arqs = {}
        network_arqs = {}

        try:
            if accel_uuids:
                arqs = self._get_bound_arq_resources(
                    context, instance, accel_uuids)
                spec_arqs, network_arqs = self._split_network_arqs(
                    arqs, requested_networks)
                LOG.debug("ARQs for spec:%s, ARQs for network:%s",
                    spec_arqs, network_arqs)
        except (Exception, eventlet.timeout.Timeout) as exc:
            LOG.exception(exc)
            # ARQs created for instance or ports.
            # The port binding isn't done yet.
            # Unbind port won't clean port ARQs.
            compute_utils.delete_arqs_if_needed(
                context, instance, accel_uuids)
            msg = _('Failure getting accelerator requests.')
            raise exception.BuildAbortException(
                    reason=msg, instance_uuid=instance.uuid)

        try:
            LOG.debug('Start building networks asynchronously for instance.',
                      instance=instance)
            network_info = self._build_networks_for_instance(context, instance,
                    requested_networks, security_groups,
                    resource_provider_mapping, network_arqs)
            resources['network_info'] = network_info
        except (exception.InstanceNotFound,
                exception.UnexpectedDeletingTaskStateError):
            raise
        except exception.UnexpectedTaskStateError as e:
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=e.format_message())
        except Exception:
            # Because this allocation is async any failures are likely to occur
            # when the driver accesses network_info during spawn().
            LOG.exception('Failed to allocate network(s)',
                          instance=instance)
            msg = _('Failed to allocate the network(s), not rescheduling.')
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=msg)

        try:
            # Perform any driver preparation work for the driver.
            self.driver.prepare_for_spawn(instance)

            # Depending on a virt driver, some network configuration is
            # necessary before preparing block devices.
            self.driver.prepare_networks_before_block_device_mapping(
                instance, network_info)

            # Verify that all the BDMs have a device_name set and assign a
            # default to the ones missing it with the help of the driver.
            self._default_block_device_names(instance, image_meta,
                                             block_device_mapping)

            LOG.debug('Start building block device mappings for instance.',
                      instance=instance)
            instance.vm_state = vm_states.BUILDING
            instance.task_state = task_states.BLOCK_DEVICE_MAPPING
            instance.save()

            block_device_info = self._prep_block_device(context, instance,
                    block_device_mapping)
            resources['block_device_info'] = block_device_info
        except (exception.InstanceNotFound,
                exception.UnexpectedDeletingTaskStateError):
            with excutils.save_and_reraise_exception():
                self._build_resources_cleanup(instance, network_info)
        except (exception.UnexpectedTaskStateError,
                exception.InstanceUnacceptable,
                exception.OverQuota, exception.InvalidBDM) as e:
            self._build_resources_cleanup(instance, network_info)
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=e.format_message())
        except Exception:
            LOG.exception('Failure prepping block device',
                          instance=instance)
            self._build_resources_cleanup(instance, network_info)
            msg = _('Failure prepping block device.')
            raise exception.BuildAbortException(instance_uuid=instance.uuid,
                    reason=msg)

        resources['accel_info'] = list(spec_arqs.values())
        try:
            yield resources
        except Exception as exc:
            with excutils.save_and_reraise_exception() as ctxt:
                if not isinstance(exc, (
                        exception.InstanceNotFound,
                        exception.UnexpectedDeletingTaskStateError)):
                    LOG.exception('Instance failed to spawn',
                                  instance=instance)
                # Make sure the async call finishes
                if network_info is not None:
                    network_info.wait(do_raise=False)
                # if network_info is empty we're likely here because of
                # network allocation failure. Since nothing can be reused on
                # rescheduling it's better to deallocate network to eliminate
                # the chance of orphaned ports in neutron
                deallocate_networks = False if network_info else True
                try:
                    self._shutdown_instance(context, instance,
                            block_device_mapping, requested_networks,
                            try_deallocate_networks=deallocate_networks)
                except Exception as exc2:
                    ctxt.reraise = False
                    LOG.warning('Could not clean up failed build,'
                                ' not rescheduling. Error: %s',
                                str(exc2))
                    raise exception.BuildAbortException(
                            instance_uuid=instance.uuid,
                            reason=str(exc))
                finally:
                    # Call Cyborg to delete accelerator requests
                    if accel_uuids:
                        # ARQs created for instance or ports.
                        # Port bind may not successful.
                        # unbind port won't clean port ARQs.
                        compute_utils.delete_arqs_if_needed(
                            context, instance, accel_uuids)

    def _get_bound_arq_resources(self, context, instance, arq_uuids):
        """Get bound accelerator requests.

        The ARQ binding was kicked off in the conductor as an async
        operation. Here we wait for the notification from Cyborg.

        If the notification arrived before this point, which can happen
        in many/most cases (see [1]), it will be lost. To handle that,
        we use exit_wait_early.
        [1] https://review.opendev.org/#/c/631244/46/nova/compute/
            manager.py@2627

        :param instance: instance object
        :param arq_uuids: List of accelerator request (ARQ) UUIDs.
        :returns: List of ARQs for which bindings have completed,
                  successfully or otherwise
        """

        cyclient = cyborg.get_client(context)
        if arq_uuids is None:
            arqs = cyclient.get_arqs_for_instance(instance.uuid)
            arq_uuids = [arq['uuid'] for arq in arqs]
        events = [('accelerator-request-bound', arq_uuid)
                  for arq_uuid in arq_uuids]

        timeout = CONF.arq_binding_timeout
        with self.virtapi.wait_for_instance_event(
                instance, events, deadline=timeout):
            resolved_arqs = cyclient.get_arqs_for_instance(
                    instance.uuid, only_resolved=True)
            # Events for these resolved ARQs may have already arrived.
            # Such 'early' events need to be ignored.
            early_events = [('accelerator-request-bound', arq['uuid'])
                             for arq in resolved_arqs]
            if early_events:
                self.virtapi.exit_wait_early(early_events)

        # Since a timeout in wait_for_instance_event will raise, we get
        # here only if all binding events have been received.
        resolved_uuids = [arq['uuid'] for arq in resolved_arqs]
        if sorted(resolved_uuids) != sorted(arq_uuids):
            # Query Cyborg to get all.
            arqs = cyclient.get_arqs_for_instance(instance.uuid)
        else:
            arqs = resolved_arqs
        return arqs

    def _split_network_arqs(self, arqs, requested_networks):
        """splif arq request by exra spec from ARQ requested by port.

        Return ARQ groups tuple:(spec_arqs, port_arqs)
        Each item in the tuple is a dict like:
            {
                arq1_uuid: arq1
            }
        """
        port_arqs = {}
        spec_arqs = {}
        port_arqs_uuids = [req_net.arq_uuid for req_net in requested_networks]
        for arq in arqs:
            if arq['uuid'] in port_arqs_uuids:
                port_arqs.update({arq['uuid']: arq})
            else:
                spec_arqs.update({arq['uuid']: arq})

        return spec_arqs, port_arqs

    def _cleanup_allocated_networks(self, context, instance,
            requested_networks):
        """Cleanup networks allocated for instance.

        :param context: nova request context
        :param instance: nova.objects.instance.Instance object
        :param requested_networks: nova.objects.NetworkRequestList
        """
        LOG.debug('Unplugging VIFs for instance', instance=instance)

        network_info = instance.get_network_info()

        # NOTE(stephenfin) to avoid nova destroying the instance without
        # unplugging the interface, refresh network_info if it is empty.
        if not network_info:
            try:
                network_info = self.network_api.get_instance_nw_info(
                    context, instance,
                )
            except Exception as exc:
                LOG.warning(
                    'Failed to update network info cache when cleaning up '
                    'allocated networks. Stale VIFs may be left on this host.'
                    'Error: %s', str(exc)
                )
                return

        try:
            self.driver.unplug_vifs(instance, network_info)
        except NotImplementedError:
            # This is an optional method so ignore things if it doesn't exist
            LOG.debug(
                'Virt driver does not provide unplug_vifs method, so it '
                'is not possible determine if VIFs should be unplugged.'
            )
        except exception.NovaException as exc:
            # It's possible that the instance never got as far as plugging
            # VIFs, in which case we would see an exception which can be
            # mostly ignored
            LOG.warning(
                'Cleaning up VIFs failed for instance. Error: %s',
                str(exc), instance=instance,
            )
        else:
            LOG.debug('Unplugged VIFs for instance', instance=instance)

        try:
            self._deallocate_network(context, instance, requested_networks)
        except Exception:
            LOG.exception('Failed to deallocate networks', instance=instance)
            return

        instance.system_metadata['network_allocated'] = 'False'
        try:
            instance.save()
        except exception.InstanceNotFound:
            # NOTE(alaski): It's possible that we're cleaning up the networks
            # because the instance was deleted.  If that's the case then this
            # exception will be raised by instance.save()
            pass

    def _try_deallocate_network(self, context, instance,
                                requested_networks=None):

        # During auto-scale cleanup, we could be deleting a large number
        # of servers at the same time and overloading parts of the system,
        # so we retry a few times in case of connection failures to the
        # networking service.
        @loopingcall.RetryDecorator(
            max_retry_count=3, inc_sleep_time=2, max_sleep_time=12,
            exceptions=(keystone_exception.connection.ConnectFailure,))
        def _deallocate_network_with_retries():
            try:
                self._deallocate_network(
                    context, instance, requested_networks)
            except keystone_exception.connection.ConnectFailure as e:
                # Provide a warning that something is amiss.
                with excutils.save_and_reraise_exception():
                    LOG.warning('Failed to deallocate network for instance; '
                                'retrying. Error: %s', str(e),
                                instance=instance)

        try:
            # tear down allocated network structure
            _deallocate_network_with_retries()
        except Exception as ex:
            with excutils.save_and_reraise_exception():
                LOG.error('Failed to deallocate network for instance. '
                          'Error: %s', ex, instance=instance)
                self._set_instance_obj_error_state(instance)

    def _get_power_off_values(self, instance, clean_shutdown):
        """Get the timing configuration for powering down this instance."""
        if clean_shutdown:
            timeout = compute_utils.get_value_from_system_metadata(instance,
                          key='image_os_shutdown_timeout', type=int,
                          default=CONF.shutdown_timeout)
            retry_interval = CONF.compute.shutdown_retry_interval
        else:
            timeout = 0
            retry_interval = 0

        return timeout, retry_interval

    def _power_off_instance(self, instance, clean_shutdown=True):
        """Power off an instance on this host."""
        timeout, retry_interval = self._get_power_off_values(
            instance, clean_shutdown)
        self.driver.power_off(instance, timeout, retry_interval)

    def _shutdown_instance(self, context, instance,
                           bdms, requested_networks=None, notify=True,
                           try_deallocate_networks=True):
        """Shutdown an instance on this host.

        :param:context: security context
        :param:instance: a nova.objects.Instance object
        :param:bdms: the block devices for the instance to be torn
                     down
        :param:requested_networks: the networks on which the instance
                                   has ports
        :param:notify: true if a final usage notification should be
                       emitted
        :param:try_deallocate_networks: false if we should avoid
                                        trying to teardown networking
        """
        context = context.elevated()
        LOG.info('Terminating instance', instance=instance)

        if notify:
            self._notify_about_instance_usage(context, instance,
                                              "shutdown.start")
            compute_utils.notify_about_instance_action(context, instance,
                    self.host, action=fields.NotificationAction.SHUTDOWN,
                    phase=fields.NotificationPhase.START, bdms=bdms)

        network_info = instance.get_network_info()

        # NOTE(arnaudmorin) to avoid nova destroying the instance without
        # unplugging the interface, refresh network_info if it is empty.
        if not network_info:
            network_info = self.network_api.get_instance_nw_info(
                context, instance)

        # NOTE(vish) get bdms before destroying the instance
        vol_bdms = [bdm for bdm in bdms if bdm.is_volume]
        block_device_info = self._get_instance_block_device_info(
            context, instance, bdms=bdms)

        # NOTE(melwitt): attempt driver destroy before releasing ip, may
        #                want to keep ip allocated for certain failures
        try:
            LOG.debug('Start destroying the instance on the hypervisor.',
                      instance=instance)
            with timeutils.StopWatch() as timer:
                self.driver.destroy(context, instance, network_info,
                                    block_device_info)
            LOG.info('Took %0.2f seconds to destroy the instance on the '
                     'hypervisor.', timer.elapsed(), instance=instance)
        except exception.InstancePowerOffFailure:
            # if the instance can't power off, don't release the ip
            with excutils.save_and_reraise_exception():
                pass
        except Exception:
            with excutils.save_and_reraise_exception():
                # deallocate ip and fail without proceeding to
                # volume api calls, preserving current behavior
                if try_deallocate_networks:
                    self._try_deallocate_network(context, instance,
                                                 requested_networks)

        if try_deallocate_networks:
            self._try_deallocate_network(context, instance, requested_networks)

        timer.restart()
        connector = None
        for bdm in vol_bdms:
            try:
                if bdm.attachment_id:
                    self.volume_api.attachment_delete(context,
                                                      bdm.attachment_id)
                else:
                    # NOTE(vish): actual driver detach done in driver.destroy,
                    #             so just tell cinder that we are done with it.
                    if connector is None:
                        connector = self.driver.get_volume_connector(instance)
                    self.volume_api.terminate_connection(context,
                                                         bdm.volume_id,
                                                         connector)
                    self.volume_api.detach(context, bdm.volume_id,
                                           instance.uuid)

            except exception.VolumeAttachmentNotFound as exc:
                LOG.debug('Ignoring VolumeAttachmentNotFound: %s', exc,
                          instance=instance)
            except exception.DiskNotFound as exc:
                LOG.debug('Ignoring DiskNotFound: %s', exc,
                          instance=instance)
            except exception.VolumeNotFound as exc:
                LOG.debug('Ignoring VolumeNotFound: %s', exc,
                          instance=instance)
            except (cinder_exception.EndpointNotFound,
                    keystone_exception.EndpointNotFound) as exc:
                LOG.warning('Ignoring EndpointNotFound for '
                            'volume %(volume_id)s: %(exc)s',
                            {'exc': exc, 'volume_id': bdm.volume_id},
                            instance=instance)
            except cinder_exception.ClientException as exc:
                LOG.warning('Ignoring unknown cinder exception for '
                            'volume %(volume_id)s: %(exc)s',
                            {'exc': exc, 'volume_id': bdm.volume_id},
                            instance=instance)
            except Exception as exc:
                LOG.warning('Ignoring unknown exception for '
                            'volume %(volume_id)s: %(exc)s',
                            {'exc': exc, 'volume_id': bdm.volume_id},
                            instance=instance)
        if vol_bdms:
            LOG.info('Took %(time).2f seconds to detach %(num)s volumes '
                     'for instance.',
                     {'time': timer.elapsed(), 'num': len(vol_bdms)},
                     instance=instance)

        if notify:
            self._notify_about_instance_usage(context, instance,
                                              "shutdown.end")
            compute_utils.notify_about_instance_action(context, instance,
                    self.host, action=fields.NotificationAction.SHUTDOWN,
                    phase=fields.NotificationPhase.END, bdms=bdms)

    def _cleanup_volumes(self, context, instance, bdms, raise_exc=True,
                         detach=True):
        original_exception = None
        for bdm in bdms:
            if detach and bdm.volume_id:
                try:
                    LOG.debug("Detaching volume: %s", bdm.volume_id,
                              instance_uuid=instance.uuid)
                    destroy = bdm.delete_on_termination
                    self._detach_volume(context, bdm, instance,
                                        destroy_bdm=destroy)
                except Exception as exc:
                    original_exception = exc
                    LOG.warning('Failed to detach volume: %(volume_id)s '
                                'due to %(exc)s',
                                {'volume_id': bdm.volume_id, 'exc': exc})

            if bdm.volume_id and bdm.delete_on_termination:
                try:
                    LOG.debug("Deleting volume: %s", bdm.volume_id,
                              instance_uuid=instance.uuid)
                    self.volume_api.delete(context, bdm.volume_id)
                except Exception as exc:
                    original_exception = exc
                    LOG.warning('Failed to delete volume: %(volume_id)s '
                                'due to %(exc)s',
                                {'volume_id': bdm.volume_id, 'exc': exc})
        if original_exception is not None and raise_exc:
            raise original_exception

    def _delete_instance(self, context, instance, bdms):
        """Delete an instance on this host.

        :param context: nova request context
        :param instance: nova.objects.instance.Instance object
        :param bdms: nova.objects.block_device.BlockDeviceMappingList object
        """
        events = self.instance_events.clear_events_for_instance(instance)
        if events:
            LOG.debug('Events pending at deletion: %(events)s',
                      {'events': ','.join(events.keys())},
                      instance=instance)
        self._notify_about_instance_usage(context, instance,
                                          "delete.start")
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.DELETE,
                phase=fields.NotificationPhase.START, bdms=bdms)

        self._shutdown_instance(context, instance, bdms)

        # NOTE(vish): We have already deleted the instance, so we have
        #             to ignore problems cleaning up the volumes. It
        #             would be nice to let the user know somehow that
        #             the volume deletion failed, but it is not
        #             acceptable to have an instance that can not be
        #             deleted. Perhaps this could be reworked in the
        #             future to set an instance fault the first time
        #             and to only ignore the failure if the instance
        #             is already in ERROR.

        # NOTE(ameeda): The volumes have already been detached during
        #               the above _shutdown_instance() call and this is
        #               why detach is not requested from
        #               _cleanup_volumes() in this case

        self._cleanup_volumes(context, instance, bdms,
                raise_exc=False, detach=False)
        # Delete Cyborg ARQs if the instance has a device profile.
        compute_utils.delete_arqs_if_needed(context, instance)
        # if a delete task succeeded, always update vm state and task
        # state without expecting task state to be DELETING
        instance.vm_state = vm_states.DELETED
        instance.task_state = None
        instance.power_state = power_state.NOSTATE
        instance.terminated_at = timeutils.utcnow()
        instance.save()

        self._complete_deletion(context, instance)
        # only destroy the instance in the db if the _complete_deletion
        # doesn't raise and therefore allocation is successfully
        # deleted in placement
        instance.destroy()

        self._notify_about_instance_usage(context, instance, "delete.end")
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.DELETE,
                phase=fields.NotificationPhase.END, bdms=bdms)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def terminate_instance(self, context, instance, bdms):
        """Terminate an instance on this host."""
        @utils.synchronized(instance.uuid)
        def do_terminate_instance(instance, bdms):
            # NOTE(mriedem): If we are deleting the instance while it was
            # booting from volume, we could be racing with a database update of
            # the BDM volume_id. Since the compute API passes the BDMs over RPC
            # to compute here, the BDMs may be stale at this point. So check
            # for any volume BDMs that don't have volume_id set and if we
            # detect that, we need to refresh the BDM list before proceeding.
            # TODO(mriedem): Move this into _delete_instance and make the bdms
            # parameter optional.
            for bdm in list(bdms):
                if bdm.is_volume and not bdm.volume_id:
                    LOG.debug('There are potentially stale BDMs during '
                              'delete, refreshing the BlockDeviceMappingList.',
                              instance=instance)
                    bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                        context, instance.uuid)
                    break
            try:
                self._delete_instance(context, instance, bdms)
            except exception.InstanceNotFound:
                LOG.info("Instance disappeared during terminate",
                         instance=instance)
            except Exception:
                # As we're trying to delete always go to Error if something
                # goes wrong that _delete_instance can't handle.
                with excutils.save_and_reraise_exception():
                    LOG.exception('Setting instance vm_state to ERROR',
                                  instance=instance)
                    self._set_instance_obj_error_state(instance)

        do_terminate_instance(instance, bdms)

    # NOTE(johannes): This is probably better named power_off_instance
    # so it matches the driver method, but because of other issues, we
    # can't use that name in grizzly.
    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def stop_instance(self, context, instance, clean_shutdown):
        """Stopping an instance on this host."""

        @utils.synchronized(instance.uuid)
        def do_stop_instance():
            current_power_state = self._get_power_state(instance)
            LOG.debug('Stopping instance; current vm_state: %(vm_state)s, '
                      'current task_state: %(task_state)s, current DB '
                      'power_state: %(db_power_state)s, current VM '
                      'power_state: %(current_power_state)s',
                      {'vm_state': instance.vm_state,
                       'task_state': instance.task_state,
                       'db_power_state': instance.power_state,
                       'current_power_state': current_power_state},
                      instance_uuid=instance.uuid)

            # NOTE(mriedem): If the instance is already powered off, we are
            # possibly tearing down and racing with other operations, so we can
            # expect the task_state to be None if something else updates the
            # instance and we're not locking it.
            expected_task_state = [task_states.POWERING_OFF]
            # The list of power states is from _sync_instance_power_state.
            if current_power_state in (power_state.NOSTATE,
                                       power_state.SHUTDOWN,
                                       power_state.CRASHED):
                LOG.info('Instance is already powered off in the '
                         'hypervisor when stop is called.',
                         instance=instance)
                expected_task_state.append(None)

            self._notify_about_instance_usage(context, instance,
                                              "power_off.start")

            compute_utils.notify_about_instance_action(context, instance,
                        self.host, action=fields.NotificationAction.POWER_OFF,
                        phase=fields.NotificationPhase.START)

            self._power_off_instance(instance, clean_shutdown)
            instance.power_state = self._get_power_state(instance)
            instance.vm_state = vm_states.STOPPED
            instance.task_state = None
            instance.save(expected_task_state=expected_task_state)
            self._notify_about_instance_usage(context, instance,
                                              "power_off.end")

            compute_utils.notify_about_instance_action(context, instance,
                        self.host, action=fields.NotificationAction.POWER_OFF,
                        phase=fields.NotificationPhase.END)

        do_stop_instance()

    def _power_on(self, context, instance):
        network_info = self.network_api.get_instance_nw_info(context, instance)
        block_device_info = self._get_instance_block_device_info(context,
                                                                 instance)
        accel_info = self._get_accel_info(context, instance)
        self.driver.power_on(context, instance,
                             network_info,
                             block_device_info, accel_info)

    def _delete_snapshot_of_shelved_instance(self, context, instance,
                                             snapshot_id):
        """Delete snapshot of shelved instance."""
        try:
            self.image_api.delete(context, snapshot_id)
        except (exception.ImageNotFound,
                exception.ImageNotAuthorized) as exc:
            LOG.warning("Failed to delete snapshot "
                        "from shelved instance (%s).",
                        exc.format_message(), instance=instance)
        except Exception:
            LOG.exception("Something wrong happened when trying to "
                          "delete snapshot from shelved instance.",
                          instance=instance)

    # NOTE(johannes): This is probably better named power_on_instance
    # so it matches the driver method, but because of other issues, we
    # can't use that name in grizzly.
    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def start_instance(self, context, instance):
        """Starting an instance on this host."""
        self._notify_about_instance_usage(context, instance, "power_on.start")
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.POWER_ON,
            phase=fields.NotificationPhase.START)
        self._power_on(context, instance)
        instance.power_state = self._get_power_state(instance)
        instance.vm_state = vm_states.ACTIVE
        instance.task_state = None

        # Delete an image(VM snapshot) for a shelved instance
        snapshot_id = instance.system_metadata.get('shelved_image_id')
        if snapshot_id:
            self._delete_snapshot_of_shelved_instance(context, instance,
                                                      snapshot_id)

        # Delete system_metadata for a shelved instance
        compute_utils.remove_shelved_keys_from_system_metadata(instance)

        instance.save(expected_task_state=task_states.POWERING_ON)
        self._notify_about_instance_usage(context, instance, "power_on.end")
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.POWER_ON,
            phase=fields.NotificationPhase.END)

    @messaging.expected_exceptions(NotImplementedError,
                                   exception.TriggerCrashDumpNotSupported,
                                   exception.InstanceNotRunning)
    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def trigger_crash_dump(self, context, instance):
        """Trigger crash dump in an instance."""

        self._notify_about_instance_usage(context, instance,
                                          "trigger_crash_dump.start")
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.TRIGGER_CRASH_DUMP,
                phase=fields.NotificationPhase.START)

        # This method does not change task_state and power_state because the
        # effect of a trigger depends on user's configuration.
        self.driver.trigger_crash_dump(instance)

        self._notify_about_instance_usage(context, instance,
                                          "trigger_crash_dump.end")
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.TRIGGER_CRASH_DUMP,
                phase=fields.NotificationPhase.END)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def soft_delete_instance(self, context, instance):
        """Soft delete an instance on this host."""
        with compute_utils.notify_about_instance_delete(
                self.notifier, context, instance, 'soft_delete',
                source=fields.NotificationSource.COMPUTE):
            try:
                self.driver.soft_delete(instance)
            except NotImplementedError:
                # Fallback to just powering off the instance if the
                # hypervisor doesn't implement the soft_delete method
                self.driver.power_off(instance)
            instance.power_state = self._get_power_state(instance)
            instance.vm_state = vm_states.SOFT_DELETED
            instance.task_state = None
            instance.save(expected_task_state=[task_states.SOFT_DELETING])

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def restore_instance(self, context, instance):
        """Restore a soft-deleted instance on this host."""
        self._notify_about_instance_usage(context, instance, "restore.start")
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.RESTORE,
            phase=fields.NotificationPhase.START)
        try:
            self.driver.restore(instance)
        except NotImplementedError:
            # Fallback to just powering on the instance if the hypervisor
            # doesn't implement the restore method
            self._power_on(context, instance)
        instance.power_state = self._get_power_state(instance)
        instance.vm_state = vm_states.ACTIVE
        instance.task_state = None
        instance.save(expected_task_state=task_states.RESTORING)
        self._notify_about_instance_usage(context, instance, "restore.end")
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.RESTORE,
            phase=fields.NotificationPhase.END)

    @staticmethod
    def _set_migration_status(migration, status):
        """Set the status, and guard against a None being passed in.

        This is useful as some of the compute RPC calls will not pass
        a migration object in older versions. The check can be removed when
        we move past 4.x major version of the RPC API.
        """
        if migration:
            migration.status = status
            migration.save()

    @staticmethod
    def _reimage_failed_callback(event_name, instance):
        msg = ('Cinder reported failure during reimaging '
               'with %(event)s for instance %(uuid)s')
        msg_args = {'event': event_name, 'uuid': instance.uuid}
        LOG.error(msg, msg_args)
        raise exception.ReimageException(msg % msg_args)

    def _detach_root_volume(self, context, instance, root_bdm):
        volume_id = root_bdm.volume_id
        mp = root_bdm.device_name
        old_connection_info = jsonutils.loads(root_bdm.connection_info)
        try:
            self.driver.detach_volume(context, old_connection_info,
                                      instance, root_bdm.device_name)
        except exception.DiskNotFound as err:
            LOG.warning('Ignoring DiskNotFound exception while '
                        'detaching volume %(volume_id)s from '
                        '%(mp)s : %(err)s',
                        {'volume_id': volume_id, 'mp': mp,
                         'err': err}, instance=instance)
        except exception.DeviceDetachFailed:
            with excutils.save_and_reraise_exception():
                LOG.warning('Guest refused to detach volume %(vol)s',
                            {'vol': volume_id}, instance=instance)
                self.volume_api.roll_detaching(context, volume_id)
        except Exception:
            with excutils.save_and_reraise_exception():
                LOG.exception('Failed to detach volume '
                              '%(volume_id)s from %(mp)s',
                              {'volume_id': volume_id, 'mp': mp},
                              instance=instance)
                self.volume_api.roll_detaching(context, volume_id)

    def _rebuild_volume_backed_instance(self, context, instance, bdms,
                                        image_id):
        # Get root bdm and attachment ID associated to it
        root_bdm = compute_utils.get_root_bdm(context, instance, bdms)
        old_attachment_id = root_bdm.attachment_id

        # Create a new attachment and delete the previous attachment
        # We create a new attachment first to keep the volume in
        # reserved state after old attachment is deleted and avoid any
        # races in between the attachment create and delete.
        attachment_id = None
        try:
            attachment_id = self.volume_api.attachment_create(
                context, root_bdm.volume_id, instance.uuid)['id']
            self._detach_root_volume(context, instance, root_bdm)
            root_bdm.attachment_id = attachment_id
            root_bdm.save()
            self.volume_api.attachment_delete(context,
                                              old_attachment_id)
        except exception.InstanceNotFound:
            # This means we failed to save the new attachment because
            # the instance is deleted, so (try to) delete it and abort.
            try:
                self.volume_api.attachment_delete(context,
                                                  attachment_id)
            except cinder_exception.ClientException:
                LOG.error('Failed to delete new attachment %s',
                          attachment_id)
            msg = _('Failed to rebuild volume backed instance.')
            raise exception.BuildAbortException(
                instance_uuid=instance.uuid, reason=msg)
        except cinder_exception.ClientException:
            if attachment_id:
                LOG.error('Failed to delete old attachment %s',
                          old_attachment_id)
            else:
                LOG.error('Failed to create new attachment')
            msg = _('Failed to rebuild volume backed instance.')
            raise exception.BuildAbortException(
                instance_uuid=instance.uuid, reason=msg)
        events = [('volume-reimaged', root_bdm.volume_id)]

        # Get the image requested for rebuild
        try:
            image = self.image_api.get(context, image_id)
        except exception.ImageNotFound:
            msg = _('Image %s not found.') % image_id
            LOG.error(msg)
            raise exception.BuildAbortException(
                instance_uuid=instance.uuid, reason=msg)
        image_size = int(math.ceil(float(image.get('size')) / units.Gi))
        deadline = CONF.reimage_timeout_per_gb * image_size
        error_cb = self._reimage_failed_callback

        # Call cinder to perform reimage operation and wait until an
        # external event is triggered.
        try:
            with self.virtapi.wait_for_instance_event(instance, events,
                                                      deadline=deadline,
                                                      error_callback=error_cb):
                self.volume_api.reimage_volume(
                    context, root_bdm.volume_id, image_id,
                    reimage_reserved=True)

        except Exception as ex:
            LOG.error('Failed to rebuild volume backed instance: %s',
                      str(ex), instance=instance)
            msg = _('Failed to rebuild volume backed instance.')
            raise exception.BuildAbortException(
                instance_uuid=instance.uuid, reason=msg)

    def _rebuild_default_impl(
            self, context, instance, image_meta, injected_files,
            admin_password, allocations, bdms, detach_block_devices,
            attach_block_devices, network_info=None, evacuate=False,
            block_device_info=None, preserve_ephemeral=False,
            accel_uuids=None, reimage_boot_volume=False):
        if preserve_ephemeral:
            # The default code path does not support preserving ephemeral
            # partitions.
            raise exception.PreserveEphemeralNotSupported()

        accel_info = []
        detach_root_bdm = not reimage_boot_volume
        if evacuate:
            if instance.flavor.extra_specs.get('accel:device_profile'):
                try:
                    accel_info = self._get_bound_arq_resources(
                        context, instance, accel_uuids or [])
                except (Exception, eventlet.timeout.Timeout) as exc:
                    LOG.exception(exc)
                    self._build_resources_cleanup(instance, network_info)
                    msg = _('Failure getting accelerator resources.')
                    raise exception.BuildAbortException(
                        instance_uuid=instance.uuid, reason=msg)
            detach_block_devices(context, bdms,
                                 detach_root_bdm=detach_root_bdm)
        else:
            self._power_off_instance(instance, clean_shutdown=True)
            detach_block_devices(context, bdms,
                                 detach_root_bdm=detach_root_bdm)
            if reimage_boot_volume:
                # Previously, the calls reaching here were for image
                # backed instance rebuild and didn't have a root bdm
                # so now we need to handle the case for root bdm.
                # For the root BDM, we are doing attach/detach operations
                # manually as we want to maintain a 'reserved' state
                # throughout the reimage process from the cinder side so
                # we are excluding the root BDM from certain operations
                # here i.e. deleteing it's mapping before the destroy call.
                block_device_info_copy = copy.deepcopy(block_device_info)
                root_bdm = compute_utils.get_root_bdm(context, instance, bdms)
                mapping = block_device_info_copy["block_device_mapping"]
                # drop root bdm from the mapping
                mapping = [
                  bdm for bdm in mapping
                  if bdm["volume_id"] != root_bdm.volume_id
                ]
                self.driver.destroy(context, instance,
                                    network_info=network_info,
                                    block_device_info=block_device_info_copy)
            else:
                self.driver.destroy(context, instance,
                    network_info=network_info,
                    block_device_info=block_device_info)
            try:
                accel_info = self._get_accel_info(context, instance)
            except Exception as exc:
                LOG.exception(exc)
                self._build_resources_cleanup(instance, network_info)
                msg = _('Failure getting accelerator resources.')
                raise exception.BuildAbortException(
                    instance_uuid=instance.uuid, reason=msg)
            if reimage_boot_volume:
                is_volume_backed = compute_utils.is_volume_backed_instance(
                    context, instance, bdms)
                if is_volume_backed:
                    self._rebuild_volume_backed_instance(
                        context, instance, bdms, image_meta.id)

        instance.task_state = task_states.REBUILD_BLOCK_DEVICE_MAPPING
        instance.save(expected_task_state=[task_states.REBUILDING])

        new_block_device_info = attach_block_devices(context, instance, bdms)

        instance.task_state = task_states.REBUILD_SPAWNING
        instance.save(
            expected_task_state=[task_states.REBUILD_BLOCK_DEVICE_MAPPING])

        with instance.mutated_migration_context():
            self.driver.spawn(context, instance, image_meta, injected_files,
                              admin_password, allocations,
                              network_info=network_info,
                              block_device_info=new_block_device_info,
                              accel_info=accel_info)

    def _notify_instance_rebuild_error(self, context, instance, error, bdms):
        self._notify_about_instance_usage(context, instance,
                                          'rebuild.error', fault=error)
        compute_utils.notify_about_instance_rebuild(
            context, instance, self.host,
            phase=fields.NotificationPhase.ERROR, exception=error, bdms=bdms)

    @messaging.expected_exceptions(exception.PreserveEphemeralNotSupported,
                                   exception.BuildAbortException)
    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def rebuild_instance(self, context, instance, orig_image_ref, image_ref,
                         injected_files, new_pass, orig_sys_metadata,
                         bdms, recreate, on_shared_storage,
                         preserve_ephemeral, migration,
                         scheduled_node, limits, request_spec, accel_uuids,
                         reimage_boot_volume):
        """Destroy and re-make this instance.

        A 'rebuild' effectively purges all existing data from the system and
        remakes the VM with given 'metadata' and 'personalities'.

        :param context: `nova.RequestContext` object
        :param instance: Instance object
        :param orig_image_ref: Original image_ref before rebuild
        :param image_ref: New image_ref for rebuild
        :param injected_files: Files to inject
        :param new_pass: password to set on rebuilt instance
        :param orig_sys_metadata: instance system metadata from pre-rebuild
        :param bdms: block-device-mappings to use for rebuild
        :param recreate: True if the instance is being evacuated (e.g. the
            hypervisor it was on failed) - cleanup of old state will be
            skipped.
        :param on_shared_storage: True if instance files on shared storage.
                                  If not provided then information from the
                                  driver will be used to decide if the instance
                                  files are available or not on the target host
        :param preserve_ephemeral: True if the default ephemeral storage
                                   partition must be preserved on rebuild
        :param migration: a Migration object if one was created for this
                          rebuild operation (if it's a part of evacuate)
        :param scheduled_node: A node of the host chosen by the scheduler. If a
                               host was specified by the user, this will be
                               None
        :param limits: Overcommit limits set by the scheduler. If a host was
                       specified by the user, this will be None
        :param request_spec: a RequestSpec object used to schedule the instance
        :param accel_uuids: a list of cyborg ARQ uuids
        :param reimage_boot_volume: Boolean to specify whether the user has
                                    explicitly requested to rebuild a boot
                                    volume

        """
        # recreate=True means the instance is being evacuated from a failed
        # host to a new destination host (this host). The 'recreate' variable
        # name is confusing, so rename it to evacuate here at the top, which
        # is simpler than renaming a parameter in an RPC versioned method.
        evacuate = recreate
        context = context.elevated()

        if evacuate:
            LOG.info("Evacuating instance", instance=instance)
        else:
            LOG.info("Rebuilding instance", instance=instance)

        if evacuate:
            # This is an evacuation to a new host, so we need to perform a
            # resource claim.
            rebuild_claim = self.rt.rebuild_claim
        else:
            # This is a rebuild to the same host, so we don't need to make
            # a claim since the instance is already on this host.
            rebuild_claim = claims.NopClaim

        if image_ref:
            image_meta = objects.ImageMeta.from_image_ref(
                context, self.image_api, image_ref)
        elif evacuate:
            # For evacuate the API does not send down the image_ref since the
            # image does not change so just get it from what was stashed in
            # the instance system_metadata when the instance was created (or
            # last rebuilt). This also works for volume-backed instances.
            image_meta = instance.image_meta
        else:
            image_meta = objects.ImageMeta()

        # NOTE(mriedem): On an evacuate, we need to update
        # the instance's host and node properties to reflect it's
        # destination node for the evacuate.
        if not scheduled_node:
            if evacuate:
                try:
                    compute_node = self._get_compute_info(context, self.host)
                    scheduled_node = compute_node.hypervisor_hostname
                except exception.ComputeHostNotFound:
                    LOG.exception('Failed to get compute_info for %s',
                                  self.host)
            else:
                scheduled_node = instance.node

        allocs = self.reportclient.get_allocations_for_consumer(
                    context, instance.uuid)

        # If the resource claim or group policy validation fails before we
        # do anything to the guest or its networking/volumes we want to keep
        # the current status rather than put the instance into ERROR status.
        instance_state = instance.vm_state
        with self._error_out_instance_on_exception(
                context, instance, instance_state=instance_state):
            try:
                self._do_rebuild_instance_with_claim(
                    context, instance, orig_image_ref,
                    image_meta, injected_files, new_pass, orig_sys_metadata,
                    bdms, evacuate, on_shared_storage, preserve_ephemeral,
                    migration, request_spec, allocs, rebuild_claim,
                    scheduled_node, limits, accel_uuids, reimage_boot_volume)
            except (exception.ComputeResourcesUnavailable,
                    exception.RescheduledException) as e:
                if isinstance(e, exception.ComputeResourcesUnavailable):
                    LOG.debug("Could not rebuild instance on this host, not "
                              "enough resources available.", instance=instance)
                else:
                    # RescheduledException is raised by the late server group
                    # policy check during evacuation if a parallel scheduling
                    # violated the policy.
                    # We catch the RescheduledException here but we don't have
                    # the plumbing to do an actual reschedule so we abort the
                    # operation.
                    LOG.debug("Could not rebuild instance on this host, "
                              "late server group check failed.",
                              instance=instance)
                # NOTE(ndipanov): We just abort the build for now and leave a
                # migration record for potential cleanup later
                self._set_migration_status(migration, 'failed')
                # Since the claim failed, we need to remove the allocation
                # created against the destination node. Note that we can only
                # get here when evacuating to a destination node. Rebuilding
                # on the same host (not evacuate) uses the NopClaim which will
                # not raise ComputeResourcesUnavailable.
                self.rt.delete_allocation_for_evacuated_instance(
                    context, instance, scheduled_node, node_type='destination')
                self._notify_instance_rebuild_error(context, instance, e, bdms)
                # Wrap this in InstanceFaultRollback so that the
                # _error_out_instance_on_exception context manager keeps the
                # vm_state unchanged.
                raise exception.InstanceFaultRollback(
                    inner_exception=exception.BuildAbortException(
                        instance_uuid=instance.uuid,
                        reason=e.format_message()))
            except (exception.InstanceNotFound,
                    exception.UnexpectedDeletingTaskStateError) as e:
                LOG.debug('Instance was deleted while rebuilding',
                          instance=instance)
                self._set_migration_status(migration, 'failed')
                self._notify_instance_rebuild_error(context, instance, e, bdms)
            except Exception as e:
                self._set_migration_status(migration, 'failed')
                if evacuate or scheduled_node is not None:
                    self.rt.delete_allocation_for_evacuated_instance(
                        context, instance, scheduled_node,
                        node_type='destination')
                self._notify_instance_rebuild_error(context, instance, e, bdms)
                raise
            else:
                # NOTE(gibi): Let the resource tracker set the instance
                # host and drop the migration context as we need to hold the
                # COMPUTE_RESOURCE_SEMAPHORE to avoid the race with
                # _update_available_resources. See bug 1896463.
                self.rt.finish_evacuation(instance, scheduled_node, migration)

    def _do_rebuild_instance_with_claim(
            self, context, instance, orig_image_ref, image_meta,
            injected_files, new_pass, orig_sys_metadata, bdms, evacuate,
            on_shared_storage, preserve_ephemeral, migration, request_spec,
            allocations, rebuild_claim, scheduled_node, limits, accel_uuids,
            reimage_boot_volume):
        """Helper to avoid deep nesting in the top-level method."""

        provider_mapping = None
        if evacuate:
            provider_mapping = self._get_request_group_mapping(request_spec)

            if provider_mapping:
                compute_utils.\
                    update_pci_request_spec_with_allocated_interface_name(
                        context, self.reportclient,
                        instance.pci_requests.requests, provider_mapping)

        claim_context = rebuild_claim(
            context, instance, scheduled_node, allocations,
            limits=limits, image_meta=image_meta, migration=migration)

        with claim_context:
            self._do_rebuild_instance(
                context, instance, orig_image_ref, image_meta, injected_files,
                new_pass, orig_sys_metadata, bdms, evacuate, on_shared_storage,
                preserve_ephemeral, migration, request_spec, allocations,
                provider_mapping, accel_uuids, reimage_boot_volume)

    @staticmethod
    def _get_image_name(image_meta):
        if image_meta.obj_attr_is_set("name"):
            return image_meta.name
        else:
            return ''

    def _do_rebuild_instance(
            self, context, instance, orig_image_ref, image_meta,
            injected_files, new_pass, orig_sys_metadata, bdms, evacuate,
            on_shared_storage, preserve_ephemeral, migration, request_spec,
            allocations, request_group_resource_providers_mapping,
            accel_uuids, reimage_boot_volume):
        orig_vm_state = instance.vm_state

        if evacuate:
            if request_spec:
                # NOTE(gibi): Do a late check of server group policy as
                # parallel scheduling could violate such policy. This will
                # cause the evacuate to fail as rebuild does not implement
                # reschedule.
                hints = self._get_scheduler_hints({}, request_spec)
                self._validate_instance_group_policy(context, instance, hints)

            if not self.driver.capabilities.get("supports_evacuate", False):
                raise exception.InstanceEvacuateNotSupported

            self._check_instance_exists(instance)

            if on_shared_storage is None:
                LOG.debug('on_shared_storage is not provided, using driver '
                          'information to decide if the instance needs to '
                          'be evacuated')
                on_shared_storage = self.driver.instance_on_disk(instance)

            elif (on_shared_storage !=
                    self.driver.instance_on_disk(instance)):
                # To cover case when admin expects that instance files are
                # on shared storage, but not accessible and vice versa
                raise exception.InvalidSharedStorage(
                        _("Invalid state of instance files on shared"
                            " storage"))

            if on_shared_storage:
                LOG.info('disk on shared storage, evacuating using'
                         ' existing disk')
            elif instance.image_ref:
                orig_image_ref = instance.image_ref
                LOG.info("disk not on shared storage, evacuating from "
                         "image: '%s'", str(orig_image_ref))
            else:
                LOG.info('disk on volume, evacuating using existing '
                         'volume')

        # We check trusted certs capabilities for both evacuate (rebuild on
        # another host) and rebuild (rebuild on the same host) because for
        # evacuate we need to make sure an instance with trusted certs can
        # have the image verified with those certs during rebuild, and for
        # rebuild we could be rebuilding a server that started out with no
        # trusted certs on this host, and then was rebuilt with trusted certs
        # for a new image, in which case we need to validate that new image
        # with the trusted certs during the rebuild.
        self._check_trusted_certs(instance)

        # This instance.exists message should contain the original
        # image_ref, not the new one.  Since the DB has been updated
        # to point to the new one... we have to override it.
        orig_image_ref_url = self.image_api.generate_image_url(orig_image_ref,
                                                               context)
        extra_usage_info = {'image_ref_url': orig_image_ref_url}
        compute_utils.notify_usage_exists(
                self.notifier, context, instance, self.host,
                current_period=True, system_metadata=orig_sys_metadata,
                extra_usage_info=extra_usage_info)

        # This message should contain the new image_ref
        extra_usage_info = {'image_name': self._get_image_name(image_meta)}
        self._notify_about_instance_usage(context, instance,
                "rebuild.start", extra_usage_info=extra_usage_info)
        # NOTE: image_name is not included in the versioned notification
        # because we already provide the image_uuid in the notification
        # payload and the image details can be looked up via the uuid.
        compute_utils.notify_about_instance_rebuild(
            context, instance, self.host,
            phase=fields.NotificationPhase.START,
            bdms=bdms)

        instance.power_state = self._get_power_state(instance)
        instance.task_state = task_states.REBUILDING
        instance.save(expected_task_state=[task_states.REBUILDING])

        if evacuate:
            self.network_api.setup_networks_on_host(
                    context, instance, self.host)
            # For nova-network this is needed to move floating IPs
            # For neutron this updates the host in the port binding
            # TODO(cfriesen): this network_api call and the one above
            # are so similar, we should really try to unify them.
            self.network_api.setup_instance_network_on_host(
                context, instance, self.host, migration,
                provider_mappings=request_group_resource_providers_mapping)
            # TODO(mriedem): Consider decorating setup_instance_network_on_host
            # with @api.refresh_cache and then we wouldn't need this explicit
            # call to get_instance_nw_info.
            network_info = self.network_api.get_instance_nw_info(context,
                                                                 instance)
        else:
            network_info = instance.get_network_info()

        if bdms is None:
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid)

        block_device_info = \
            self._get_instance_block_device_info(
                    context, instance, bdms=bdms)

        def detach_block_devices(context, bdms, detach_root_bdm=True):
            for bdm in bdms:
                # Previously, the calls made to this method by rebuild
                # instance operation were for image backed instances which
                # assumed we only had attached volumes and no root BDM.
                # Now we need to handle case for root BDM which we are
                # doing manually so skipping the attachment create/delete
                # calls from here.
                # The detach_root_bdm parameter is only passed while
                # rebuilding the volume backed instance so we don't have
                # to worry about other callers as they won't satisfy this
                # condition.
                # For evacuate case, we have detach_root_bdm always True
                # since we don't have reimage_boot_volume parameter in
                # this case so this will not be executed.
                if not detach_root_bdm and bdm.is_root:
                    continue
                if bdm.is_volume:
                    # NOTE (ildikov): Having the attachment_id set in the BDM
                    # means that it's the new Cinder attach/detach flow
                    # (available from v3.44). In that case we explicitly
                    # attach and detach the volumes through attachment level
                    # operations. In this scenario _detach_volume will delete
                    # the existing attachment which would make the volume
                    # status change to 'available' if we don't pre-create
                    # another empty attachment before deleting the old one.
                    attachment_id = None
                    if bdm.attachment_id:
                        attachment_id = self.volume_api.attachment_create(
                            context, bdm['volume_id'], instance.uuid)['id']
                    self._detach_volume(context, bdm, instance,
                                        destroy_bdm=False)
                    if attachment_id:
                        bdm.attachment_id = attachment_id
                        bdm.save()

        files = self._decode_files(injected_files)

        kwargs = dict(
            context=context,
            instance=instance,
            image_meta=image_meta,
            injected_files=files,
            admin_password=new_pass,
            allocations=allocations,
            bdms=bdms,
            detach_block_devices=detach_block_devices,
            attach_block_devices=self._prep_block_device,
            block_device_info=block_device_info,
            network_info=network_info,
            preserve_ephemeral=preserve_ephemeral,
            evacuate=evacuate,
            accel_uuids=accel_uuids,
            reimage_boot_volume=reimage_boot_volume)
        try:
            with instance.mutated_migration_context():
                self.driver.rebuild(**kwargs)
        except NotImplementedError:
            # NOTE(rpodolyaka): driver doesn't provide specialized version
            # of rebuild, fall back to the default implementation
            self._rebuild_default_impl(**kwargs)
        self._update_instance_after_spawn(instance)
        instance.save(expected_task_state=[task_states.REBUILD_SPAWNING])

        if orig_vm_state == vm_states.STOPPED:
            LOG.info("bringing vm to original state: '%s'",
                     orig_vm_state, instance=instance)
            instance.vm_state = vm_states.ACTIVE
            instance.task_state = task_states.POWERING_OFF
            instance.progress = 0
            instance.save()
            self.stop_instance(context, instance, False)
        # TODO(melwitt): We should clean up instance console tokens here in the
        # case of evacuate. The instance is on a new host and will need to
        # establish a new console connection.
        self._update_scheduler_instance_info(context, instance)
        self._notify_about_instance_usage(
                context, instance, "rebuild.end",
                network_info=network_info,
                extra_usage_info=extra_usage_info)
        compute_utils.notify_about_instance_rebuild(
            context, instance, self.host,
            phase=fields.NotificationPhase.END,
            bdms=bdms)

    def _handle_bad_volumes_detached(self, context, instance, bad_devices,
                                     block_device_info):
        """Handle cases where the virt-layer had to detach non-working volumes
        in order to complete an operation.
        """
        for bdm in block_device_info['block_device_mapping']:
            if bdm.get('mount_device') in bad_devices:
                try:
                    volume_id = bdm['connection_info']['data']['volume_id']
                except KeyError:
                    continue

                # NOTE(sirp): ideally we'd just call
                # `compute_api.detach_volume` here but since that hits the
                # DB directly, that's off limits from within the
                # compute-manager.
                #
                # API-detach
                LOG.info("Detaching from volume api: %s", volume_id)
                self.volume_api.begin_detaching(context, volume_id)

                # Manager-detach
                self.detach_volume(context, volume_id, instance)

    def _get_accel_info(self, context, instance):
        dp_name = instance.flavor.extra_specs.get('accel:device_profile')
        if dp_name:
            cyclient = cyborg.get_client(context)
            accel_info = cyclient.get_arqs_for_instance(instance.uuid)
        else:
            accel_info = []
        return accel_info

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def reboot_instance(self, context, instance, block_device_info,
                        reboot_type):
        @utils.synchronized(instance.uuid)
        def do_reboot_instance(context, instance, block_device_info,
                               reboot_type):
            self._reboot_instance(context, instance, block_device_info,
                                  reboot_type)
        do_reboot_instance(context, instance, block_device_info, reboot_type)

    def _reboot_instance(self, context, instance, block_device_info,
                         reboot_type):
        """Reboot an instance on this host."""
        # acknowledge the request made it to the manager
        if reboot_type == "SOFT":
            instance.task_state = task_states.REBOOT_PENDING
            expected_states = task_states.soft_reboot_states
        else:
            instance.task_state = task_states.REBOOT_PENDING_HARD
            expected_states = task_states.hard_reboot_states

        context = context.elevated()
        LOG.info("Rebooting instance", instance=instance)

        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            context, instance.uuid)
        block_device_info = self._get_instance_block_device_info(
            context, instance, bdms=bdms)

        network_info = self.network_api.get_instance_nw_info(context, instance)

        accel_info = self._get_accel_info(context, instance)

        self._notify_about_instance_usage(context, instance, "reboot.start")
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.REBOOT,
            phase=fields.NotificationPhase.START,
            bdms=bdms
        )

        instance.power_state = self._get_power_state(instance)
        instance.save(expected_task_state=expected_states)

        if instance.power_state != power_state.RUNNING:
            state = instance.power_state
            running = power_state.RUNNING
            LOG.warning('trying to reboot a non-running instance:'
                        ' (state: %(state)s expected: %(running)s)',
                        {'state': state, 'running': running},
                        instance=instance)

        def bad_volumes_callback(bad_devices):
            self._handle_bad_volumes_detached(
                    context, instance, bad_devices, block_device_info)

        try:
            # Don't change it out of rescue mode
            if instance.vm_state == vm_states.RESCUED:
                new_vm_state = vm_states.RESCUED
            else:
                new_vm_state = vm_states.ACTIVE
            new_power_state = None
            if reboot_type == "SOFT":
                instance.task_state = task_states.REBOOT_STARTED
                expected_state = task_states.REBOOT_PENDING
            else:
                instance.task_state = task_states.REBOOT_STARTED_HARD
                expected_state = task_states.REBOOT_PENDING_HARD
            instance.save(expected_task_state=expected_state)
            self.driver.reboot(context, instance,
                               network_info,
                               reboot_type,
                               block_device_info=block_device_info,
                               accel_info=accel_info,
                               bad_volumes_callback=bad_volumes_callback)

        except Exception as error:
            with excutils.save_and_reraise_exception() as ctxt:
                exc_info = sys.exc_info()
                # if the reboot failed but the VM is running don't
                # put it into an error state
                new_power_state = self._get_power_state(instance)
                if new_power_state == power_state.RUNNING:
                    LOG.warning('Reboot failed but instance is running',
                                instance=instance)
                    compute_utils.add_instance_fault_from_exc(context,
                            instance, error, exc_info)
                    self._notify_about_instance_usage(context, instance,
                            'reboot.error', fault=error)
                    compute_utils.notify_about_instance_action(
                        context, instance, self.host,
                        action=fields.NotificationAction.REBOOT,
                        phase=fields.NotificationPhase.ERROR,
                        exception=error, bdms=bdms
                    )
                    ctxt.reraise = False
                else:
                    LOG.error('Cannot reboot instance: %s', error,
                              instance=instance)
                    self._set_instance_obj_error_state(instance)

        if not new_power_state:
            new_power_state = self._get_power_state(instance)
        try:
            instance.power_state = new_power_state
            instance.vm_state = new_vm_state
            instance.task_state = None
            instance.save()
        except exception.InstanceNotFound:
            LOG.warning("Instance disappeared during reboot",
                        instance=instance)

        self._notify_about_instance_usage(context, instance, "reboot.end")
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.REBOOT,
            phase=fields.NotificationPhase.END,
            bdms=bdms
        )

    @delete_image_on_error
    def _do_snapshot_instance(self, context, image_id, instance):
        self._snapshot_instance(context, image_id, instance,
                                task_states.IMAGE_BACKUP)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def backup_instance(self, context, image_id, instance, backup_type,
                        rotation):
        """Backup an instance on this host.

        :param backup_type: daily | weekly
        :param rotation: int representing how many backups to keep around
        """
        self._do_snapshot_instance(context, image_id, instance)
        self._rotate_backups(context, instance, backup_type, rotation)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    @delete_image_on_error
    def snapshot_instance(self, context, image_id, instance):
        """Snapshot an instance on this host.

        :param context: security context
        :param image_id: glance.db.sqlalchemy.models.Image.Id
        :param instance: a nova.objects.instance.Instance object
        """
        # NOTE(dave-mcnally) the task state will already be set by the api
        # but if the compute manager has crashed/been restarted prior to the
        # request getting here the task state may have been cleared so we set
        # it again and things continue normally
        try:
            instance.task_state = task_states.IMAGE_SNAPSHOT
            instance.save(
                        expected_task_state=task_states.IMAGE_SNAPSHOT_PENDING)
        except exception.InstanceNotFound:
            # possibility instance no longer exists, no point in continuing
            LOG.debug("Instance not found, could not set state %s "
                      "for instance.",
                      task_states.IMAGE_SNAPSHOT, instance=instance)
            return

        except exception.UnexpectedDeletingTaskStateError:
            LOG.debug("Instance being deleted, snapshot cannot continue",
                      instance=instance)
            return

        with self._snapshot_semaphore:
            self._snapshot_instance(context, image_id, instance,
                                    task_states.IMAGE_SNAPSHOT)

    def _snapshot_instance(self, context, image_id, instance,
                           expected_task_state):
        context = context.elevated()

        instance.power_state = self._get_power_state(instance)
        try:
            instance.save()

            LOG.info('instance snapshotting', instance=instance)

            if instance.power_state != power_state.RUNNING:
                state = instance.power_state
                running = power_state.RUNNING
                LOG.warning('trying to snapshot a non-running instance: '
                            '(state: %(state)s expected: %(running)s)',
                            {'state': state, 'running': running},
                            instance=instance)

            self._notify_about_instance_usage(
                context, instance, "snapshot.start")
            compute_utils.notify_about_instance_snapshot(context, instance,
                self.host, phase=fields.NotificationPhase.START,
                snapshot_image_id=image_id)

            def update_task_state(task_state,
                                  expected_state=expected_task_state):
                instance.task_state = task_state
                instance.save(expected_task_state=expected_state)

            with timeutils.StopWatch() as timer:
                self.driver.snapshot(context, instance, image_id,
                                     update_task_state)
            LOG.info('Took %0.2f seconds to snapshot the instance on '
                     'the hypervisor.', timer.elapsed(), instance=instance)

            instance.task_state = None
            instance.save(expected_task_state=task_states.IMAGE_UPLOADING)

            self._notify_about_instance_usage(context, instance,
                                              "snapshot.end")
            compute_utils.notify_about_instance_snapshot(context, instance,
                self.host, phase=fields.NotificationPhase.END,
                snapshot_image_id=image_id)
        except (exception.InstanceNotFound,
                exception.InstanceNotRunning,
                exception.UnexpectedDeletingTaskStateError):
            # the instance got deleted during the snapshot
            # Quickly bail out of here
            msg = 'Instance disappeared during snapshot'
            LOG.debug(msg, instance=instance)
            try:
                image = self.image_api.get(context, image_id)
                if image['status'] != 'active':
                    self.image_api.delete(context, image_id)
            except exception.ImageNotFound:
                LOG.debug('Image not found during clean up %s', image_id)
            except Exception:
                LOG.warning("Error while trying to clean up image %s",
                            image_id, instance=instance)
        except exception.ImageNotFound:
            instance.task_state = None
            instance.save()
            LOG.warning("Image not found during snapshot", instance=instance)

    @messaging.expected_exceptions(NotImplementedError)
    @wrap_exception()
    def volume_snapshot_create(self, context, instance, volume_id,
                               create_info):
        try:
            self.driver.volume_snapshot_create(context, instance, volume_id,
                                               create_info)
        except exception.InstanceNotRunning:
            # Libvirt driver can raise this exception
            LOG.debug('Instance disappeared during volume snapshot create',
                      instance=instance)

    @messaging.expected_exceptions(NotImplementedError)
    @wrap_exception()
    def volume_snapshot_delete(self, context, instance, volume_id,
                               snapshot_id, delete_info):
        try:
            self.driver.volume_snapshot_delete(context, instance, volume_id,
                                               snapshot_id, delete_info)
        except exception.InstanceNotRunning:
            # Libvirt driver can raise this exception
            LOG.debug('Instance disappeared during volume snapshot delete',
                      instance=instance)

    @wrap_instance_fault
    def _rotate_backups(self, context, instance, backup_type, rotation):
        """Delete excess backups associated to an instance.

        Instances are allowed a fixed number of backups (the rotation number);
        this method deletes the oldest backups that exceed the rotation
        threshold.

        :param context: security context
        :param instance: Instance dict
        :param backup_type: a user-defined type, like "daily" or "weekly" etc.
        :param rotation: int representing how many backups to keep around;
            None if rotation shouldn't be used (as in the case of snapshots)
        """
        filters = {'property-image_type': 'backup',
                   'property-backup_type': backup_type,
                   'property-instance_uuid': instance.uuid}

        images = self.image_api.get_all(context, filters=filters,
                                        sort_key='created_at', sort_dir='desc')
        num_images = len(images)
        LOG.debug("Found %(num_images)d images (rotation: %(rotation)d)",
                  {'num_images': num_images, 'rotation': rotation},
                  instance=instance)

        if num_images > rotation:
            # NOTE(sirp): this deletes all backups that exceed the rotation
            # limit
            excess = len(images) - rotation
            LOG.debug("Rotating out %d backups", excess,
                      instance=instance)
            for i in range(excess):
                image = images.pop()
                image_id = image['id']
                LOG.debug("Deleting image %s", image_id,
                          instance=instance)
                try:
                    self.image_api.delete(context, image_id)
                except exception.ImageNotFound:
                    LOG.info("Failed to find image %(image_id)s to "
                             "delete", {'image_id': image_id},
                             instance=instance)
                except (exception.ImageDeleteConflict, Exception) as exc:
                    LOG.info("Failed to delete image %(image_id)s during "
                             "deleting excess backups. "
                             "Continuing for next image.. %(exc)s",
                             {'image_id': image_id, 'exc': exc},
                             instance=instance)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def set_admin_password(self, context, instance, new_pass):
        """Set the root/admin password for an instance on this host.

        This is generally only called by API password resets after an
        image has been built.

        @param context: Nova auth context.
        @param instance: Nova instance object.
        @param new_pass: The admin password for the instance.
        """

        context = context.elevated()
        current_power_state = self._get_power_state(instance)
        expected_state = power_state.RUNNING

        if current_power_state != expected_state:
            instance.task_state = None
            instance.save(expected_task_state=task_states.UPDATING_PASSWORD)
            _msg = _('instance %s is not running') % instance.uuid
            raise exception.InstancePasswordSetFailed(
                instance=instance.uuid, reason=_msg)

        try:
            self.driver.set_admin_password(instance, new_pass)
            LOG.info("Admin password set", instance=instance)
            instance.task_state = None
            instance.save(
                expected_task_state=task_states.UPDATING_PASSWORD)
        except exception.InstanceAgentNotEnabled:
            with excutils.save_and_reraise_exception():
                LOG.debug('Guest agent is not enabled for the instance.',
                          instance=instance)
                instance.task_state = None
                instance.save(
                    expected_task_state=task_states.UPDATING_PASSWORD)
        except exception.SetAdminPasswdNotSupported:
            with excutils.save_and_reraise_exception():
                LOG.info('set_admin_password is not supported '
                         'by this driver or guest instance.',
                         instance=instance)
                instance.task_state = None
                instance.save(
                    expected_task_state=task_states.UPDATING_PASSWORD)
        except NotImplementedError:
            LOG.warning('set_admin_password is not implemented '
                        'by this driver or guest instance.',
                        instance=instance)
            instance.task_state = None
            instance.save(
                expected_task_state=task_states.UPDATING_PASSWORD)
            raise NotImplementedError(_('set_admin_password is not '
                                        'implemented by this driver or guest '
                                        'instance.'))
        except exception.UnexpectedTaskStateError:
            # interrupted by another (most likely delete) task
            # do not retry
            raise
        except Exception:
            # Catch all here because this could be anything.
            LOG.exception('set_admin_password failed', instance=instance)
            # We create a new exception here so that we won't
            # potentially reveal password information to the
            # API caller.  The real exception is logged above
            _msg = _('error setting admin password')
            raise exception.InstancePasswordSetFailed(
                instance=instance.uuid, reason=_msg)

    def _get_rescue_image(self, context, instance, rescue_image_ref=None):
        """Determine what image should be used to boot the rescue VM."""
        # 1. If rescue_image_ref is passed in, use that for rescue.
        # 2. Else, use the base image associated with instance's current image.
        #       The idea here is to provide the customer with a rescue
        #       environment which they are familiar with.
        #       So, if they built their instance off of a Debian image,
        #       their rescue VM will also be Debian.
        # 3. As a last resort, use instance's current image.
        if not rescue_image_ref:
            system_meta = utils.instance_sys_meta(instance)
            rescue_image_ref = system_meta.get('image_base_image_ref')

        if not rescue_image_ref:
            LOG.warning('Unable to find a different image to use for '
                        'rescue VM, using instance\'s current image',
                        instance=instance)
            rescue_image_ref = instance.image_ref

        return objects.ImageMeta.from_image_ref(
            context, self.image_api, rescue_image_ref)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def rescue_instance(self, context, instance, rescue_password,
                        rescue_image_ref, clean_shutdown):
        context = context.elevated()
        LOG.info('Rescuing', instance=instance)

        admin_password = (rescue_password if rescue_password else
                      utils.generate_password())

        network_info = self.network_api.get_instance_nw_info(context, instance)

        rescue_image_meta = self._get_rescue_image(context, instance,
                                                   rescue_image_ref)

        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                                              context, instance.uuid)
        block_device_info = self._get_instance_block_device_info(
                                context, instance, bdms=bdms)

        extra_usage_info = {'rescue_image_name':
                            self._get_image_name(rescue_image_meta)}
        self._notify_about_instance_usage(context, instance,
                "rescue.start", extra_usage_info=extra_usage_info,
                network_info=network_info)
        compute_utils.notify_about_instance_rescue_action(
            context, instance, self.host, rescue_image_ref,
            phase=fields.NotificationPhase.START)

        try:
            self._power_off_instance(instance, clean_shutdown)

            self.driver.rescue(context, instance, network_info,
                               rescue_image_meta, admin_password,
                               block_device_info)
        except Exception as e:
            LOG.exception("Error trying to Rescue Instance",
                          instance=instance)
            self._set_instance_obj_error_state(instance)
            raise exception.InstanceNotRescuable(
                instance_id=instance.uuid,
                reason=_("Driver Error: %s") % e)

        compute_utils.notify_usage_exists(self.notifier, context, instance,
                                          self.host, current_period=True)

        instance.vm_state = vm_states.RESCUED
        instance.task_state = None
        instance.power_state = self._get_power_state(instance)
        instance.launched_at = timeutils.utcnow()
        instance.save(expected_task_state=task_states.RESCUING)

        self._notify_about_instance_usage(context, instance,
                "rescue.end", extra_usage_info=extra_usage_info,
                network_info=network_info)
        compute_utils.notify_about_instance_rescue_action(
            context, instance, self.host, rescue_image_ref,
            phase=fields.NotificationPhase.END)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def unrescue_instance(self, context, instance):
        orig_context = context
        context = context.elevated()
        LOG.info('Unrescuing', instance=instance)

        network_info = self.network_api.get_instance_nw_info(context, instance)
        self._notify_about_instance_usage(context, instance,
                "unrescue.start", network_info=network_info)
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.UNRESCUE,
            phase=fields.NotificationPhase.START)

        with self._error_out_instance_on_exception(context, instance):
            self.driver.unrescue(orig_context, instance)

        instance.vm_state = vm_states.ACTIVE
        instance.task_state = None
        instance.power_state = self._get_power_state(instance)
        instance.save(expected_task_state=task_states.UNRESCUING)

        self._notify_about_instance_usage(context,
                                          instance,
                                          "unrescue.end",
                                          network_info=network_info)
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.UNRESCUE,
            phase=fields.NotificationPhase.END)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def confirm_resize(self, context, instance, migration):
        """Confirms a migration/resize and deletes the 'old' instance.

        This is called from the API and runs on the source host.

        Nothing needs to happen on the destination host at this point since
        the instance is already running there. This routine just cleans up the
        source host.
        """
        @utils.synchronized(instance.uuid)
        def do_confirm_resize(context, instance, migration):
            LOG.debug("Going to confirm migration %s", migration.id,
                      instance=instance)

            if migration.status == 'confirmed':
                LOG.info("Migration %s is already confirmed",
                         migration.id, instance=instance)
                return

            if migration.status not in ('finished', 'confirming'):
                LOG.warning("Unexpected confirmation status '%(status)s' "
                            "of migration %(id)s, exit confirmation process",
                            {"status": migration.status, "id": migration.id},
                            instance=instance)
                return

            # NOTE(wangpan): Get the instance from db, if it has been
            #                deleted, we do nothing and return here
            expected_attrs = ['metadata', 'system_metadata', 'flavor']
            try:
                instance = objects.Instance.get_by_uuid(
                        context, instance.uuid,
                        expected_attrs=expected_attrs)
            except exception.InstanceNotFound:
                LOG.info("Instance is not found during confirmation",
                         instance=instance)
                return

            with self._error_out_instance_on_exception(context, instance):
                try:
                    self._confirm_resize(
                        context, instance, migration=migration)
                except Exception:
                    # Something failed when cleaning up the source host so
                    # log a traceback and leave a hint about hard rebooting
                    # the server to correct its state in the DB.
                    with excutils.save_and_reraise_exception(logger=LOG):
                        LOG.exception(
                            'Confirm resize failed on source host %s. '
                            'Resource allocations in the placement service '
                            'will be removed regardless because the instance '
                            'is now on the destination host %s. You can try '
                            'hard rebooting the instance to correct its '
                            'state.', self.host, migration.dest_compute,
                            instance=instance)
                finally:
                    # Whether an error occurred or not, at this point the
                    # instance is on the dest host. Avoid leaking allocations
                    # in placement by deleting them here...
                    self._delete_allocation_after_move(
                        context, instance, migration)
                    # ...inform the scheduler about the move...
                    self._delete_scheduler_instance_info(
                        context, instance.uuid)
                    # ...and unset the cached flavor information (this is done
                    # last since the resource tracker relies on it for its
                    # periodic tasks)
                    self._delete_stashed_flavor_info(instance)

        do_confirm_resize(context, instance, migration)

    def _get_updated_nw_info_with_pci_mapping(self, nw_info, pci_mapping):
        # NOTE(adrianc): This method returns a copy of nw_info if modifications
        # are made else it returns the original nw_info.
        updated_nw_info = nw_info
        if nw_info and pci_mapping:
            updated_nw_info = copy.deepcopy(nw_info)
            for vif in updated_nw_info:
                if vif['vnic_type'] in network_model.VNIC_TYPES_SRIOV:
                    try:
                        vif_pci_addr = vif['profile']['pci_slot']
                        new_addr = pci_mapping[vif_pci_addr].address
                        vif['profile']['pci_slot'] = new_addr
                        LOG.debug("Updating VIF's PCI address for VIF %(id)s. "
                                  "Original value %(orig_val)s, "
                                  "new value %(new_val)s",
                                  {'id': vif['id'],
                                   'orig_val': vif_pci_addr,
                                   'new_val': new_addr})
                    except (KeyError, AttributeError):
                        with excutils.save_and_reraise_exception():
                            # NOTE(adrianc): This should never happen. If we
                            # get here it means there is some inconsistency
                            # with either 'nw_info' or 'pci_mapping'.
                            LOG.error("Unexpected error when updating network "
                                      "information with PCI mapping.")
        return updated_nw_info

    def _confirm_resize(self, context, instance, migration=None):
        """Destroys the source instance."""
        self._notify_about_instance_usage(context, instance,
                                          "resize.confirm.start")
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.RESIZE_CONFIRM,
            phase=fields.NotificationPhase.START)

        # NOTE(tr3buchet): tear down networks on source host
        self.network_api.setup_networks_on_host(context, instance,
                           migration.source_compute, teardown=True)

        # TODO(stephenfin): These next three calls should be bundled
        network_info = self.network_api.get_instance_nw_info(context,
                                                             instance)

        # NOTE(adrianc): Populate old PCI device in VIF profile
        # to allow virt driver to properly unplug it from Hypervisor.
        pci_mapping = (instance.migration_context.
                       get_pci_mapping_for_migration(True))
        network_info = self._get_updated_nw_info_with_pci_mapping(
            network_info, pci_mapping)

        self.driver.confirm_migration(context, migration, instance,
                                      network_info)

        # Free up the old_flavor usage from the resource tracker for this host.
        self.rt.drop_move_claim_at_source(context, instance, migration)

        # NOTE(mriedem): The old_vm_state could be STOPPED but the user
        # might have manually powered up the instance to confirm the
        # resize/migrate, so we need to check the current power state
        # on the instance and set the vm_state appropriately. We default
        # to ACTIVE because if the power state is not SHUTDOWN, we
        # assume _sync_instance_power_state will clean it up.
        p_state = instance.power_state
        vm_state = None
        if p_state == power_state.SHUTDOWN:
            vm_state = vm_states.STOPPED
            LOG.debug("Resized/migrated instance is powered off. "
                      "Setting vm_state to '%s'.", vm_state,
                      instance=instance)
        else:
            vm_state = vm_states.ACTIVE

        instance.vm_state = vm_state
        instance.task_state = None
        instance.save(expected_task_state=[None, task_states.DELETING,
                                           task_states.SOFT_DELETING])

        self._notify_about_instance_usage(
            context, instance, "resize.confirm.end",
            network_info=network_info)
        compute_utils.notify_about_instance_action(context, instance,
               self.host, action=fields.NotificationAction.RESIZE_CONFIRM,
               phase=fields.NotificationPhase.END)

    def _delete_allocation_after_move(self, context, instance, migration):
        """Deletes resource allocations held by the migration record against
        the source compute node resource provider after a confirmed cold /
        successful live migration.
        """
        try:
            # NOTE(danms): We're finishing on the source node, so try
            # to delete the allocation based on the migration uuid
            self.reportclient.delete_allocation_for_instance(
                context, migration.uuid, consumer_type='migration',
                force=False)
        except exception.AllocationDeleteFailed:
            LOG.error('Deleting allocation in placement for migration '
                      '%(migration_uuid)s failed. The instance '
                      '%(instance_uuid)s will be put to ERROR state '
                      'but the allocation held by the migration is '
                      'leaked.',
                      {'instance_uuid': instance.uuid,
                       'migration_uuid': migration.uuid})
            raise

    def _delete_stashed_flavor_info(self, instance):
        """Remove information about the flavor change after a resize."""
        instance.old_flavor = None
        instance.new_flavor = None
        instance.system_metadata.pop('old_vm_state', None)
        instance.save()

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def confirm_snapshot_based_resize_at_source(
            self, ctxt, instance, migration):
        """Confirms a snapshot-based resize on the source host.

        Cleans the guest from the source hypervisor including disks and drops
        the MoveClaim which will free up "old_flavor" usage from the
        ResourceTracker.

        Deletes the allocations held by the migration consumer against the
        source compute node resource provider.

        :param ctxt: nova auth request context targeted at the source cell
        :param instance: Instance object being resized which should have the
            "old_flavor" attribute set
        :param migration: Migration object for the resize operation
        """

        @utils.synchronized(instance.uuid)
        def do_confirm():
            LOG.info('Confirming resize on source host.', instance=instance)
            with self._error_out_instance_on_exception(ctxt, instance):
                # TODO(mriedem): Could probably make this try/except/finally
                # a context manager to share with confirm_resize().
                try:
                    self._confirm_snapshot_based_resize_at_source(
                        ctxt, instance, migration)
                except Exception:
                    # Something failed when cleaning up the source host so
                    # log a traceback and leave a hint about hard rebooting
                    # the server to correct its state in the DB.
                    with excutils.save_and_reraise_exception(logger=LOG):
                        LOG.exception(
                            'Confirm resize failed on source host %s. '
                            'Resource allocations in the placement service '
                            'will be removed regardless because the instance '
                            'is now on the destination host %s. You can try '
                            'hard rebooting the instance to correct its '
                            'state.', self.host, migration.dest_compute,
                            instance=instance)
                finally:
                    # Whether an error occurred or not, at this point the
                    # instance is on the dest host so to avoid leaking
                    # allocations in placement, delete them here.
                    # TODO(mriedem): Should we catch and just log
                    # AllocationDeleteFailed? What is the user's recourse if
                    # we got this far but this fails? At this point the
                    # instance is on the target host and the allocations
                    # could just be manually cleaned up by the operator.
                    self._delete_allocation_after_move(ctxt, instance,
                                                       migration)
        do_confirm()

    def _confirm_snapshot_based_resize_at_source(
            self, ctxt, instance, migration):
        """Private version of confirm_snapshot_based_resize_at_source

        This allows the main method to be decorated with error handlers.

        :param ctxt: nova auth request context targeted at the source cell
        :param instance: Instance object being resized which should have the
            "old_flavor" attribute set
        :param migration: Migration object for the resize operation
        """
        # Cleanup the guest from the hypervisor including local disks.
        network_info = self.network_api.get_instance_nw_info(ctxt, instance)
        LOG.debug('Cleaning up guest from source hypervisor including disks.',
                  instance=instance)

        # FIXME(mriedem): Per bug 1809095, _confirm_resize calls
        # _get_updated_nw_info_with_pci_mapping here prior to unplugging
        # VIFs on the source, but in our case we have already unplugged
        # VIFs during prep_snapshot_based_resize_at_source, so what do we
        # need to do about those kinds of ports? Do we need to wait to unplug
        # VIFs until confirm like normal resize?

        # Note that prep_snapshot_based_resize_at_source already destroyed the
        # guest which disconnected volumes and unplugged VIFs but did not
        # destroy disks in case something failed during the resize and the
        # instance needed to be rebooted or rebuilt on the source host. Now
        # that we are confirming the resize we want to cleanup the disks left
        # on the source host. We call cleanup() instead of destroy() to avoid
        # any InstanceNotFound confusion from the driver since the guest was
        # already destroyed on this host. block_device_info=None and
        # destroy_vifs=False means cleanup() will not try to disconnect volumes
        # or unplug VIFs.
        self.driver.cleanup(
            ctxt, instance, network_info, block_device_info=None,
            destroy_disks=True, destroy_vifs=False)

        # Delete port bindings for the source host.
        self._confirm_snapshot_based_resize_delete_port_bindings(
            ctxt, instance)

        # Delete volume attachments for the source host.
        self._delete_volume_attachments(ctxt, instance.get_bdms())

        # Free up the old_flavor usage from the resource tracker for this host.
        self.rt.drop_move_claim_at_source(ctxt, instance, migration)

    def _confirm_snapshot_based_resize_delete_port_bindings(
            self, ctxt, instance):
        """Delete port bindings for the source host when confirming
        snapshot-based resize on the source host."

        :param ctxt: nova auth RequestContext
        :param instance: Instance object that was resized/cold migrated
        """
        LOG.debug('Deleting port bindings for source host.',
                  instance=instance)
        try:
            self.network_api.cleanup_instance_network_on_host(
                ctxt, instance, self.host)
        except exception.PortBindingDeletionFailed as e:
            # Do not let this stop us from cleaning up since the guest
            # is already gone.
            LOG.error('Failed to delete port bindings from source host. '
                      'Error: %s', str(e), instance=instance)

    def _delete_volume_attachments(self, ctxt, bdms):
        """Deletes volume attachment records for the given bdms.

        This method will log but not re-raise any exceptions if the volume
        attachment delete fails.

        :param ctxt: nova auth request context used to make
            DELETE /attachments/{attachment_id} requests to cinder.
        :param bdms: objects.BlockDeviceMappingList representing volume
            attachments to delete based on BlockDeviceMapping.attachment_id.
        """
        for bdm in bdms:
            if bdm.attachment_id:
                try:
                    self.volume_api.attachment_delete(ctxt, bdm.attachment_id)
                except Exception as e:
                    LOG.error('Failed to delete volume attachment with ID %s. '
                              'Error: %s', bdm.attachment_id, str(e),
                              instance_uuid=bdm.instance_uuid)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def revert_snapshot_based_resize_at_dest(self, ctxt, instance, migration):
        """Reverts a snapshot-based resize at the destination host.

        Cleans the guest from the destination compute service host hypervisor
        and related resources (ports, volumes) and frees resource usage from
        the compute service on that host.

        :param ctxt: nova auth request context targeted at the target cell
        :param instance: Instance object whose vm_state is "resized" and
            task_state is "resize_reverting".
        :param migration: Migration object whose status is "reverting".
        """
        # A resize revert is essentially a resize back to the old size, so we
        # need to send a usage event here.
        compute_utils.notify_usage_exists(
            self.notifier, ctxt, instance, self.host, current_period=True)

        @utils.synchronized(instance.uuid)
        def do_revert():
            LOG.info('Reverting resize on destination host.',
                     instance=instance)
            with self._error_out_instance_on_exception(ctxt, instance):
                self._revert_snapshot_based_resize_at_dest(
                    ctxt, instance, migration)
        do_revert()

        # Broadcast to all schedulers that the instance is no longer on
        # this host and clear any waiting callback events. This is best effort
        # so if anything fails just log it.
        try:
            self._delete_scheduler_instance_info(ctxt, instance.uuid)
            self.instance_events.clear_events_for_instance(instance)
        except Exception as e:
            LOG.warning('revert_snapshot_based_resize_at_dest failed during '
                        'post-processing. Error: %s', e, instance=instance)

    def _revert_snapshot_based_resize_at_dest(
            self, ctxt, instance, migration):
        """Private version of revert_snapshot_based_resize_at_dest.

        This allows the main method to be decorated with error handlers.

        :param ctxt: nova auth request context targeted at the target cell
        :param instance: Instance object whose vm_state is "resized" and
            task_state is "resize_reverting".
        :param migration: Migration object whose status is "reverting".
        """
        # Cleanup the guest from the hypervisor including local disks.
        network_info = self.network_api.get_instance_nw_info(ctxt, instance)
        bdms = instance.get_bdms()
        block_device_info = self._get_instance_block_device_info(
            ctxt, instance, bdms=bdms)
        LOG.debug('Destroying guest from destination hypervisor including '
                  'disks.', instance=instance)
        self.driver.destroy(
            ctxt, instance, network_info, block_device_info=block_device_info)

        # Activate source host port bindings. We need to do this before
        # deleting the (active) dest host port bindings in
        # setup_networks_on_host otherwise the ports will be unbound and
        # finish on the source will fail.
        # migrate_instance_start uses migration.dest_compute for the port
        # binding host and since we want to activate the source host port
        # bindings, we need to temporarily mutate the migration object.
        with utils.temporary_mutation(
                migration, dest_compute=migration.source_compute):
            LOG.debug('Activating port bindings for source host %s.',
                      migration.source_compute, instance=instance)
            # TODO(mriedem): https://review.opendev.org/#/c/594139/ would allow
            # us to remove this and make setup_networks_on_host do it.
            # TODO(mriedem): Should we try/except/log any errors but continue?
            self.network_api.migrate_instance_start(
                ctxt, instance, migration)

        # Delete port bindings for the target host.
        LOG.debug('Deleting port bindings for target host %s.',
                  self.host, instance=instance)
        try:
            # Note that deleting the destination host port bindings does
            # not automatically activate the source host port bindings.
            self.network_api.cleanup_instance_network_on_host(
                ctxt, instance, self.host)
        except exception.PortBindingDeletionFailed as e:
            # Do not let this stop us from cleaning up since the guest
            # is already gone.
            LOG.error('Failed to delete port bindings from target host. '
                      'Error: %s', str(e), instance=instance)

        # Delete any volume attachments remaining for this target host.
        LOG.debug('Deleting volume attachments for target host.',
                  instance=instance)
        self._delete_volume_attachments(ctxt, bdms)

        # Free up the new_flavor usage from the resource tracker for this host.
        self.rt.drop_move_claim_at_dest(ctxt, instance, migration)

    def _revert_instance_flavor_host_node(self, instance, migration):
        """Revert host, node and flavor fields after a resize-revert."""
        self._set_instance_info(instance, instance.old_flavor)
        instance.host = migration.source_compute
        instance.node = migration.source_node
        instance.save(expected_task_state=[task_states.RESIZE_REVERTING])

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def finish_revert_snapshot_based_resize_at_source(
            self, ctxt, instance, migration):
        """Reverts a snapshot-based resize at the source host.

        Spawn the guest and re-connect volumes/VIFs on the source host and
        revert the instance to use the old_flavor for resource usage reporting.

        Updates allocations in the placement service to move the source node
        allocations, held by the migration record, to the instance and drop
        the allocations held by the instance on the destination node.

        :param ctxt: nova auth request context targeted at the target cell
        :param instance: Instance object whose vm_state is "resized" and
            task_state is "resize_reverting".
        :param migration: Migration object whose status is "reverting".
        """

        @utils.synchronized(instance.uuid)
        def do_revert():
            LOG.info('Reverting resize on source host.', instance=instance)
            with self._error_out_instance_on_exception(ctxt, instance):
                self._finish_revert_snapshot_based_resize_at_source(
                    ctxt, instance, migration)

        try:
            do_revert()
        finally:
            self._delete_stashed_flavor_info(instance)

        # Broadcast to all schedulers that the instance is on this host.
        # This is best effort so if anything fails just log it.
        try:
            self._update_scheduler_instance_info(ctxt, instance)
        except Exception as e:
            LOG.warning('finish_revert_snapshot_based_resize_at_source failed '
                        'during post-processing. Error: %s', e,
                        instance=instance)

    def _finish_revert_snapshot_based_resize_at_source(
            self, ctxt, instance, migration):
        """Private version of finish_revert_snapshot_based_resize_at_source.

        This allows the main method to be decorated with error handlers.

        :param ctxt: nova auth request context targeted at the source cell
        :param instance: Instance object whose vm_state is "resized" and
            task_state is "resize_reverting".
        :param migration: Migration object whose status is "reverting".
        """
        # Get stashed old_vm_state information to determine if guest should
        # be powered on after spawn; we default to ACTIVE for backwards
        # compatibility if old_vm_state is not set
        old_vm_state = instance.system_metadata.get(
            'old_vm_state', vm_states.ACTIVE)

        # Revert the flavor and host/node fields to their previous values
        self._revert_instance_flavor_host_node(instance, migration)

        # Move the allocations against the source compute node resource
        # provider, held by the migration, to the instance which will drop
        # the destination compute node resource provider allocations held by
        # the instance. This puts the allocations against the source node
        # back to the old_flavor and owned by the instance.
        try:
            self._revert_allocation(ctxt, instance, migration)
        except exception.AllocationMoveFailed:
            # Log the error but do not re-raise because we want to continue to
            # process ports and volumes below.
            LOG.error('Reverting allocation in placement for migration '
                      '%(migration_uuid)s failed. You may need to manually '
                      'remove the allocations for the migration consumer '
                      'against the source node resource provider '
                      '%(source_provider)s and the allocations for the '
                      'instance consumer against the destination node '
                      'resource provider %(dest_provider)s and then run the '
                      '"nova-manage placement heal_allocations" command.',
                      {'instance_uuid': instance.uuid,
                       'migration_uuid': migration.uuid,
                       'source_provider': migration.source_node,
                       'dest_provider': migration.dest_node},
                      instance=instance)

        bdms = instance.get_bdms()
        # prep_snapshot_based_resize_at_source created empty volume attachments
        # that we need to update here to get the connection_info before calling
        # driver.finish_revert_migration which will connect the volumes to this
        # host.
        LOG.debug('Updating volume attachments for target host %s.',
                  self.host, instance=instance)
        # TODO(mriedem): We should probably make _update_volume_attachments
        # (optionally) graceful to errors so we (1) try to process all
        # attachments and (2) continue to process networking below.
        self._update_volume_attachments(ctxt, instance, bdms)

        LOG.debug('Updating port bindings for source host %s.',
                  self.host, instance=instance)
        # TODO(mriedem): Calculate provider mappings when we support
        # cross-cell resize/migrate with ports having resource requests.
        # NOTE(hanrong): we need to change migration.dest_compute to
        # source host temporarily.
        # "network_api.migrate_instance_finish" will setup the network
        # for the instance on the destination host. For revert resize,
        # the instance will back to the source host, the setup of the
        # network for instance should be on the source host. So set
        # the migration.dest_compute to source host at here.
        with utils.temporary_mutation(
            migration, dest_compute=migration.source_compute
        ):
            self.network_api.migrate_instance_finish(
                ctxt, instance, migration, provider_mappings=None)
        network_info = self.network_api.get_instance_nw_info(ctxt, instance)

        # Remember that prep_snapshot_based_resize_at_source destroyed the
        # guest but left the disks intact so we cannot call spawn() here but
        # finish_revert_migration should do the job.
        block_device_info = self._get_instance_block_device_info(
            ctxt, instance, bdms=bdms)
        power_on = old_vm_state == vm_states.ACTIVE
        driver_error = None
        try:
            self.driver.finish_revert_migration(
                ctxt, instance, network_info, migration,
                block_device_info=block_device_info, power_on=power_on)
        except Exception as e:
            driver_error = e
            # Leave a hint about hard rebooting the guest and reraise so the
            # instance is put into ERROR state.
            with excutils.save_and_reraise_exception(logger=LOG):
                LOG.error('An error occurred during finish_revert_migration. '
                          'The instance may need to be hard rebooted. Error: '
                          '%s', driver_error, instance=instance)
        else:
            # Perform final cleanup of the instance in the database.
            instance.drop_migration_context()
            # If the original vm_state was STOPPED, set it back to STOPPED.
            vm_state = vm_states.ACTIVE if power_on else vm_states.STOPPED
            self._update_instance_after_spawn(instance, vm_state=vm_state)
            instance.save(expected_task_state=[task_states.RESIZE_REVERTING])
        finally:
            # Complete any volume attachments so the volumes are in-use. We
            # do this regardless of finish_revert_migration failing because
            # the instance is back on this host now and we do not want to leave
            # the volumes in a pending state in case the instance is hard
            # rebooted.
            LOG.debug('Completing volume attachments for instance on source '
                      'host.', instance=instance)
            with excutils.save_and_reraise_exception(
                    reraise=driver_error is not None, logger=LOG):
                self._complete_volume_attachments(ctxt, bdms)

        migration.status = 'reverted'
        migration.save()

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def revert_resize(self, context, instance, migration, request_spec):
        """Destroys the new instance on the destination machine.

        Reverts the model changes, and powers on the old instance on the
        source machine.

        """
        # NOTE(comstud): A revert_resize is essentially a resize back to
        # the old size, so we need to send a usage event here.
        compute_utils.notify_usage_exists(self.notifier, context, instance,
                                          self.host, current_period=True)

        with self._error_out_instance_on_exception(context, instance):
            # NOTE(tr3buchet): tear down networks on destination host
            self.network_api.setup_networks_on_host(context, instance,
                                                    teardown=True)

            self.network_api.migrate_instance_start(context,
                                                    instance,
                                                    migration)

            network_info = self.network_api.get_instance_nw_info(context,
                                                                 instance)
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid)
            block_device_info = self._get_instance_block_device_info(
                                context, instance, bdms=bdms)

            destroy_disks = not self._is_instance_storage_shared(
                context, instance, host=migration.source_compute)
            self.driver.destroy(context, instance, network_info,
                                block_device_info, destroy_disks)

            self._terminate_volume_connections(context, instance, bdms)

            # Free up the new_flavor usage from the resource tracker for this
            # host.
            self.rt.drop_move_claim_at_dest(context, instance, migration)

            # RPC cast back to the source host to finish the revert there.
            self.compute_rpcapi.finish_revert_resize(context, instance,
                    migration, migration.source_compute, request_spec)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def finish_revert_resize(self, context, instance, migration, request_spec):
        """Finishes the second half of reverting a resize on the source host.

        Bring the original source instance state back (active/shutoff) and
        revert the resized attributes in the database.

        """
        try:
            self._finish_revert_resize(
                context, instance, migration, request_spec)
        finally:
            self._delete_stashed_flavor_info(instance)

    def _finish_revert_resize(
        self, context, instance, migration, request_spec=None,
    ):
        """Inner version of finish_revert_resize."""
        with self._error_out_instance_on_exception(context, instance):
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)
            self._notify_about_instance_usage(
                    context, instance, "resize.revert.start")
            compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.RESIZE_REVERT,
                    phase=fields.NotificationPhase.START, bdms=bdms)

            # Get stashed old_vm_state information to determine if guest should
            # be powered on after spawn; we default to ACTIVE for backwards
            # compatibility if old_vm_state is not set
            old_vm_state = instance.system_metadata.get(
                'old_vm_state', vm_states.ACTIVE)

            # Revert the flavor and host/node fields to their previous values
            self._revert_instance_flavor_host_node(instance, migration)

            try:
                source_allocations = self._revert_allocation(
                    context, instance, migration)
            except exception.AllocationMoveFailed:
                LOG.error('Reverting allocation in placement for migration '
                          '%(migration_uuid)s failed. The instance '
                          '%(instance_uuid)s will be put into ERROR state but '
                          'the allocation held by the migration is leaked.',
                          {'instance_uuid': instance.uuid,
                           'migration_uuid': migration.uuid})
                raise

            provider_mappings = self._fill_provider_mapping_based_on_allocs(
                context, source_allocations, request_spec)

            self.network_api.setup_networks_on_host(context, instance,
                                                    migration.source_compute)
            # NOTE(hanrong): we need to change migration.dest_compute to
            # source host temporarily. "network_api.migrate_instance_finish"
            # will setup the network for the instance on the destination host.
            # For revert resize, the instance will back to the source host, the
            # setup of the network for instance should be on the source host.
            # So set the migration.dest_compute to source host at here.
            with utils.temporary_mutation(
                    migration, dest_compute=migration.source_compute):
                self.network_api.migrate_instance_finish(context,
                                                         instance,
                                                         migration,
                                                         provider_mappings)
            network_info = self.network_api.get_instance_nw_info(context,
                                                                 instance)

            # revert_resize deleted any volume attachments for the instance
            # and created new ones to be used on this host, but we
            # have to update those attachments with the host connector so the
            # BDM.connection_info will get set in the call to
            # _get_instance_block_device_info below with refresh_conn_info=True
            # and then the volumes can be re-connected via the driver on this
            # host.
            self._update_volume_attachments(context, instance, bdms)

            block_device_info = self._get_instance_block_device_info(
                    context, instance, refresh_conn_info=True, bdms=bdms)

            power_on = old_vm_state != vm_states.STOPPED
            self.driver.finish_revert_migration(
                context, instance, network_info, migration, block_device_info,
                power_on)

            instance.drop_migration_context()
            instance.launched_at = timeutils.utcnow()
            instance.save(expected_task_state=task_states.RESIZE_REVERTING)

            # Complete any volume attachments so the volumes are in-use.
            self._complete_volume_attachments(context, bdms)

            # if the original vm state was STOPPED, set it back to STOPPED
            LOG.info("Updating instance to original state: '%s'",
                     old_vm_state, instance=instance)
            if power_on:
                instance.vm_state = vm_states.ACTIVE
                instance.task_state = None
                instance.save()
            else:
                instance.task_state = task_states.POWERING_OFF
                instance.save()
                self.stop_instance(context, instance=instance,
                                   clean_shutdown=True)

            self._notify_about_instance_usage(
                    context, instance, "resize.revert.end")
            compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.RESIZE_REVERT,
                    phase=fields.NotificationPhase.END, bdms=bdms)

    def _fill_provider_mapping_based_on_allocs(
            self, context, allocations, request_spec):
        """Fills and returns the request group - resource provider mapping
        based on the allocation passed in.

        :param context: The security context
        :param allocation: allocation dict keyed by RP UUID.
        :param request_spec: The RequestSpec object associated with the
            operation
        :returns: None if the request_spec is None. Otherwise a mapping
            between RequestGroup requester_id, currently Neutron port_id,
            and a list of resource provider UUIDs providing resource for
            that RequestGroup.
        """
        if request_spec:
            # NOTE(gibi): We need to re-calculate the resource provider -
            # port mapping as we have to have the neutron ports allocate
            # from the source compute after revert.
            scheduler_utils.fill_provider_mapping_based_on_allocation(
                context, self.reportclient, request_spec, allocations)
            provider_mappings = self._get_request_group_mapping(
                request_spec)
        else:
            # TODO(sbauza): Remove this conditional once we only support RPC
            # API 6.0
            # NOTE(gibi): The compute RPC is pinned to be older than 5.2
            # and therefore request_spec is not sent. We cannot calculate
            # the provider mappings. If the instance has ports with
            # resource request then the port update will fail in
            # _update_port_binding_for_instance() called via
            # migrate_instance_finish() in finish_revert_resize.
            provider_mappings = None
        return provider_mappings

    def _revert_allocation(self, context, instance, migration):
        """Revert an allocation that is held by migration to our instance."""

        # Fetch the original allocation that the instance had on the source
        # node, which are now held by the migration
        orig_alloc = self.reportclient.get_allocations_for_consumer(
            context, migration.uuid)
        if not orig_alloc:
            LOG.error('Did not find resource allocations for migration '
                      '%s on source node %s. Unable to revert source node '
                      'allocations back to the instance.',
                      migration.uuid, migration.source_node, instance=instance)
            return False

        LOG.info('Swapping old allocation on %(rp_uuids)s held by migration '
                 '%(mig)s for instance',
                 {'rp_uuids': orig_alloc.keys(), 'mig': migration.uuid},
                 instance=instance)
        # FIXME(gibi): This method is flawed in that it does not handle
        # allocations against sharing providers in any special way. This leads
        # to duplicate allocations against the sharing provider during
        # migration.
        # TODO(cdent): Should we be doing anything with return values here?
        self.reportclient.move_allocations(context, migration.uuid,
                                           instance.uuid)
        return orig_alloc

    def _prep_resize(
        self, context, image, instance, flavor, filter_properties, node,
        migration, request_spec, clean_shutdown=True,
    ):

        if not filter_properties:
            filter_properties = {}

        if not instance.host:
            self._set_instance_obj_error_state(instance)
            msg = _('Instance has no source host')
            raise exception.MigrationError(reason=msg)

        same_host = instance.host == self.host
        # if the flavor IDs match, it's migrate; otherwise resize
        if same_host and flavor.id == instance['instance_type_id']:
            # check driver whether support migrate to same host
            if not self.driver.capabilities.get(
                    'supports_migrate_to_same_host', False):
                # Raise InstanceFaultRollback so that the
                # _error_out_instance_on_exception context manager in
                # prep_resize will set the instance.vm_state properly.
                raise exception.InstanceFaultRollback(
                    inner_exception=exception.UnableToMigrateToSelf(
                        instance_id=instance.uuid, host=self.host))

        # NOTE(danms): Stash the new flavor to avoid having to
        # look it up in the database later
        instance.new_flavor = flavor
        # NOTE(mriedem): Stash the old vm_state so we can set the
        # resized/reverted instance back to the same state later.
        vm_state = instance.vm_state
        LOG.debug('Stashing vm_state: %s', vm_state, instance=instance)
        instance.system_metadata['old_vm_state'] = vm_state
        instance.save()

        if not isinstance(request_spec, objects.RequestSpec):
            # Prior to compute RPC API 5.1 conductor would pass a legacy dict
            # version of the request spec to compute and since Stein compute
            # could be sending that back to conductor on reschedule, so if we
            # got a dict convert it to an object.
            # TODO(mriedem): We can drop this compat code when we only support
            # compute RPC API >=6.0.
            request_spec = objects.RequestSpec.from_primitives(
                context, request_spec, filter_properties)
            # We don't have to set the new flavor on the request spec because
            # if we got here it was due to a reschedule from the compute and
            # the request spec would already have the new flavor in it from the
            # else block below.

        provider_mapping = self._get_request_group_mapping(request_spec)

        if provider_mapping:
            try:
                compute_utils.\
                    update_pci_request_spec_with_allocated_interface_name(
                        context, self.reportclient,
                        instance.pci_requests.requests, provider_mapping)
            except (exception.AmbiguousResourceProviderForPCIRequest,
                    exception.UnexpectedResourceProviderNameForPCIRequest
                    ) as e:
                raise exception.BuildAbortException(
                    reason=str(e), instance_uuid=instance.uuid)

        limits = filter_properties.get('limits', {})
        allocs = self.reportclient.get_allocations_for_consumer(
            context, instance.uuid)
        with self.rt.resize_claim(
            context, instance, flavor, node, migration, allocs,
            image_meta=image, limits=limits,
        ) as claim:
            LOG.info('Migrating', instance=instance)
            # RPC cast to the source host to start the actual resize/migration.
            self.compute_rpcapi.resize_instance(
                context, instance, claim.migration, image,
                flavor, request_spec, clean_shutdown)

    def _send_prep_resize_notifications(
            self, context, instance, phase, flavor):
        """Send "resize.prep.*" notifications.

        :param context: nova auth request context
        :param instance: The instance being resized
        :param phase: The phase of the action (NotificationPhase enum)
        :param flavor: The (new) flavor for the resize (same as existing
            instance.flavor for a cold migration)
        """
        # Only send notify_usage_exists if it's the "start" phase.
        if phase == fields.NotificationPhase.START:
            compute_utils.notify_usage_exists(
                self.notifier, context, instance, self.host,
                current_period=True)

        # Send extra usage info about the flavor if it's the "end" phase for
        # the legacy unversioned notification.
        extra_usage_info = None
        if phase == fields.NotificationPhase.END:
            extra_usage_info = dict(
                new_instance_type=flavor.name,
                new_instance_type_id=flavor.id)
        self._notify_about_instance_usage(
            context, instance, "resize.prep.%s" % phase,
            extra_usage_info=extra_usage_info)

        # Send the versioned notification.
        compute_utils.notify_about_resize_prep_instance(
            context, instance, self.host, phase, flavor)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def prep_resize(self, context, image, instance, flavor,
                    request_spec, filter_properties, node,
                    clean_shutdown, migration, host_list):
        """Initiates the process of moving a running instance to another host.

        Possibly changes the VCPU, RAM and disk size in the process.

        This is initiated from conductor and runs on the destination host.

        The main purpose of this method is performing some checks on the
        destination host and making a claim for resources. If the claim fails
        then a reschedule to another host may be attempted which involves
        calling back to conductor to start the process over again.
        """
        if node is None:
            node = self._get_nodename(instance, refresh=True)

        # Pass instance_state=instance.vm_state because we can resize
        # a STOPPED server and we don't want to set it back to ACTIVE
        # in case _prep_resize fails.
        instance_state = instance.vm_state
        with self._error_out_instance_on_exception(
                context, instance, instance_state=instance_state),\
                errors_out_migration_ctxt(migration):

            self._send_prep_resize_notifications(
                context, instance, fields.NotificationPhase.START,
                flavor)
            try:
                scheduler_hints = self._get_scheduler_hints(filter_properties,
                                                            request_spec)
                # Error out if this host cannot accept the new instance due
                # to anti-affinity. At this point the migration is already
                # in-progress, so this is the definitive moment to abort due to
                # the policy violation. Also, exploding here is covered by the
                # cleanup methods in except block.
                try:
                    self._validate_instance_group_policy(context, instance,
                                                         scheduler_hints)
                except exception.RescheduledException as e:
                    raise exception.InstanceFaultRollback(inner_exception=e)

                self._prep_resize(context, image, instance,
                                  flavor, filter_properties,
                                  node, migration, request_spec,
                                  clean_shutdown)
            except exception.BuildAbortException:
                # NOTE(gibi): We failed
                # update_pci_request_spec_with_allocated_interface_name so
                # there is no reason to re-schedule. Just revert the allocation
                # and fail the migration.
                with excutils.save_and_reraise_exception():
                    self._revert_allocation(context, instance, migration)
            except Exception:
                # Since we hit a failure, we're either rescheduling or dead
                # and either way we need to cleanup any allocations created
                # by the scheduler for the destination node.
                self._revert_allocation(context, instance, migration)
                # try to re-schedule the resize elsewhere:
                exc_info = sys.exc_info()
                self._reschedule_resize_or_reraise(context, instance,
                        exc_info, flavor, request_spec,
                        filter_properties, host_list)
            finally:
                self._send_prep_resize_notifications(
                    context, instance, fields.NotificationPhase.END,
                    flavor)

    def _reschedule_resize_or_reraise(self, context, instance, exc_info,
            flavor, request_spec, filter_properties, host_list):
        """Try to re-schedule the resize or re-raise the original error to
        error out the instance.
        """
        if not filter_properties:
            filter_properties = {}

        rescheduled = False
        instance_uuid = instance.uuid

        try:
            retry = filter_properties.get('retry')
            if retry:
                LOG.debug('Rescheduling, attempt %d', retry['num_attempts'],
                          instance_uuid=instance_uuid)

                # reset the task state
                task_state = task_states.RESIZE_PREP
                self._instance_update(context, instance, task_state=task_state)

                if exc_info:
                    # stringify to avoid circular ref problem in json
                    # serialization
                    retry['exc'] = traceback.format_exception_only(
                        exc_info[0], exc_info[1])

                scheduler_hint = {'filter_properties': filter_properties}

                self.compute_task_api.resize_instance(
                    context, instance, scheduler_hint, flavor,
                    request_spec=request_spec, host_list=host_list)

                rescheduled = True
            else:
                # no retry information, do not reschedule.
                LOG.debug('Retry info not present, will not reschedule',
                          instance_uuid=instance_uuid)
                rescheduled = False
        except Exception as error:
            rescheduled = False
            LOG.exception("Error trying to reschedule",
                          instance_uuid=instance_uuid)
            compute_utils.add_instance_fault_from_exc(context,
                    instance, error,
                    exc_info=sys.exc_info())
            self._notify_about_instance_usage(context, instance,
                    'resize.error', fault=error)
            compute_utils.notify_about_instance_action(
                context, instance, self.host,
                action=fields.NotificationAction.RESIZE,
                phase=fields.NotificationPhase.ERROR,
                exception=error,
            )

        if rescheduled:
            self._log_original_error(exc_info, instance_uuid)
            compute_utils.add_instance_fault_from_exc(context,
                    instance, exc_info[1], exc_info=exc_info)
            self._notify_about_instance_usage(context, instance,
                    'resize.error', fault=exc_info[1])
            compute_utils.notify_about_instance_action(
                context, instance, self.host,
                action=fields.NotificationAction.RESIZE,
                phase=fields.NotificationPhase.ERROR,
                exception=exc_info[1],
            )
        else:
            # not re-scheduling
            exc = exc_info[1] or exc_info[0]()
            if exc.__traceback__ is not exc_info[2]:
                raise exc.with_traceback(exc_info[2])
            raise exc

    @messaging.expected_exceptions(exception.MigrationPreCheckError)
    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def prep_snapshot_based_resize_at_dest(
            self, ctxt, instance, flavor, nodename, migration, limits):
        """Performs pre-cross-cell resize resource claim on the dest host.

        This runs on the destination host in a cross-cell resize operation
        before the resize is actually started.

        Performs a resize_claim for resources that are not claimed in placement
        like PCI devices and NUMA topology.

        Note that this is different from same-cell prep_resize in that this:

        * Does not RPC cast to the source compute, that is orchestrated from
          conductor.
        * This does not reschedule on failure, conductor handles that since
          conductor is synchronously RPC calling this method. As such, the
          reverts_task_state decorator is not used on this method.

        :param ctxt: user auth request context
        :param instance: the instance being resized
        :param flavor: the flavor being resized to (unchanged for cold migrate)
        :param nodename: Name of the target compute node
        :param migration: nova.objects.Migration object for the operation
        :param limits: nova.objects.SchedulerLimits object of resource limits
        :returns: nova.objects.MigrationContext; the migration context created
            on the destination host during the resize_claim.
        :raises: nova.exception.MigrationPreCheckError if the pre-check
            validation fails for the given host selection
        """
        LOG.debug('Checking if we can cross-cell migrate instance to this '
                  'host (%s).', self.host, instance=instance)
        self._send_prep_resize_notifications(
            ctxt, instance, fields.NotificationPhase.START, flavor)
        # TODO(mriedem): update_pci_request_spec_with_allocated_interface_name
        # should be called here if the request spec has request group mappings,
        # e.g. for things like QoS ports with resource requests. Do it outside
        # the try/except so if it raises BuildAbortException we do not attempt
        # to reschedule.
        try:
            # Get the allocations within the try/except block in case we get
            # an error so MigrationPreCheckError is raised up.
            allocations = self.reportclient.get_allocs_for_consumer(
                ctxt, instance.uuid)['allocations']
            # Claim resources on this target host using the new flavor which
            # will create the MigrationContext object. Note that in the future
            # if we want to do other validation here we should do it within
            # the MoveClaim context so we can drop the claim if anything fails.
            self.rt.resize_claim(
                ctxt, instance, flavor, nodename, migration, allocations,
                image_meta=instance.image_meta, limits=limits)
        except Exception as ex:
            err = str(ex)
            LOG.warning(
                'Cross-cell resize pre-checks failed for this host (%s). '
                'Cleaning up. Failure: %s', self.host, err,
                instance=instance, exc_info=True)
            raise exception.MigrationPreCheckError(
                reason=(_("Pre-checks failed on host '%(host)s'. "
                          "Error: %(error)s") %
                        {'host': self.host, 'error': err}))
        finally:
            self._send_prep_resize_notifications(
                ctxt, instance, fields.NotificationPhase.END, flavor)

        # ResourceTracker.resize_claim() sets instance.migration_context.
        return instance.migration_context

    @messaging.expected_exceptions(exception.InstancePowerOffFailure)
    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def prep_snapshot_based_resize_at_source(
            self, ctxt, instance, migration, snapshot_id=None):
        """Prepares the instance at the source host for cross-cell resize

        Performs actions like powering off the guest, upload snapshot data if
        the instance is not volume-backed, disconnecting volumes, unplugging
        VIFs and activating the destination host port bindings.

        :param ctxt: user auth request context targeted at source cell
        :param instance: nova.objects.Instance; the instance being resized.
            The expected instance.task_state is "resize_migrating" when calling
            this method, and the expected task_state upon successful completion
            is "resize_migrated".
        :param migration: nova.objects.Migration object for the operation.
            The expected migration.status is "pre-migrating" when calling this
            method and the expected status upon successful completion is
            "post-migrating".
        :param snapshot_id: ID of the image snapshot to upload if not a
            volume-backed instance
        :raises: nova.exception.InstancePowerOffFailure if stopping the
            instance fails
        """
        LOG.info('Preparing for snapshot based resize on source host %s.',
                 self.host, instance=instance)
        # Note that if anything fails here, the migration-based allocations
        # created in conductor should be reverted by conductor as well,
        # see MigrationTask.rollback.
        self._prep_snapshot_based_resize_at_source(
            ctxt, instance, migration, snapshot_id=snapshot_id)

    @delete_image_on_error
    def _snapshot_for_resize(self, ctxt, image_id, instance):
        """Uploads snapshot for the instance during a snapshot-based resize

        If the snapshot operation fails the image will be deleted.

        :param ctxt: the nova auth request context for the resize operation
        :param image_id: the snapshot image ID
        :param instance: the instance to snapshot/resize
        """
        LOG.debug('Uploading snapshot data for image %s', image_id,
                  instance=instance)
        # Note that we do not track the snapshot phase task states
        # during resize since we do not want to reflect those into the
        # actual instance.task_state.
        update_task_state = lambda *args, **kwargs: None
        with timeutils.StopWatch() as timer:
            self.driver.snapshot(ctxt, instance, image_id, update_task_state)
            LOG.debug('Took %0.2f seconds to snapshot the instance on '
                      'the hypervisor.', timer.elapsed(), instance=instance)

    def _prep_snapshot_based_resize_at_source(
            self, ctxt, instance, migration, snapshot_id=None):
        """Private method for prep_snapshot_based_resize_at_source so calling
        code can handle errors and perform rollbacks as necessary.
        """
        # Fetch and update the instance.info_cache.
        network_info = self.network_api.get_instance_nw_info(ctxt, instance)
        # Get the BDMs attached to this instance on this source host.
        bdms = instance.get_bdms()
        # Send the resize.start notification.
        self._send_resize_instance_notifications(
            ctxt, instance, bdms, network_info, fields.NotificationPhase.START)
        # Update the migration status from "pre-migrating" to "migrating".
        migration.status = 'migrating'
        migration.save()

        # Since the instance is going to be left on the source host during the
        # resize, we need to power it off so we do not have the instance
        # potentially running in two places.
        LOG.debug('Stopping instance', instance=instance)
        try:
            self._power_off_instance(instance)
        except Exception as e:
            LOG.exception('Failed to power off instance.', instance=instance)
            raise exception.InstancePowerOffFailure(reason=str(e))
        instance.power_state = self._get_power_state(instance)

        # If a snapshot image ID was provided, we need to snapshot the guest
        # disk image and upload it to the image service.
        if snapshot_id:
            self._snapshot_for_resize(ctxt, snapshot_id, instance)

        block_device_info = self._get_instance_block_device_info(
            ctxt, instance, bdms=bdms)

        # If something fails at this point the instance must go to ERROR
        # status for operator intervention or to reboot/rebuild the instance.
        with self._error_out_instance_on_exception(
                ctxt, instance, instance_state=vm_states.ERROR):

            # Destroy the guest on the source host which will disconnect
            # volumes and unplug VIFs. Note that we DO NOT destroy disks since
            # we want to leave those on the source host in case of a later
            # failure and disks are needed to recover the guest or in case the
            # resize is reverted.
            LOG.debug('Destroying guest on source host but retaining disks.',
                      instance=instance)
            self.driver.destroy(
                ctxt, instance, network_info,
                block_device_info=block_device_info, destroy_disks=False)

            # At this point the volumes are disconnected from this source host.
            # Delete the old volume attachment records and create new empty
            # ones which will be used later if the resize is reverted.
            LOG.debug('Deleting volume attachments for the source host.',
                      instance=instance)
            self._terminate_volume_connections(ctxt, instance, bdms)

            # At this point the VIFs are unplugged from this source host.
            # Activate the dest host port bindings created by conductor.
            self.network_api.migrate_instance_start(ctxt, instance, migration)

            # Update the migration status from "migrating" to "post-migrating".
            migration.status = 'post-migrating'
            migration.save()

            # At this point, the traditional resize_instance would update the
            # instance host/node values to point at the dest host/node because
            # that is where the disk is transferred during resize_instance, but
            # with cross-cell resize the instance is not yet at the dest host
            # so we do not make that update here.
            instance.task_state = task_states.RESIZE_MIGRATED
            instance.save(expected_task_state=task_states.RESIZE_MIGRATING)

        self._send_resize_instance_notifications(
            ctxt, instance, bdms, network_info,
            fields.NotificationPhase.END)
        self.instance_events.clear_events_for_instance(instance)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def resize_instance(self, context, instance, image,
                        migration, flavor, clean_shutdown,
                        request_spec):
        """Starts the migration of a running instance to another host.

        This is initiated from the destination host's ``prep_resize`` routine
        and runs on the source host.
        """
        try:
            self._resize_instance(context, instance, image, migration,
                                  flavor, clean_shutdown, request_spec)
        except Exception:
            with excutils.save_and_reraise_exception():
                self._revert_allocation(context, instance, migration)

    def _resize_instance(
        self, context, instance, image, migration, flavor,
        clean_shutdown, request_spec,
    ):
        # Pass instance_state=instance.vm_state because we can resize
        # a STOPPED server and we don't want to set it back to ACTIVE
        # in case migrate_disk_and_power_off raises InstanceFaultRollback.
        instance_state = instance.vm_state
        with self._error_out_instance_on_exception(
                context, instance, instance_state=instance_state), \
             errors_out_migration_ctxt(migration):
            network_info = self.network_api.get_instance_nw_info(context,
                                                                 instance)

            migration.status = 'migrating'
            migration.save()

            instance.task_state = task_states.RESIZE_MIGRATING
            instance.save(expected_task_state=task_states.RESIZE_PREP)

            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid)
            self._send_resize_instance_notifications(
                context, instance, bdms, network_info,
                fields.NotificationPhase.START)

            block_device_info = self._get_instance_block_device_info(
                                context, instance, bdms=bdms)

            timeout, retry_interval = self._get_power_off_values(
                instance, clean_shutdown)
            disk_info = self.driver.migrate_disk_and_power_off(
                context, instance, migration.dest_host,
                flavor, network_info,
                block_device_info,
                timeout, retry_interval)

            self._terminate_volume_connections(context, instance, bdms)

            self.network_api.migrate_instance_start(context,
                                                    instance,
                                                    migration)

            migration.status = 'post-migrating'
            migration.save()

            instance.host = migration.dest_compute
            instance.node = migration.dest_node
            # NOTE(gibi): as the instance now tracked on the destination we
            # have to make sure that the source compute resource track can
            # track this instance as a migration. For that the resource tracker
            # needs to see the old_flavor set on the instance. The old_flavor
            # setting used to be done on the destination host in finish_resize
            # but that is racy with a source host update_available_resource
            # periodic run
            instance.old_flavor = instance.flavor
            instance.task_state = task_states.RESIZE_MIGRATED
            instance.save(expected_task_state=task_states.RESIZE_MIGRATING)

            # RPC cast to the destination host to finish the resize/migration.
            self.compute_rpcapi.finish_resize(context, instance,
                migration, image, disk_info, migration.dest_compute,
                request_spec)

        self._send_resize_instance_notifications(
            context, instance, bdms, network_info,
            fields.NotificationPhase.END)
        self.instance_events.clear_events_for_instance(instance)

    def _send_resize_instance_notifications(
            self, context, instance, bdms, network_info, phase):
        """Send "resize.(start|end)" notifications.

        :param context: nova auth request context
        :param instance: The instance being resized
        :param bdms: BlockDeviceMappingList for the BDMs associated with the
            instance
        :param network_info: NetworkInfo for the instance info cache of ports
        :param phase: The phase of the action (NotificationPhase enum, either
            ``start`` or ``end``)
        """
        action = fields.NotificationAction.RESIZE
        # Send the legacy unversioned notification.
        self._notify_about_instance_usage(
            context, instance, "%s.%s" % (action, phase),
            network_info=network_info)
        # Send the versioned notification.
        compute_utils.notify_about_instance_action(
            context, instance, self.host, action=action, phase=phase,
            bdms=bdms)

    def _terminate_volume_connections(self, context, instance, bdms):
        connector = None
        for bdm in bdms:
            if bdm.is_volume:
                if bdm.attachment_id:
                    # NOTE(jdg): So here's the thing, the idea behind the new
                    # attach API's was to have a new code fork/path that we
                    # followed, we're not going to do that so we have to do
                    # some extra work in here to make it *behave* just like the
                    # old code. Cinder doesn't allow disconnect/reconnect (you
                    # just delete the attachment and get a new one)
                    # attachments in the new attach code so we have to do
                    # a delete and create without a connector (reserve),
                    # in other words, beware
                    attachment_id = self.volume_api.attachment_create(
                        context, bdm.volume_id, instance.uuid)['id']
                    self.volume_api.attachment_delete(context,
                                                      bdm.attachment_id)
                    bdm.attachment_id = attachment_id
                    bdm.save()

                else:
                    if connector is None:
                        connector = self.driver.get_volume_connector(instance)
                    self.volume_api.terminate_connection(context,
                                                         bdm.volume_id,
                                                         connector)

    @staticmethod
    def _set_instance_info(instance, flavor):
        instance.instance_type_id = flavor.id
        instance.memory_mb = flavor.memory_mb
        instance.vcpus = flavor.vcpus
        instance.root_gb = flavor.root_gb
        instance.ephemeral_gb = flavor.ephemeral_gb
        instance.flavor = flavor

    def _update_volume_attachments(self, context, instance, bdms):
        """Updates volume attachments using the virt driver host connector.

        :param context: nova.context.RequestContext - user request context
        :param instance: nova.objects.Instance
        :param bdms: nova.objects.BlockDeviceMappingList - the list of block
                     device mappings for the given instance
        """
        if bdms:
            connector = None
            for bdm in bdms:
                if bdm.is_volume and bdm.attachment_id:
                    if connector is None:
                        connector = self.driver.get_volume_connector(instance)
                    self.volume_api.attachment_update(
                        context, bdm.attachment_id, connector, bdm.device_name)

    def _complete_volume_attachments(self, context, bdms):
        """Completes volume attachments for the instance

        :param context: nova.context.RequestContext - user request context
        :param bdms: nova.objects.BlockDeviceMappingList - the list of block
                     device mappings for the given instance
        """
        if bdms:
            for bdm in bdms:
                if bdm.is_volume and bdm.attachment_id:
                    self.volume_api.attachment_complete(
                        context, bdm.attachment_id)

    def _finish_resize(self, context, instance, migration, disk_info,
                       image_meta, bdms, request_spec):
        resize_instance = False  # indicates disks have been resized
        old_instance_type_id = migration.old_instance_type_id
        new_instance_type_id = migration.new_instance_type_id
        old_flavor = instance.flavor  # the current flavor is now old
        # NOTE(mriedem): Get the old_vm_state so we know if we should
        # power on the instance. If old_vm_state is not set we need to default
        # to ACTIVE for backwards compatibility
        old_vm_state = instance.system_metadata.get('old_vm_state',
                                                    vm_states.ACTIVE)
        # NOTE(gibi): this is already set by the resize_instance on the source
        # node before calling finish_resize on destination but during upgrade
        # it can be that the source node is not having the fix for bug 1944759
        # yet. This assignment can be removed in Z release.
        instance.old_flavor = old_flavor

        if old_instance_type_id != new_instance_type_id:
            new_flavor = instance.new_flavor  # this is set in _prep_resize
            # Set the flavor-related fields on the instance object including
            # making instance.flavor = new_flavor.
            self._set_instance_info(instance, new_flavor)
            for key in ('root_gb', 'swap', 'ephemeral_gb'):
                if old_flavor[key] != new_flavor[key]:
                    resize_instance = True
                    break
        instance.apply_migration_context()

        # NOTE(tr3buchet): setup networks on destination host
        self.network_api.setup_networks_on_host(context, instance,
                                                migration.dest_compute)
        provider_mappings = self._get_request_group_mapping(request_spec)

        # For neutron, migrate_instance_finish updates port bindings for this
        # host including any PCI devices claimed for SR-IOV ports.
        self.network_api.migrate_instance_finish(
            context, instance, migration, provider_mappings)

        network_info = self.network_api.get_instance_nw_info(context, instance)

        instance.task_state = task_states.RESIZE_FINISH
        instance.save(expected_task_state=task_states.RESIZE_MIGRATED)

        self._send_finish_resize_notifications(
            context, instance, bdms, network_info,
            fields.NotificationPhase.START)

        # We need to update any volume attachments using the destination
        # host connector so that we can update the BDM.connection_info
        # before calling driver.finish_migration otherwise the driver
        # won't know how to connect the volumes to this host.
        # Note that _get_instance_block_device_info with
        # refresh_conn_info=True will update the BDM.connection_info value
        # in the database so we must do this before calling that method.
        self._update_volume_attachments(context, instance, bdms)

        block_device_info = self._get_instance_block_device_info(
            context, instance, refresh_conn_info=True, bdms=bdms)

        # NOTE(mriedem): If the original vm_state was STOPPED, we don't
        # automatically power on the instance after it's migrated
        power_on = old_vm_state != vm_states.STOPPED

        # NOTE(sbauza): During a migration, the original allocation is against
        # the migration UUID while the target allocation (for the destination
        # node) is related to the instance UUID, so here we need to pass the
        # new ones.
        allocations = self.reportclient.get_allocs_for_consumer(
            context, instance.uuid)['allocations']

        try:
            self.driver.finish_migration(context, migration, instance,
                                         disk_info,
                                         network_info,
                                         image_meta, resize_instance,
                                         allocations,
                                         block_device_info, power_on)
        except Exception:
            # Note that we do not rollback port bindings to the source host
            # because resize_instance (on the source host) updated the
            # instance.host to point to *this* host (the destination host)
            # so the port bindings pointing at this host are correct even
            # though we failed to create the guest.
            with excutils.save_and_reraise_exception():
                # If we failed to create the guest on this host, reset the
                # instance flavor-related fields to the old flavor. An
                # error handler like reverts_task_state will save the changes.
                if old_instance_type_id != new_instance_type_id:
                    self._set_instance_info(instance, old_flavor)

        # Now complete any volume attachments that were previously updated.
        self._complete_volume_attachments(context, bdms)

        migration.status = 'finished'
        migration.save()

        instance.vm_state = vm_states.RESIZED
        instance.task_state = None
        instance.launched_at = timeutils.utcnow()
        instance.save(expected_task_state=task_states.RESIZE_FINISH)

        return network_info

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def finish_resize(self, context, disk_info, image, instance,
                      migration, request_spec):
        """Completes the migration process.

        Sets up the newly transferred disk and turns on the instance at its
        new host machine.

        """
        try:
            self._finish_resize_helper(context, disk_info, image, instance,
                                       migration, request_spec)
        except Exception:
            with excutils.save_and_reraise_exception():
                # At this point, resize_instance (which runs on the source) has
                # already updated the instance host/node values to point to
                # this (the dest) compute, so we need to leave the allocations
                # against the dest node resource provider intact and drop the
                # allocations against the source node resource provider. If the
                # user tries to recover the server by hard rebooting it, it
                # will happen on this host so that's where the allocations
                # should go. Note that this is the same method called from
                # confirm_resize to cleanup the source node allocations held
                # by the migration record.
                LOG.info('Deleting allocations for old flavor on source node '
                         '%s after finish_resize failure. You may be able to '
                         'recover the instance by hard rebooting it.',
                         migration.source_compute, instance=instance)
                self._delete_allocation_after_move(
                    context, instance, migration)

    def _finish_resize_helper(self, context, disk_info, image, instance,
                              migration, request_spec):
        """Completes the migration process.

        The caller must revert the instance's allocations if the migration
        process failed.
        """
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            context, instance.uuid)

        with self._error_out_instance_on_exception(context, instance):
            image_meta = objects.ImageMeta.from_dict(image)
            network_info = self._finish_resize(context, instance, migration,
                                               disk_info, image_meta, bdms,
                                               request_spec)

        # TODO(melwitt): We should clean up instance console tokens here. The
        # instance is on a new host and will need to establish a new console
        # connection.
        self._update_scheduler_instance_info(context, instance)
        self._send_finish_resize_notifications(
            context, instance, bdms, network_info,
            fields.NotificationPhase.END)

    def _send_finish_resize_notifications(
            self, context, instance, bdms, network_info, phase):
        """Send notifications for the finish_resize flow.

        :param context: nova auth request context
        :param instance: The instance being resized
        :param bdms: BlockDeviceMappingList for the BDMs associated with the
            instance
        :param network_info: NetworkInfo for the instance info cache of ports
        :param phase: The phase of the action (NotificationPhase enum, either
            ``start`` or ``end``)
        """
        # Send the legacy unversioned notification.
        self._notify_about_instance_usage(
            context, instance, "finish_resize.%s" % phase,
            network_info=network_info)
        # Send the versioned notification.
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.RESIZE_FINISH, phase=phase,
            bdms=bdms)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def finish_snapshot_based_resize_at_dest(
            self, ctxt, instance, migration, snapshot_id):
        """Finishes the snapshot-based resize at the destination compute.

        Sets up block devices and networking on the destination compute and
        spawns the guest.

        :param ctxt: nova auth request context targeted at the target cell DB
        :param instance: The Instance object being resized with the
            ``migration_context`` field set. Upon successful completion of this
            method the vm_state should be "resized", the task_state should be
            None, and migration context, host/node and flavor-related fields
            should be set on the instance.
        :param migration: The Migration object for this resize operation. Upon
            successful completion of this method the migration status should
            be "finished".
        :param snapshot_id: ID of the image snapshot created for a
            non-volume-backed instance, else None.
        """
        LOG.info('Finishing snapshot based resize on destination host %s.',
                 self.host, instance=instance)
        with self._error_out_instance_on_exception(ctxt, instance):
            # Note that if anything fails here, the migration-based allocations
            # created in conductor should be reverted by conductor as well,
            # see MigrationTask.rollback.
            self._finish_snapshot_based_resize_at_dest(
                ctxt, instance, migration, snapshot_id)

    def _finish_snapshot_based_resize_at_dest(
            self, ctxt, instance, migration, snapshot_id):
        """Private variant of finish_snapshot_based_resize_at_dest so the
        caller can handle reverting resource allocations on failure and perform
        other generic error handling.
        """
        # Figure out the image metadata to use when spawning the guest.
        origin_image_ref = instance.image_ref
        if snapshot_id:
            instance.image_ref = snapshot_id
            image_meta = objects.ImageMeta.from_image_ref(
                ctxt, self.image_api, snapshot_id)
        else:
            # Just use what is already on the volume-backed instance.
            image_meta = instance.image_meta

        resize = migration.migration_type == 'resize'
        instance.old_flavor = instance.flavor
        if resize:
            flavor = instance.new_flavor
            # If we are resizing to a new flavor we need to set the
            # flavor-related fields on the instance.
            # NOTE(mriedem): This is likely where storing old/new_flavor on
            # the MigrationContext would make this cleaner.
            self._set_instance_info(instance, flavor)

        instance.apply_migration_context()
        instance.task_state = task_states.RESIZE_FINISH
        instance.save(expected_task_state=task_states.RESIZE_MIGRATED)

        # This seems a bit late to be sending the start notification but
        # it is what traditional resize has always done as well and it does
        # contain the changes to the instance with the new_flavor and
        # task_state.
        bdms = instance.get_bdms()
        network_info = instance.get_network_info()
        self._send_finish_resize_notifications(
            ctxt, instance, bdms, network_info,
            fields.NotificationPhase.START)

        # Setup volumes and networking and spawn the guest in the hypervisor.
        self._finish_snapshot_based_resize_at_dest_spawn(
            ctxt, instance, migration, image_meta, bdms)

        # If we spawned from a temporary snapshot image we can delete that now,
        # similar to how unshelve works.
        if snapshot_id:
            instance.image_ref = origin_image_ref
            compute_utils.delete_image(
                ctxt, instance, self.image_api, snapshot_id)

        migration.status = 'finished'
        migration.save()

        self._update_instance_after_spawn(instance, vm_state=vm_states.RESIZED)
        # Setting the host/node values will make the ResourceTracker continue
        # to track usage for this instance on this host.
        instance.host = migration.dest_compute
        instance.node = migration.dest_node
        instance.save(expected_task_state=task_states.RESIZE_FINISH)

        # Broadcast to all schedulers that the instance is on this host.
        self._update_scheduler_instance_info(ctxt, instance)
        self._send_finish_resize_notifications(
            ctxt, instance, bdms, network_info,
            fields.NotificationPhase.END)

    def _finish_snapshot_based_resize_at_dest_spawn(
            self, ctxt, instance, migration, image_meta, bdms):
        """Sets up volumes and networking and spawns the guest on the dest host

        If the instance was stopped when the resize was initiated the guest
        will be created but remain in a shutdown power state.

        If the spawn fails, port bindings are rolled back to the source host
        and volume connections are terminated for this dest host.

        :param ctxt: nova auth request context
        :param instance: Instance object being migrated
        :param migration: Migration object for the operation
        :param image_meta: ImageMeta object used during driver.spawn
        :param bdms: BlockDeviceMappingList of BDMs for the instance
        """
        # Update the volume attachments using this host's connector.
        # That will update the BlockDeviceMapping.connection_info which
        # will be used to connect the volumes on this host during spawn().
        block_device_info = self._prep_block_device(ctxt, instance, bdms)

        allocations = self.reportclient.get_allocations_for_consumer(
            ctxt, instance.uuid)

        # We do not call self.network_api.setup_networks_on_host here because
        # for neutron that sets up the port migration profile which is only
        # used during live migration with DVR. Yes it is gross knowing what
        # that method does internally. We could change this when bug 1814837
        # is fixed if setup_networks_on_host is made smarter by passing the
        # migration record and the method checks the migration_type.

        # Activate the port bindings for this host.
        # FIXME(mriedem): We're going to have the same issue as bug 1813789
        # here because this will update the port bindings and send the
        # network-vif-plugged event and that means when driver.spawn waits for
        # it we might have already gotten the event and neutron won't send
        # another one so we could timeout.
        # TODO(mriedem): Calculate provider mappings when we support cross-cell
        # resize/migrate with ports having resource requests.
        self.network_api.migrate_instance_finish(
            ctxt, instance, migration, provider_mappings=None)
        network_info = self.network_api.get_instance_nw_info(ctxt, instance)

        # If the original vm_state was STOPPED, we do not automatically
        # power on the instance after it is migrated.
        power_on = instance.system_metadata['old_vm_state'] == vm_states.ACTIVE
        try:
            # NOTE(mriedem): If this instance uses a config drive, it will get
            # rebuilt here which means any personality files will be lost,
            # similar to unshelve. If the instance is not using a config drive
            # and getting metadata from the metadata API service, personality
            # files would be lost regardless of the move operation.
            self.driver.spawn(
                ctxt, instance, image_meta, injected_files=[],
                admin_password=None, allocations=allocations,
                network_info=network_info, block_device_info=block_device_info,
                power_on=power_on)
        except Exception:
            with excutils.save_and_reraise_exception(logger=LOG):
                # Rollback port bindings to the source host.
                try:
                    # This is gross but migrate_instance_start looks at the
                    # migration.dest_compute to determine where to activate the
                    # port bindings and we want the source compute port
                    # bindings to be re-activated. Remember at this point the
                    # instance.host is still pointing at the source compute.
                    # TODO(mriedem): Maybe we should be calling
                    # setup_instance_network_on_host here to deal with pci
                    # devices?
                    with utils.temporary_mutation(
                            migration, dest_compute=migration.source_compute):
                        self.network_api.migrate_instance_start(
                            ctxt, instance, migration)
                except Exception:
                    LOG.exception(
                        'Failed to activate port bindings on the source '
                        'host: %s', migration.source_compute,
                        instance=instance)

                # Rollback volume connections on this host.
                for bdm in bdms:
                    if bdm.is_volume:
                        try:
                            self._remove_volume_connection(
                                ctxt, bdm, instance, delete_attachment=True)
                        except Exception:
                            LOG.exception('Failed to remove volume connection '
                                          'on this host %s for volume %s.',
                                          self.host, bdm.volume_id,
                                          instance=instance)

    @wrap_exception()
    @wrap_instance_fault
    def add_fixed_ip_to_instance(self, context, network_id, instance):
        """Calls network_api to add new fixed_ip to instance
        then injects the new network info and resets instance networking.

        """
        self._notify_about_instance_usage(
                context, instance, "create_ip.start")

        network_info = self.network_api.add_fixed_ip_to_instance(context,
                                                                 instance,
                                                                 network_id)
        self._inject_network_info(instance, network_info)

        # NOTE(russellb) We just want to bump updated_at.  See bug 1143466.
        instance.updated_at = timeutils.utcnow()
        instance.save()

        self._notify_about_instance_usage(
            context, instance, "create_ip.end", network_info=network_info)

    @wrap_exception()
    @wrap_instance_fault
    def remove_fixed_ip_from_instance(self, context, address, instance):
        """Calls network_api to remove existing fixed_ip from instance
        by injecting the altered network info and resetting
        instance networking.
        """
        self._notify_about_instance_usage(
                context, instance, "delete_ip.start")

        network_info = self.network_api.remove_fixed_ip_from_instance(context,
                                                                      instance,
                                                                      address)
        self._inject_network_info(instance, network_info)

        # NOTE(russellb) We just want to bump updated_at.  See bug 1143466.
        instance.updated_at = timeutils.utcnow()
        instance.save()

        self._notify_about_instance_usage(
            context, instance, "delete_ip.end", network_info=network_info)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def pause_instance(self, context, instance):
        """Pause an instance on this host."""
        context = context.elevated()
        LOG.info('Pausing', instance=instance)
        self._notify_about_instance_usage(context, instance, 'pause.start')
        compute_utils.notify_about_instance_action(context, instance,
               self.host, action=fields.NotificationAction.PAUSE,
               phase=fields.NotificationPhase.START)
        self.driver.pause(instance)
        instance.power_state = self._get_power_state(instance)
        instance.vm_state = vm_states.PAUSED
        instance.task_state = None
        instance.save(expected_task_state=task_states.PAUSING)
        self._notify_about_instance_usage(context, instance, 'pause.end')
        compute_utils.notify_about_instance_action(context, instance,
               self.host, action=fields.NotificationAction.PAUSE,
               phase=fields.NotificationPhase.END)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def unpause_instance(self, context, instance):
        """Unpause a paused instance on this host."""
        context = context.elevated()
        LOG.info('Unpausing', instance=instance)
        self._notify_about_instance_usage(context, instance, 'unpause.start')
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.UNPAUSE,
            phase=fields.NotificationPhase.START)
        self.driver.unpause(instance)
        instance.power_state = self._get_power_state(instance)
        instance.vm_state = vm_states.ACTIVE
        instance.task_state = None
        instance.save(expected_task_state=task_states.UNPAUSING)
        self._notify_about_instance_usage(context, instance, 'unpause.end')
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.UNPAUSE,
            phase=fields.NotificationPhase.END)

    @wrap_exception()
    def host_power_action(self, context, action):
        """Reboots, shuts down or powers up the host."""
        return self.driver.host_power_action(action)

    @wrap_exception()
    def host_maintenance_mode(self, context, host, mode):
        """Start/Stop host maintenance window. On start, it triggers
        guest VMs evacuation.
        """
        return self.driver.host_maintenance_mode(host, mode)

    def _update_compute_provider_status(self, context, enabled):
        """Adds or removes the COMPUTE_STATUS_DISABLED trait for this host.

        For each ComputeNode managed by this service, adds or removes the
        COMPUTE_STATUS_DISABLED traits to/from the associated resource provider
        in Placement.

        :param context: nova auth RequestContext
        :param enabled: True if the node is enabled in which case the trait
            would be removed, False if the node is disabled in which case
            the trait would be added.
        :raises: ComputeHostNotFound if there are no compute nodes found in
            the ResourceTracker for this service.
        """
        # Get the compute node(s) on this host. Remember that ironic can be
        # managing more than one compute node.
        nodes = self.rt.compute_nodes.values()
        if not nodes:
            raise exception.ComputeHostNotFound(host=self.host)
        # For each node, we want to add (or remove) the COMPUTE_STATUS_DISABLED
        # trait on the related resource provider in placement so the scheduler
        # (pre-)filters the provider based on its status.
        for node in nodes:
            try:
                self.virtapi.update_compute_provider_status(
                    context, node.uuid, enabled)
            except (exception.ResourceProviderTraitRetrievalFailed,
                    exception.ResourceProviderUpdateConflict,
                    exception.ResourceProviderUpdateFailed,
                    exception.TraitRetrievalFailed) as e:
                # This is best effort so just log a warning and continue.
                LOG.warning('An error occurred while updating '
                            'COMPUTE_STATUS_DISABLED trait on compute node '
                            'resource provider %s. The trait will be '
                            'synchronized when the update_available_resource '
                            'periodic task runs. Error: %s',
                            node.uuid, e.format_message())
            except Exception:
                LOG.exception('An error occurred while updating '
                              'COMPUTE_STATUS_DISABLED trait on compute node '
                              'resource provider %s. The trait will be '
                              'synchronized when the '
                              'update_available_resource periodic task runs.',
                              node.uuid)

    @wrap_exception()
    def set_host_enabled(self, context, enabled):
        """Sets the specified host's ability to accept new instances.

        This method will add or remove the COMPUTE_STATUS_DISABLED trait
        to/from the associated compute node resource provider(s) for this
        compute service.
        """
        try:
            self._update_compute_provider_status(context, enabled)
        except exception.ComputeHostNotFound:
            LOG.warning('Unable to add/remove trait COMPUTE_STATUS_DISABLED. '
                        'No ComputeNode(s) found for host: %s', self.host)

        try:
            return self.driver.set_host_enabled(enabled)
        except NotImplementedError:
            # Only the xenapi driver implements set_host_enabled but we don't
            # want NotImplementedError to get raised back to the API. We still
            # need to honor the compute RPC API contract and return 'enabled'
            # or 'disabled' though.
            return 'enabled' if enabled else 'disabled'

    @wrap_exception()
    def get_host_uptime(self, context):
        """Returns the result of calling "uptime" on the target host."""
        return self.driver.get_host_uptime()

    @wrap_exception()
    @wrap_instance_fault
    def get_diagnostics(self, context, instance):
        """Retrieve diagnostics for an instance on this host."""
        current_power_state = self._get_power_state(instance)
        if current_power_state == power_state.RUNNING:
            LOG.info("Retrieving diagnostics", instance=instance)
            return self.driver.get_diagnostics(instance)
        else:
            raise exception.InstanceInvalidState(
                attr='power state',
                instance_uuid=instance.uuid,
                state=power_state.STATE_MAP[instance.power_state],
                method='get_diagnostics')

    @wrap_exception()
    @wrap_instance_fault
    def get_instance_diagnostics(self, context, instance):
        """Retrieve diagnostics for an instance on this host."""
        current_power_state = self._get_power_state(instance)
        if current_power_state == power_state.RUNNING:
            LOG.info("Retrieving diagnostics", instance=instance)
            return self.driver.get_instance_diagnostics(instance)
        else:
            raise exception.InstanceInvalidState(
                attr='power state',
                instance_uuid=instance.uuid,
                state=power_state.STATE_MAP[instance.power_state],
                method='get_diagnostics')

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def suspend_instance(self, context, instance):
        """Suspend the given instance."""
        context = context.elevated()

        # Store the old state
        instance.system_metadata['old_vm_state'] = instance.vm_state
        self._notify_about_instance_usage(context, instance, 'suspend.start')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.SUSPEND,
                phase=fields.NotificationPhase.START)
        with self._error_out_instance_on_exception(context, instance,
             instance_state=instance.vm_state):
            self.driver.suspend(context, instance)
        instance.power_state = self._get_power_state(instance)
        instance.vm_state = vm_states.SUSPENDED
        instance.task_state = None
        instance.save(expected_task_state=task_states.SUSPENDING)
        self._notify_about_instance_usage(context, instance, 'suspend.end')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.SUSPEND,
                phase=fields.NotificationPhase.END)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def resume_instance(self, context, instance):
        """Resume the given suspended instance."""
        context = context.elevated()
        LOG.info('Resuming', instance=instance)

        self._notify_about_instance_usage(context, instance, 'resume.start')

        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            context, instance.uuid)
        block_device_info = self._get_instance_block_device_info(
            context, instance, bdms=bdms)

        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.RESUME,
            phase=fields.NotificationPhase.START, bdms=bdms)

        network_info = self.network_api.get_instance_nw_info(context, instance)

        with self._error_out_instance_on_exception(context, instance,
             instance_state=instance.vm_state):
            self.driver.resume(context, instance, network_info,
                               block_device_info)

        instance.power_state = self._get_power_state(instance)

        # We default to the ACTIVE state for backwards compatibility
        instance.vm_state = instance.system_metadata.pop('old_vm_state',
                                                         vm_states.ACTIVE)

        instance.task_state = None
        instance.save(expected_task_state=task_states.RESUMING)
        self._notify_about_instance_usage(context, instance, 'resume.end')
        compute_utils.notify_about_instance_action(context, instance,
            self.host, action=fields.NotificationAction.RESUME,
            phase=fields.NotificationPhase.END, bdms=bdms)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def shelve_instance(self, context, instance, image_id,
                        clean_shutdown, accel_uuids):
        """Shelve an instance.

        This should be used when you want to take a snapshot of the instance.
        It also adds system_metadata that can be used by a periodic task to
        offload the shelved instance after a period of time.

        :param context: request context
        :param instance: an Instance object
        :param image_id: an image id to snapshot to.
        :param clean_shutdown: give the GuestOS a chance to stop
        :param accel_uuids: the accelerators uuids for the instance
        """

        @utils.synchronized(instance.uuid)
        def do_shelve_instance():
            self._shelve_instance(context, instance, image_id, clean_shutdown,
                                  accel_uuids)
        do_shelve_instance()

    def _shelve_instance(self, context, instance, image_id,
                         clean_shutdown, accel_uuids=None):
        LOG.info('Shelving', instance=instance)
        offload = CONF.shelved_offload_time == 0
        if offload:
            # Get the BDMs early so we can pass them into versioned
            # notifications since _shelve_offload_instance needs the
            # BDMs anyway.
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)
        else:
            bdms = None
        compute_utils.notify_usage_exists(self.notifier, context, instance,
                                          self.host, current_period=True)
        self._notify_about_instance_usage(context, instance, 'shelve.start')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.SHELVE,
                phase=fields.NotificationPhase.START, bdms=bdms)

        def update_task_state(task_state, expected_state=task_states.SHELVING):
            shelving_state_map = {
                    task_states.IMAGE_PENDING_UPLOAD:
                        task_states.SHELVING_IMAGE_PENDING_UPLOAD,
                    task_states.IMAGE_UPLOADING:
                        task_states.SHELVING_IMAGE_UPLOADING,
                    task_states.SHELVING: task_states.SHELVING}
            task_state = shelving_state_map[task_state]
            expected_state = shelving_state_map[expected_state]
            instance.task_state = task_state
            instance.save(expected_task_state=expected_state)
        # Do not attempt a clean shutdown of a paused guest since some
        # hypervisors will fail the clean shutdown if the guest is not
        # running.
        if instance.power_state == power_state.PAUSED:
            clean_shutdown = False
        self._power_off_instance(instance, clean_shutdown)
        self.driver.snapshot(context, instance, image_id, update_task_state)

        instance.system_metadata['shelved_at'] = timeutils.utcnow().isoformat()
        instance.system_metadata['shelved_image_id'] = image_id
        instance.system_metadata['shelved_host'] = self.host
        instance.vm_state = vm_states.SHELVED
        instance.task_state = None
        if offload:
            instance.task_state = task_states.SHELVING_OFFLOADING
        instance.power_state = self._get_power_state(instance)
        instance.save(expected_task_state=[
                task_states.SHELVING,
                task_states.SHELVING_IMAGE_UPLOADING])

        self._notify_about_instance_usage(context, instance, 'shelve.end')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.SHELVE,
                phase=fields.NotificationPhase.END, bdms=bdms)

        if offload:
            self._shelve_offload_instance(
                context, instance, clean_shutdown=False, bdms=bdms,
                accel_uuids=accel_uuids)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def shelve_offload_instance(self, context, instance, clean_shutdown,
            accel_uuids):
        """Remove a shelved instance from the hypervisor.

        This frees up those resources for use by other instances, but may lead
        to slower unshelve times for this instance.  This method is used by
        volume backed instances since restoring them doesn't involve the
        potentially large download of an image.

        :param context: request context
        :param instance: nova.objects.instance.Instance
        :param clean_shutdown: give the GuestOS a chance to stop
        :param accel_uuids: the accelerators uuids for the instance
        """

        @utils.synchronized(instance.uuid)
        def do_shelve_offload_instance():
            self._shelve_offload_instance(context, instance, clean_shutdown,
                                          accel_uuids=accel_uuids)
        do_shelve_offload_instance()

    def _shelve_offload_instance(self, context, instance, clean_shutdown,
                                 bdms=None, accel_uuids=None):
        LOG.info('Shelve offloading', instance=instance)
        if bdms is None:
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)
        self._notify_about_instance_usage(context, instance,
                'shelve_offload.start')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.SHELVE_OFFLOAD,
                phase=fields.NotificationPhase.START, bdms=bdms)

        self._power_off_instance(instance, clean_shutdown)
        current_power_state = self._get_power_state(instance)
        network_info = self.network_api.get_instance_nw_info(context, instance)

        block_device_info = self._get_instance_block_device_info(context,
                                                                 instance,
                                                                 bdms=bdms)
        self.driver.destroy(context, instance, network_info,
                block_device_info)

        # the instance is going to be removed from the host so we want to
        # terminate all the connections with the volume server and the host
        self._terminate_volume_connections(context, instance, bdms)

        # NOTE(brinzhang): Free up the accelerator resource occupied
        # in the cyborg service.
        if accel_uuids:
            cyclient = cyborg.get_client(context)
            cyclient.delete_arqs_for_instance(instance.uuid)

        # Free up the resource allocations in the placement service.
        # This should happen *before* the vm_state is changed to
        # SHELVED_OFFLOADED in case client-side code is polling the API to
        # schedule more instances (or unshelve) once this server is offloaded.
        self.rt.delete_allocation_for_shelve_offloaded_instance(context,
                                                                instance)

        instance.power_state = current_power_state
        # NOTE(mriedem): The vm_state has to be set before updating the
        # resource tracker, see vm_states.ALLOW_RESOURCE_REMOVAL. The host/node
        # values cannot be nulled out until after updating the resource tracker
        # though.
        instance.vm_state = vm_states.SHELVED_OFFLOADED
        instance.task_state = None
        instance.save(expected_task_state=[task_states.SHELVING,
                                           task_states.SHELVING_OFFLOADING])

        # NOTE(ndipanov): Free resources from the resource tracker
        self._update_resource_tracker(context, instance)

        # NOTE(sfinucan): RPC calls should no longer be attempted against this
        # instance, so ensure any calls result in errors
        self._nil_out_instance_obj_host_and_node(instance)
        instance.save(expected_task_state=None)

        # TODO(melwitt): We should clean up instance console tokens here. The
        # instance has no host at this point and will need to establish a new
        # console connection in the future after it is unshelved.
        self._delete_scheduler_instance_info(context, instance.uuid)
        self._notify_about_instance_usage(context, instance,
                'shelve_offload.end')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.SHELVE_OFFLOAD,
                phase=fields.NotificationPhase.END, bdms=bdms)

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def unshelve_instance(self, context, instance, image,
                          filter_properties, node, request_spec, accel_uuids):
        """Unshelve the instance.

        :param context: request context
        :param instance: a nova.objects.instance.Instance object
        :param image: an image to build from.  If None we assume a
            volume backed instance.
        :param filter_properties: dict containing limits, retry info etc.
        :param node: target compute node
        :param request_spec: the RequestSpec object used to schedule the
            instance
        :param accel_uuids: the accelerators uuids for the instance
        """
        if filter_properties is None:
            filter_properties = {}

        @utils.synchronized(instance.uuid)
        def do_unshelve_instance():
            self._unshelve_instance(
                context, instance, image, filter_properties, node,
                request_spec, accel_uuids)
        do_unshelve_instance()

    def _unshelve_instance_key_scrub(self, instance):
        """Remove data from the instance that may cause side effects."""
        cleaned_keys = dict(
                key_data=instance.key_data,
                auto_disk_config=instance.auto_disk_config)
        instance.key_data = None
        instance.auto_disk_config = False
        return cleaned_keys

    def _unshelve_instance_key_restore(self, instance, keys):
        """Restore previously scrubbed keys before saving the instance."""
        instance.update(keys)

    def _unshelve_instance(self, context, instance, image, filter_properties,
                           node, request_spec, accel_uuids):
        LOG.info('Unshelving', instance=instance)
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)

        self._notify_about_instance_usage(context, instance, 'unshelve.start')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.UNSHELVE,
                phase=fields.NotificationPhase.START, bdms=bdms)

        instance.task_state = task_states.SPAWNING
        instance.save()

        block_device_info = self._prep_block_device(context, instance, bdms)
        scrubbed_keys = self._unshelve_instance_key_scrub(instance)

        if node is None:
            node = self._get_nodename(instance)

        limits = filter_properties.get('limits', {})

        allocations = self.reportclient.get_allocations_for_consumer(
            context, instance.uuid)

        shelved_image_ref = instance.image_ref
        if image:
            instance.image_ref = image['id']
            image_meta = objects.ImageMeta.from_dict(image)
        else:
            image_meta = objects.ImageMeta.from_dict(
                utils.get_image_from_system_metadata(
                    instance.system_metadata))

        provider_mappings = self._get_request_group_mapping(request_spec)

        try:
            if provider_mappings:
                update = (
                    compute_utils.
                        update_pci_request_spec_with_allocated_interface_name)
                update(
                    context, self.reportclient, instance.pci_requests.requests,
                    provider_mappings)

            accel_info = []
            if accel_uuids:
                try:
                    accel_info = self._get_bound_arq_resources(
                        context, instance, accel_uuids)
                except (Exception, eventlet.timeout.Timeout) as exc:
                    LOG.exception('Failure getting accelerator requests '
                                  'with the exception: %s', exc,
                                  instance=instance)
                    self._build_resources_cleanup(instance, None)
                    raise

            with self.rt.instance_claim(context, instance, node, allocations,
                                        limits):
                self.network_api.setup_instance_network_on_host(
                    context, instance, self.host,
                    provider_mappings=provider_mappings)
                network_info = self.network_api.get_instance_nw_info(
                    context, instance)

                self.driver.spawn(context, instance, image_meta,
                                  injected_files=[],
                                  admin_password=None,
                                  allocations=allocations,
                                  network_info=network_info,
                                  block_device_info=block_device_info,
                                  accel_info=accel_info)
        except Exception:
            with excutils.save_and_reraise_exception(logger=LOG):
                LOG.exception('Instance failed to spawn',
                              instance=instance)
                # Set the image_ref back to initial image_ref because instance
                # object might have been saved with image['id']
                # https://bugs.launchpad.net/nova/+bug/1934094
                instance.image_ref = shelved_image_ref
                # Cleanup allocations created by the scheduler on this host
                # since we failed to spawn the instance. We do this both if
                # the instance claim failed with ComputeResourcesUnavailable
                # or if we did claim but the spawn failed, because aborting the
                # instance claim will not remove the allocations.
                self.reportclient.delete_allocation_for_instance(
                    context, instance.uuid, force=True)
                # FIXME: Umm, shouldn't we be rolling back port bindings too?
                self._terminate_volume_connections(context, instance, bdms)
                # The reverts_task_state decorator on unshelve_instance will
                # eventually save these updates.
                self._nil_out_instance_obj_host_and_node(instance)

        if image:
            instance.image_ref = shelved_image_ref
            self._delete_snapshot_of_shelved_instance(context, instance,
                                                      image['id'])

        self._unshelve_instance_key_restore(instance, scrubbed_keys)
        self._update_instance_after_spawn(instance)
        # Delete system_metadata for a shelved instance
        compute_utils.remove_shelved_keys_from_system_metadata(instance)

        instance.save(expected_task_state=task_states.SPAWNING)
        self._update_scheduler_instance_info(context, instance)
        self._notify_about_instance_usage(context, instance, 'unshelve.end')
        compute_utils.notify_about_instance_action(context, instance,
                self.host, action=fields.NotificationAction.UNSHELVE,
                phase=fields.NotificationPhase.END, bdms=bdms)

    def _inject_network_info(self, instance, network_info):
        """Inject network info for the given instance."""
        LOG.debug('Inject network info', instance=instance)
        LOG.debug('network_info to inject: |%s|', network_info,
                  instance=instance)

        self.driver.inject_network_info(instance, network_info)

    @wrap_instance_fault
    def inject_network_info(self, context, instance):
        """Inject network info, but don't return the info."""
        network_info = self.network_api.get_instance_nw_info(context, instance)
        self._inject_network_info(instance, network_info)

    @messaging.expected_exceptions(NotImplementedError,
                                   exception.ConsoleNotAvailable,
                                   exception.InstanceNotFound)
    @wrap_exception()
    @wrap_instance_fault
    def get_console_output(self, context, instance, tail_length):
        """Send the console output for the given instance."""
        context = context.elevated()
        LOG.info("Get console output", instance=instance)
        output = self.driver.get_console_output(context, instance)

        if type(output) is str:
            output = output.encode("latin-1")

        if tail_length is not None:
            output = self._tail_log(output, tail_length)

        return output.decode('ascii', 'replace')

    def _tail_log(self, log, length):
        try:
            length = int(length)
        except ValueError:
            length = 0

        if length == 0:
            return b''
        else:
            return b'\n'.join(log.split(b'\n')[-int(length):])

    @messaging.expected_exceptions(exception.ConsoleTypeInvalid,
                                   exception.InstanceNotReady,
                                   exception.InstanceNotFound,
                                   exception.ConsoleTypeUnavailable,
                                   NotImplementedError)
    @wrap_exception()
    @wrap_instance_fault
    def get_vnc_console(self, context, console_type, instance):
        """Return connection information for a vnc console."""
        context = context.elevated()
        LOG.debug("Getting vnc console", instance=instance)

        if not CONF.vnc.enabled:
            raise exception.ConsoleTypeUnavailable(console_type=console_type)

        if console_type == 'novnc':
            # For essex, novncproxy_base_url must include the full path
            # including the html file (like http://myhost/vnc_auto.html)
            access_url_base = CONF.vnc.novncproxy_base_url
        else:
            raise exception.ConsoleTypeInvalid(console_type=console_type)

        try:
            # Retrieve connect info from driver, and then decorate with our
            # access info token
            console = self.driver.get_vnc_console(context, instance)
            console_auth = objects.ConsoleAuthToken(
                context=context,
                console_type=console_type,
                host=console.host,
                port=console.port,
                internal_access_path=console.internal_access_path,
                instance_uuid=instance.uuid,
                access_url_base=access_url_base,
            )
            console_auth.authorize(CONF.consoleauth.token_ttl)
            connect_info = console.get_connection_info(
                console_auth.token, console_auth.access_url)

        except exception.InstanceNotFound:
            if instance.vm_state != vm_states.BUILDING:
                raise
            raise exception.InstanceNotReady(instance_id=instance.uuid)

        return connect_info

    @messaging.expected_exceptions(exception.ConsoleTypeInvalid,
                                   exception.InstanceNotReady,
                                   exception.InstanceNotFound,
                                   exception.ConsoleTypeUnavailable,
                                   NotImplementedError)
    @wrap_exception()
    @wrap_instance_fault
    def get_spice_console(self, context, console_type, instance):
        """Return connection information for a spice console."""
        context = context.elevated()
        LOG.debug("Getting spice console", instance=instance)

        if not CONF.spice.enabled:
            raise exception.ConsoleTypeUnavailable(console_type=console_type)

        if console_type != 'spice-html5':
            raise exception.ConsoleTypeInvalid(console_type=console_type)

        try:
            # Retrieve connect info from driver, and then decorate with our
            # access info token
            console = self.driver.get_spice_console(context, instance)
            console_auth = objects.ConsoleAuthToken(
                context=context,
                console_type=console_type,
                host=console.host,
                port=console.port,
                internal_access_path=console.internal_access_path,
                instance_uuid=instance.uuid,
                access_url_base=CONF.spice.html5proxy_base_url,
            )
            console_auth.authorize(CONF.consoleauth.token_ttl)
            connect_info = console.get_connection_info(
                console_auth.token, console_auth.access_url)

        except exception.InstanceNotFound:
            if instance.vm_state != vm_states.BUILDING:
                raise
            raise exception.InstanceNotReady(instance_id=instance.uuid)

        return connect_info

    @messaging.expected_exceptions(exception.ConsoleTypeInvalid,
                                   exception.InstanceNotReady,
                                   exception.InstanceNotFound,
                                   exception.ConsoleTypeUnavailable,
                                   NotImplementedError)
    @wrap_exception()
    @wrap_instance_fault
    def get_rdp_console(self, context, console_type, instance):
        """Return connection information for a RDP console."""
        context = context.elevated()
        LOG.debug("Getting RDP console", instance=instance)

        if not CONF.rdp.enabled:
            raise exception.ConsoleTypeUnavailable(console_type=console_type)

        if console_type != 'rdp-html5':
            raise exception.ConsoleTypeInvalid(console_type=console_type)

        try:
            # Retrieve connect info from driver, and then decorate with our
            # access info token
            console = self.driver.get_rdp_console(context, instance)
            console_auth = objects.ConsoleAuthToken(
                context=context,
                console_type=console_type,
                host=console.host,
                port=console.port,
                internal_access_path=console.internal_access_path,
                instance_uuid=instance.uuid,
                access_url_base=CONF.rdp.html5_proxy_base_url,
            )
            console_auth.authorize(CONF.consoleauth.token_ttl)
            connect_info = console.get_connection_info(
                console_auth.token, console_auth.access_url)

        except exception.InstanceNotFound:
            if instance.vm_state != vm_states.BUILDING:
                raise
            raise exception.InstanceNotReady(instance_id=instance.uuid)

        return connect_info

    @messaging.expected_exceptions(exception.ConsoleTypeInvalid,
                                   exception.InstanceNotReady,
                                   exception.InstanceNotFound,
                                   exception.ConsoleTypeUnavailable,
                                   NotImplementedError)
    @wrap_exception()
    @wrap_instance_fault
    def get_mks_console(self, context, console_type, instance):
        """Return connection information for a MKS console."""
        context = context.elevated()
        LOG.debug("Getting MKS console", instance=instance)

        if not CONF.mks.enabled:
            raise exception.ConsoleTypeUnavailable(console_type=console_type)

        if console_type != 'webmks':
            raise exception.ConsoleTypeInvalid(console_type=console_type)

        try:
            # Retrieve connect info from driver, and then decorate with our
            # access info token
            console = self.driver.get_mks_console(context, instance)
            console_auth = objects.ConsoleAuthToken(
                context=context,
                console_type=console_type,
                host=console.host,
                port=console.port,
                internal_access_path=console.internal_access_path,
                instance_uuid=instance.uuid,
                access_url_base=CONF.mks.mksproxy_base_url,
            )
            console_auth.authorize(CONF.consoleauth.token_ttl)
            connect_info = console.get_connection_info(
                console_auth.token, console_auth.access_url)

        except exception.InstanceNotFound:
            if instance.vm_state != vm_states.BUILDING:
                raise
            raise exception.InstanceNotReady(instance_id=instance.uuid)

        return connect_info

    @messaging.expected_exceptions(
        exception.ConsoleTypeInvalid,
        exception.InstanceNotReady,
        exception.InstanceNotFound,
        exception.ConsoleTypeUnavailable,
        exception.SocketPortRangeExhaustedException,
        exception.ImageSerialPortNumberInvalid,
        exception.ImageSerialPortNumberExceedFlavorValue,
        NotImplementedError)
    @wrap_exception()
    @wrap_instance_fault
    def get_serial_console(self, context, console_type, instance):
        """Returns connection information for a serial console."""

        LOG.debug("Getting serial console", instance=instance)

        if not CONF.serial_console.enabled:
            raise exception.ConsoleTypeUnavailable(console_type=console_type)

        context = context.elevated()

        try:
            # Retrieve connect info from driver, and then decorate with our
            # access info token
            console = self.driver.get_serial_console(context, instance)
            console_auth = objects.ConsoleAuthToken(
                context=context,
                console_type=console_type,
                host=console.host,
                port=console.port,
                internal_access_path=console.internal_access_path,
                instance_uuid=instance.uuid,
                access_url_base=CONF.serial_console.base_url,
            )
            console_auth.authorize(CONF.consoleauth.token_ttl)
            connect_info = console.get_connection_info(
                console_auth.token, console_auth.access_url)

        except exception.InstanceNotFound:
            if instance.vm_state != vm_states.BUILDING:
                raise
            raise exception.InstanceNotReady(instance_id=instance.uuid)

        return connect_info

    @messaging.expected_exceptions(exception.ConsoleTypeInvalid,
                                   exception.InstanceNotReady,
                                   exception.InstanceNotFound)
    @wrap_exception()
    @wrap_instance_fault
    def validate_console_port(self, ctxt, instance, port, console_type):
        if console_type == "spice-html5":
            console_info = self.driver.get_spice_console(ctxt, instance)
        elif console_type == "rdp-html5":
            console_info = self.driver.get_rdp_console(ctxt, instance)
        elif console_type == "serial":
            console_info = self.driver.get_serial_console(ctxt, instance)
        elif console_type == "webmks":
            console_info = self.driver.get_mks_console(ctxt, instance)
        else:
            console_info = self.driver.get_vnc_console(ctxt, instance)

        # Some drivers may return an int on console_info.port but the port
        # variable in this method is a string, so cast to be sure we are
        # comparing the correct types.
        return str(console_info.port) == port

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_fault
    def reserve_block_device_name(self, context, instance, device,
                                  volume_id, disk_bus, device_type, tag,
                                  multiattach):
        if (tag and not
                self.driver.capabilities.get('supports_tagged_attach_volume',
                                             False)):
            raise exception.VolumeTaggedAttachNotSupported()

        if (multiattach and not
                self.driver.capabilities.get('supports_multiattach', False)):
            raise exception.MultiattachNotSupportedByVirtDriver(
                volume_id=volume_id)

        @utils.synchronized(instance.uuid)
        def do_reserve():
            bdms = (
                objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid))

            # Now that we have the lock check that we haven't raced another
            # request and ensure there is no existing attachment
            if any(b for b in bdms if b.volume_id == volume_id):
                msg = _("volume %s already attached") % volume_id
                raise exception.InvalidVolume(reason=msg)

            # NOTE(ndipanov): We need to explicitly set all the fields on the
            #                 object so that obj_load_attr does not fail
            new_bdm = objects.BlockDeviceMapping(
                    context=context,
                    source_type='volume', destination_type='volume',
                    instance_uuid=instance.uuid, boot_index=None,
                    volume_id=volume_id,
                    device_name=device, guest_format=None,
                    disk_bus=disk_bus, device_type=device_type, tag=tag)

            new_bdm.device_name = self._get_device_name_for_instance(
                    instance, bdms, new_bdm)

            # NOTE(vish): create bdm here to avoid race condition
            new_bdm.create()
            return new_bdm

        return do_reserve()

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def attach_volume(self, context, instance, bdm):
        """Attach a volume to an instance."""
        driver_bdm = driver_block_device.convert_volume(bdm)

        @utils.synchronized(instance.uuid)
        def do_attach_volume(context, instance, driver_bdm):
            try:
                return self._attach_volume(context, instance, driver_bdm)
            except Exception:
                with excutils.save_and_reraise_exception():
                    bdm.destroy()

        do_attach_volume(context, instance, driver_bdm)

    def _attach_volume(self, context, instance, bdm):
        context = context.elevated()
        LOG.info('Attaching volume %(volume_id)s to %(mountpoint)s',
                 {'volume_id': bdm.volume_id,
                  'mountpoint': bdm['mount_device']},
                 instance=instance)
        compute_utils.notify_about_volume_attach_detach(
            context, instance, self.host,
            action=fields.NotificationAction.VOLUME_ATTACH,
            phase=fields.NotificationPhase.START,
            volume_id=bdm.volume_id)
        try:
            bdm.attach(context, instance, self.volume_api, self.driver,
                       do_driver_attach=True)
        except Exception as e:
            with excutils.save_and_reraise_exception():
                LOG.exception("Failed to attach %(volume_id)s "
                              "at %(mountpoint)s",
                              {'volume_id': bdm.volume_id,
                               'mountpoint': bdm['mount_device']},
                              instance=instance)
                if bdm['attachment_id']:
                    # Try to delete the attachment to make the volume
                    # available again. Note that DriverVolumeBlockDevice
                    # may have already deleted the attachment so ignore
                    # VolumeAttachmentNotFound.
                    try:
                        self.volume_api.attachment_delete(
                            context, bdm['attachment_id'])
                    except exception.VolumeAttachmentNotFound as exc:
                        LOG.debug('Ignoring VolumeAttachmentNotFound: %s',
                                  exc, instance=instance)
                else:
                    self.volume_api.unreserve_volume(context, bdm.volume_id)
                compute_utils.notify_about_volume_attach_detach(
                    context, instance, self.host,
                    action=fields.NotificationAction.VOLUME_ATTACH,
                    phase=fields.NotificationPhase.ERROR,
                    exception=e,
                    volume_id=bdm.volume_id)

        info = {'volume_id': bdm.volume_id}
        self._notify_about_instance_usage(
            context, instance, "volume.attach", extra_usage_info=info)
        compute_utils.notify_about_volume_attach_detach(
            context, instance, self.host,
            action=fields.NotificationAction.VOLUME_ATTACH,
            phase=fields.NotificationPhase.END,
            volume_id=bdm.volume_id)

    def _notify_volume_usage_detach(self, context, instance, bdm):
        if CONF.volume_usage_poll_interval <= 0:
            return

        mp = bdm.device_name
        # Handle bootable volumes which will not contain /dev/
        if '/dev/' in mp:
            mp = mp[5:]
        try:
            vol_stats = self.driver.block_stats(instance, mp)
            if vol_stats is None:
                return
        except NotImplementedError:
            return

        LOG.debug("Updating volume usage cache with totals", instance=instance)
        rd_req, rd_bytes, wr_req, wr_bytes, flush_ops = vol_stats
        vol_usage = objects.VolumeUsage(context)
        vol_usage.volume_id = bdm.volume_id
        vol_usage.instance_uuid = instance.uuid
        vol_usage.project_id = instance.project_id
        vol_usage.user_id = instance.user_id
        vol_usage.availability_zone = instance.availability_zone
        vol_usage.curr_reads = rd_req
        vol_usage.curr_read_bytes = rd_bytes
        vol_usage.curr_writes = wr_req
        vol_usage.curr_write_bytes = wr_bytes
        vol_usage.save(update_totals=True)
        self.notifier.info(context, 'volume.usage', vol_usage.to_dict())
        compute_utils.notify_about_volume_usage(context, vol_usage, self.host)

    def _detach_volume(self, context, bdm, instance, destroy_bdm=True,
                       attachment_id=None):
        """Detach a volume from an instance.

        :param context: security context
        :param bdm: nova.objects.BlockDeviceMapping volume bdm to detach
        :param instance: the Instance object to detach the volume from
        :param destroy_bdm: if True, the corresponding BDM entry will be marked
                            as deleted. Disabling this is useful for operations
                            like rebuild, when we don't want to destroy BDM
        :param attachment_id: The volume attachment_id for the given instance
                              and volume.
        """
        volume_id = bdm.volume_id
        compute_utils.notify_about_volume_attach_detach(
            context, instance, self.host,
            action=fields.NotificationAction.VOLUME_DETACH,
            phase=fields.NotificationPhase.START,
            volume_id=volume_id)

        self._notify_volume_usage_detach(context, instance, bdm)

        LOG.info('Detaching volume %(volume_id)s',
                 {'volume_id': volume_id}, instance=instance)

        driver_bdm = driver_block_device.convert_volume(bdm)
        driver_bdm.detach(context, instance, self.volume_api, self.driver,
                          attachment_id=attachment_id, destroy_bdm=destroy_bdm)

        info = dict(volume_id=volume_id)
        self._notify_about_instance_usage(
            context, instance, "volume.detach", extra_usage_info=info)
        compute_utils.notify_about_volume_attach_detach(
            context, instance, self.host,
            action=fields.NotificationAction.VOLUME_DETACH,
            phase=fields.NotificationPhase.END,
            volume_id=volume_id)

        if 'tag' in bdm and bdm.tag:
            self._delete_disk_metadata(instance, bdm)
        if destroy_bdm:
            bdm.destroy()

    def _delete_disk_metadata(self, instance, bdm):
        for device in instance.device_metadata.devices:
            if isinstance(device, objects.DiskMetadata):
                if 'serial' in device:
                    if device.serial == bdm.volume_id:
                        instance.device_metadata.devices.remove(device)
                        instance.save()
                        break
                else:
                    # NOTE(artom) We log the entire device object because all
                    # fields are nullable and may not be set
                    LOG.warning('Unable to determine whether to clean up '
                                'device metadata for disk %s', device,
                                instance=instance)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def detach_volume(self, context, volume_id, instance, attachment_id):
        """Detach a volume from an instance.

        :param context: security context
        :param volume_id: the volume id
        :param instance: the Instance object to detach the volume from
        :param attachment_id: The volume attachment_id for the given instance
                              and volume.

        """
        @utils.synchronized(instance.uuid)
        def do_detach_volume(context, volume_id, instance, attachment_id):
            bdm = objects.BlockDeviceMapping.get_by_volume_and_instance(
                    context, volume_id, instance.uuid)
            self._detach_volume(context, bdm, instance,
                                attachment_id=attachment_id)

        do_detach_volume(context, volume_id, instance, attachment_id)

    def _init_volume_connection(self, context, new_volume,
                                old_volume_id, connector, bdm,
                                new_attachment_id, mountpoint):
        new_volume_id = new_volume['id']
        if new_attachment_id is None:
            # We're dealing with an old-style attachment so initialize the
            # connection so we can get the connection_info.
            new_cinfo = self.volume_api.initialize_connection(context,
                                                              new_volume_id,
                                                              connector)
        else:
            # Check for multiattach on the new volume and if True, check to
            # see if the virt driver supports multiattach.
            # TODO(mriedem): This is copied from DriverVolumeBlockDevice
            # and should be consolidated into some common code at some point.
            vol_multiattach = new_volume.get('multiattach', False)
            virt_multiattach = self.driver.capabilities.get(
                'supports_multiattach', False)
            if vol_multiattach and not virt_multiattach:
                raise exception.MultiattachNotSupportedByVirtDriver(
                    volume_id=new_volume_id)

            # This is a new style attachment and the API created the new
            # volume attachment and passed the id to the compute over RPC.
            # At this point we need to update the new volume attachment with
            # the host connector, which will give us back the new attachment
            # connection_info.
            new_cinfo = self.volume_api.attachment_update(
                context, new_attachment_id, connector,
                mountpoint)['connection_info']

            if vol_multiattach:
                # This will be used by the volume driver to determine the
                # proper disk configuration.
                new_cinfo['multiattach'] = True

        old_cinfo = jsonutils.loads(bdm['connection_info'])
        if old_cinfo and 'serial' not in old_cinfo:
            old_cinfo['serial'] = old_volume_id
        # NOTE(lyarwood): serial is not always present in the returned
        # connection_info so set it if it is missing as we do in
        # DriverVolumeBlockDevice.attach().
        if 'serial' not in new_cinfo:
            new_cinfo['serial'] = new_volume_id
        return (old_cinfo, new_cinfo)

    def _swap_volume(self, context, instance, bdm, connector,
                     old_volume_id, new_volume, resize_to,
                     new_attachment_id, is_cinder_migration):
        new_volume_id = new_volume['id']
        mountpoint = bdm['device_name']
        failed = False
        new_cinfo = None
        try:
            old_cinfo, new_cinfo = self._init_volume_connection(
                context, new_volume, old_volume_id, connector,
                bdm, new_attachment_id, mountpoint)
            # NOTE(lyarwood): The Libvirt driver, the only virt driver
            # currently implementing swap_volume, will modify the contents of
            # new_cinfo when connect_volume is called. This is then saved to
            # the BDM in swap_volume for future use outside of this flow.
            msg = ("swap_volume: Calling driver volume swap with "
                   "connection infos: new: %(new_cinfo)s; "
                   "old: %(old_cinfo)s" %
                   {'new_cinfo': new_cinfo, 'old_cinfo': old_cinfo})
            # Both new and old info might contain password
            LOG.debug(strutils.mask_password(msg), instance=instance)

            self.driver.swap_volume(context, old_cinfo, new_cinfo, instance,
                                    mountpoint, resize_to)
            if new_attachment_id:
                self.volume_api.attachment_complete(context, new_attachment_id)
            msg = ("swap_volume: Driver volume swap returned, new "
                   "connection_info is now : %(new_cinfo)s" %
                   {'new_cinfo': new_cinfo})
            LOG.debug(strutils.mask_password(msg))
        except Exception as ex:
            failed = True
            with excutils.save_and_reraise_exception():
                compute_utils.notify_about_volume_swap(
                    context, instance, self.host,
                    fields.NotificationPhase.ERROR,
                    old_volume_id, new_volume_id, ex)
                if new_cinfo:
                    msg = ("Failed to swap volume %(old_volume_id)s "
                           "for %(new_volume_id)s")
                    LOG.exception(msg, {'old_volume_id': old_volume_id,
                                        'new_volume_id': new_volume_id},
                                  instance=instance)
                else:
                    msg = ("Failed to connect to volume %(volume_id)s "
                           "with volume at %(mountpoint)s")
                    LOG.exception(msg, {'volume_id': new_volume_id,
                                        'mountpoint': bdm['device_name']},
                                  instance=instance)

                # The API marked the volume as 'detaching' for the old volume
                # so we need to roll that back so the volume goes back to
                # 'in-use' state.
                self.volume_api.roll_detaching(context, old_volume_id)

                if new_attachment_id is None:
                    # The API reserved the new volume so it would be in
                    # 'attaching' status, so we need to unreserve it so it
                    # goes back to 'available' status.
                    self.volume_api.unreserve_volume(context, new_volume_id)
                else:
                    # This is a new style attachment for the new volume, which
                    # was created in the API. We just need to delete it here
                    # to put the new volume back into 'available' status.
                    self.volume_api.attachment_delete(
                        context, new_attachment_id)
        finally:
            # TODO(mriedem): This finally block is terribly confusing and is
            # trying to do too much. We should consider removing the finally
            # block and move whatever needs to happen on success and failure
            # into the blocks above for clarity, even if it means a bit of
            # redundant code.
            conn_volume = new_volume_id if failed else old_volume_id
            if new_cinfo:
                LOG.debug("swap_volume: removing Cinder connection "
                          "for volume %(volume)s", {'volume': conn_volume},
                          instance=instance)
                if bdm.attachment_id is None:
                    # This is the pre-3.44 flow for new-style volume
                    # attachments so just terminate the connection.
                    self.volume_api.terminate_connection(context,
                                                         conn_volume,
                                                         connector)
                else:
                    # This is a new style volume attachment. If we failed, then
                    # the new attachment was already deleted above in the
                    # exception block and we have nothing more to do here. If
                    # swap_volume was successful in the driver, then we need to
                    # "detach" the original attachment by deleting it.
                    if not failed:
                        self.volume_api.attachment_delete(
                            context, bdm.attachment_id)

            # Need to make some decisions based on whether this was
            # a Cinder initiated migration or not. The callback to
            # migration completion isn't needed in the case of a
            # nova initiated simple swap of two volume
            # "volume-update" call so skip that. The new attachment
            # scenarios will give us a new attachment record and
            # that's what we want.
            if bdm.attachment_id and not is_cinder_migration:
                # we don't callback to cinder
                comp_ret = {'save_volume_id': new_volume_id}
            else:
                # NOTE(lyarwood): The following call to
                # os-migrate-volume-completion returns a dict containing
                # save_volume_id, this volume id has two possible values :
                # 1. old_volume_id if we are migrating (retyping) volumes
                # 2. new_volume_id if we are swapping between two existing
                #    volumes
                # This volume id is later used to update the volume_id and
                # connection_info['serial'] of the BDM.
                comp_ret = self.volume_api.migrate_volume_completion(
                                                          context,
                                                          old_volume_id,
                                                          new_volume_id,
                                                          error=failed)
                LOG.debug("swap_volume: Cinder migrate_volume_completion "
                          "returned: %(comp_ret)s", {'comp_ret': comp_ret},
                          instance=instance)

        return (comp_ret, new_cinfo)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def swap_volume(self, context, old_volume_id, new_volume_id, instance,
                    new_attachment_id):
        """Replace the old volume with the new volume within the active server

        :param context: User request context
        :param old_volume_id: Original volume id
        :param new_volume_id: New volume id being swapped to
        :param instance: Instance with original_volume_id attached
        :param new_attachment_id: ID of the new attachment for new_volume_id
        """
        @utils.synchronized(instance.uuid)
        def _do_locked_swap_volume(context, old_volume_id, new_volume_id,
                                   instance, new_attachment_id):
            self._do_swap_volume(context, old_volume_id, new_volume_id,
                                 instance, new_attachment_id)
        _do_locked_swap_volume(context, old_volume_id, new_volume_id, instance,
                               new_attachment_id)

    def _do_swap_volume(self, context, old_volume_id, new_volume_id,
                        instance, new_attachment_id):
        """Replace the old volume with the new volume within the active server

        :param context: User request context
        :param old_volume_id: Original volume id
        :param new_volume_id: New volume id being swapped to
        :param instance: Instance with original_volume_id attached
        :param new_attachment_id: ID of the new attachment for new_volume_id
        """
        context = context.elevated()
        compute_utils.notify_about_volume_swap(
            context, instance, self.host,
            fields.NotificationPhase.START,
            old_volume_id, new_volume_id)

        bdm = objects.BlockDeviceMapping.get_by_volume_and_instance(
                context, old_volume_id, instance.uuid)
        connector = self.driver.get_volume_connector(instance)

        resize_to = 0
        old_volume = self.volume_api.get(context, old_volume_id)
        # Yes this is a tightly-coupled state check of what's going on inside
        # cinder, but we need this while we still support old (v1/v2) and
        # new style attachments (v3.44). Once we drop support for old style
        # attachments we could think about cleaning up the cinder-initiated
        # swap volume API flows.
        is_cinder_migration = False
        if 'migration_status' in old_volume:
            is_cinder_migration = old_volume['migration_status'] == 'migrating'
        old_vol_size = old_volume['size']
        new_volume = self.volume_api.get(context, new_volume_id)
        new_vol_size = new_volume['size']
        if new_vol_size > old_vol_size:
            resize_to = new_vol_size

        LOG.info('Swapping volume %(old_volume)s for %(new_volume)s',
                 {'old_volume': old_volume_id, 'new_volume': new_volume_id},
                 instance=instance)
        comp_ret, new_cinfo = self._swap_volume(context,
                                                instance,
                                                bdm,
                                                connector,
                                                old_volume_id,
                                                new_volume,
                                                resize_to,
                                                new_attachment_id,
                                                is_cinder_migration)

        # NOTE(lyarwood): Update the BDM with the modified new_cinfo and
        # correct volume_id returned by Cinder.
        save_volume_id = comp_ret['save_volume_id']

        # NOTE(lyarwood): Overwrite the possibly stale serial and volume_id in
        # the connection_info with the volume_id returned from Cinder. This
        # could be the case during a volume migration where the new_cinfo here
        # refers to the temporary volume *before* Cinder renames it to the
        # original volume UUID at the end of the migration.
        new_cinfo['serial'] = save_volume_id
        new_cinfo['volume_id'] = save_volume_id
        if 'data' in new_cinfo:
            new_cinfo['data']['volume_id'] = save_volume_id
        values = {
            'connection_info': jsonutils.dumps(new_cinfo),
            'source_type': 'volume',
            'destination_type': 'volume',
            'snapshot_id': None,
            'volume_id': save_volume_id,
            'no_device': None}

        if resize_to:
            values['volume_size'] = resize_to

        if new_attachment_id is not None:
            # This was a volume swap for a new-style attachment so we
            # need to update the BDM attachment_id for the new attachment.
            values['attachment_id'] = new_attachment_id

        LOG.debug("swap_volume: Updating volume %(volume_id)s BDM record with "
                  "%(updates)s", {'volume_id': bdm.volume_id,
                                  'updates': values},
                  instance=instance)
        bdm.update(values)
        bdm.save()

        compute_utils.notify_about_volume_swap(
            context, instance, self.host,
            fields.NotificationPhase.END,
            old_volume_id, new_volume_id)

    @wrap_exception()
    def remove_volume_connection(self, context, volume_id, instance):
        """Remove the volume connection on this host

        Detach the volume from this instance on this host, and if this is
        the cinder v2 flow, call cinder to terminate the connection.
        """
        try:
            # NOTE(mriedem): If the BDM was just passed directly we would not
            # need to do this DB query, but this is an RPC interface so
            # changing that requires some care.
            bdm = objects.BlockDeviceMapping.get_by_volume_and_instance(
                    context, volume_id, instance.uuid)
            # NOTE(mriedem): Normally we would pass delete_attachment=True to
            # _remove_volume_connection to delete a v3 style volume attachment,
            # but this method is RPC called from _rollback_live_migration which
            # already deletes the attachment, so because of that tight coupling
            # we cannot simply delete a v3 style attachment here without
            # needing to do some behavior modification of that
            # _rollback_live_migration flow which gets messy.
            self._remove_volume_connection(context, bdm, instance)
        except exception.NotFound:
            pass

    def _remove_volume_connection(self, context, bdm, instance,
                                  delete_attachment=False):
        """Remove the volume connection on this host

        Detach the volume from this instance on this host.

        :param context: nova auth request context
        :param bdm: BlockDeviceMapping object for a volume attached to the
            instance
        :param instance: Instance object with a volume attached represented
            by ``bdm``
        :param delete_attachment: If ``bdm.attachment_id`` is not None the
            attachment was made as a cinder v3 style attachment and if True,
            then deletes the volume attachment, otherwise just terminates
            the connection for a cinder legacy style connection.
        """
        driver_bdm = driver_block_device.convert_volume(bdm)
        driver_bdm.driver_detach(context, instance,
                                 self.volume_api, self.driver)
        if bdm.attachment_id is None:
            # cinder v2 api flow
            connector = self.driver.get_volume_connector(instance)
            self.volume_api.terminate_connection(context, bdm.volume_id,
                                                 connector)
        elif delete_attachment:
            # cinder v3 api flow
            self.volume_api.attachment_delete(context, bdm.attachment_id)

    def _deallocate_port_resource_for_instance(
        self,
        context: nova.context.RequestContext,
        instance: 'objects.Instance',
        port_id: str,
        port_allocation: ty.Dict[str, ty.Dict[str, ty.Dict[str, int]]],
    ) -> None:

        if not port_allocation:
            return

        try:
            client = self.reportclient
            client.remove_resources_from_instance_allocation(
                context, instance.uuid, port_allocation)
        except Exception as ex:
            # We always raise here as it is not a race condition where
            # somebody has already deleted the port we want to cleanup.
            # Here we see that the port exists, the allocation exists,
            # but we cannot clean it up so we will actually leak
            # allocations.
            with excutils.save_and_reraise_exception():
                LOG.warning(
                    'Failed to remove resource allocation of port %(port_id)s '
                    'for instance. Error: %(error)s',
                    {'port_id': port_id, 'error': ex},
                    instance=instance)

    def _deallocate_port_for_instance(
            self, context, instance, port_id, raise_on_failure=False,
            pci_device=None):
        try:
            result = self.network_api.deallocate_port_for_instance(
                context, instance, port_id)
            __, port_allocation = result
        except Exception as ex:
            with excutils.save_and_reraise_exception(
                    reraise=raise_on_failure):
                LOG.warning('Failed to deallocate port %(port_id)s '
                            'for instance. Error: %(error)s',
                            {'port_id': port_id, 'error': ex},
                            instance=instance)
        else:
            if pci_device:
                self.rt.unclaim_pci_devices(context, pci_device, instance)
                instance.remove_pci_device_and_request(pci_device)

            # Deallocate the resources in placement that were used by the
            # detached port.
            self._deallocate_port_resource_for_instance(
                context, instance, port_id, port_allocation)

    def _claim_pci_device_for_interface_attach(
        self,
        context: nova.context.RequestContext,
        instance: 'objects.Instance',
        pci_reqs: 'objects.InstancePCIRequests',
    ) -> ty.Optional['objects.PciDevice']:
        """Claim PCI devices if there are PCI requests

        :param context: nova.context.RequestContext
        :param instance: the objects.Instance to where the interface is being
            attached
        :param pci_reqs: A InstancePCIRequests object describing the
            needed PCI devices
        :raises InterfaceAttachPciClaimFailed: if the PCI device claim fails
        :returns: An objects.PciDevice describing the claimed PCI device for
            the interface or None if no device is requested
        """

        if not pci_reqs.requests:
            return None

        try:
            devices = self.rt.claim_pci_devices(
                context, pci_reqs, instance.numa_topology)
        except exception.PciDeviceRequestFailed:
            LOG.info('Failed to claim PCI devices during interface attach '
                     'for PCI request %s', pci_reqs, instance=instance)
            raise exception.InterfaceAttachPciClaimFailed(
                instance_uuid=instance.uuid)

        # NOTE(gibi): We assume that maximum one PCI devices is attached per
        # interface attach request.
        device = devices[0]
        instance.pci_devices.objects.append(device)

        return device

    def _allocate_port_resource_for_instance(
        self,
        context: nova.context.RequestContext,
        instance: 'objects.Instance',
        pci_reqs: 'objects.InstancePCIRequests',
        request_groups: ty.List['objects.RequestGroup'],
        request_level_params: 'objects.RequestLevelParams',
    ) -> ty.Tuple[ty.Optional[ty.Dict[str, ty.List[str]]],
                  ty.Optional[ty.Dict[str, ty.Dict[str, ty.Dict[str, int]]]]]:
        """Allocate resources for the request in placement

        :param context: nova.context.RequestContext
        :param instance: the objects.Instance to where the interface is being
            attached
        :param pci_reqs: A list of InstancePCIRequest objects describing the
            needed PCI devices
        :param request_groups: A list of RequestGroup objects describing the
            resources the port requests from placement
        :param request_level_params: A RequestLevelParams object describing the
            non group specific request of the port.
        :raises InterfaceAttachResourceAllocationFailed: if we failed to
            allocate resource in placement for the request
        :returns: A tuple of provider mappings and allocated resources or
            (None, None) if no resource allocation was needed for the request
        """

        if not request_groups:
            return None, None

        # restrict the resource request to the current compute node. The
        # compute node uuid is the uuid of the root provider of the node in
        # placement
        compute_node_uuid = objects.ComputeNode.get_by_nodename(
            context, instance.node).uuid
        # we can have multiple request groups, it would be enough to restrict
        # only one of them to the compute tree but for symetry we restrict
        # all of them
        for request_group in request_groups:
            request_group.in_tree = compute_node_uuid

        # NOTE(gibi): group policy is mandatory in a resource request if there
        # are multiple groups. The policy can only come from the flavor today
        # and a new flavor is not provided with an interface attach request and
        # the instance's current flavor might not have a policy. Still we are
        # attaching a single port where currently the two possible groups
        # (one for bandwidth and one for packet rate) will always be allocated
        # from different providers. So both possible policies (none, isolated)
        # are always fulfilled for this single port. We still has to specify
        # one so we specify the least restrictive now.
        rr = scheduler_utils.ResourceRequest.from_request_groups(
            request_groups, request_level_params, group_policy='none')
        res = self.reportclient.get_allocation_candidates(context, rr)
        alloc_reqs, provider_sums, version = res

        if not alloc_reqs:
            # no allocation candidates available, we run out of free resources
            raise exception.InterfaceAttachResourceAllocationFailed(
                instance_uuid=instance.uuid)

        # select one of the candidates and update the instance
        # allocation
        # TODO(gibi): We could loop over all possible candidates
        # if the first one selected here does not work due to race or due
        # to not having free PCI devices. However the latter is only
        # detected later in the interface attach code path.
        alloc_req = alloc_reqs[0]
        resources = alloc_req['allocations']
        provider_mappings = alloc_req['mappings']
        try:
            self.reportclient.add_resources_to_instance_allocation(
                context, instance.uuid, resources)
        except exception.AllocationUpdateFailed as e:
            # We lost a race. We could retry another candidate
            raise exception.InterfaceAttachResourceAllocationFailed(
                instance_uuid=instance.uuid) from e
        except (
            exception.ConsumerAllocationRetrievalFailed,
            keystone_exception.ClientException,
        ) as e:
            # These are non-recoverable errors so we should not retry
            raise exception.InterfaceAttachResourceAllocationFailed(
                instance_uuid=instance.uuid) from e

        try:
            update = (
                compute_utils.
                    update_pci_request_spec_with_allocated_interface_name)
            update(
                context, self.reportclient, pci_reqs.requests,
                provider_mappings)
        except (
            exception.AmbiguousResourceProviderForPCIRequest,
            exception.UnexpectedResourceProviderNameForPCIRequest
        ):
            # These are programing errors. So we clean up an re-raise to let
            # the request fail
            with excutils.save_and_reraise_exception():
                self.reportclient.remove_resources_from_instance_allocation(
                    context, instance.uuid, resources)

        return provider_mappings, resources

    # TODO(mriedem): There are likely race failures which can result in
    # NotFound and QuotaError exceptions getting traced as well.
    @messaging.expected_exceptions(
        # Do not log a traceback for user errors. We use Invalid generically
        # since this method can raise lots of different exceptions:
        # AttachInterfaceNotSupported
        # NetworkInterfaceTaggedAttachNotSupported
        # NetworkAmbiguous
        # PortNotUsable
        # PortInUse
        # PortNotUsableDNS
        # AttachSRIOVPortNotSupported
        # NetworksWithQoSPolicyNotSupported
        # InterfaceAttachResourceAllocationFailed
        exception.Invalid)
    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def attach_interface(self, context, instance, network_id, port_id,
                         requested_ip, tag):
        """Use hotplug to add an network adapter to an instance."""
        lockname = 'interface-%s-%s' % (instance.uuid, port_id)

        @utils.synchronized(lockname)
        def do_attach_interface(context, instance, network_id, port_id,
                                requested_ip, tag):
            return self._attach_interface(context, instance, network_id,
                                port_id, requested_ip, tag)

        return do_attach_interface(context, instance, network_id, port_id,
                                   requested_ip, tag)

    def _attach_interface(self, context, instance, network_id, port_id,
                          requested_ip, tag):
        if not self.driver.capabilities.get('supports_attach_interface',
                                            False):
            raise exception.AttachInterfaceNotSupported(
                instance_uuid=instance.uuid)
        if (tag and not
            self.driver.capabilities.get('supports_tagged_attach_interface',
                                         False)):
            raise exception.NetworkInterfaceTaggedAttachNotSupported()

        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.INTERFACE_ATTACH,
            phase=fields.NotificationPhase.START)

        bind_host_id = self.driver.network_binding_host_id(context, instance)

        requested_networks = objects.NetworkRequestList(
            objects=[
                objects.NetworkRequest(
                    network_id=network_id,
                    port_id=port_id,
                    address=requested_ip,
                    tag=tag,
                )
            ]
        )

        if len(requested_networks) != 1:
            LOG.warning(
                "Interface attach only supports one interface per attach "
                "request", instance=instance)
            raise exception.InterfaceAttachFailed(instance_uuid=instance.uuid)

        pci_numa_affinity_policy = hardware.get_pci_numa_policy_constraint(
            instance.flavor, instance.image_meta)
        pci_reqs = objects.InstancePCIRequests(
            requests=[], instance_uuid=instance.uuid)
        _, request_groups, req_lvl_params = (
            self.network_api.create_resource_requests(
                context,
                requested_networks,
                pci_reqs,
                affinity_policy=pci_numa_affinity_policy
            )
        )

        result = self._allocate_port_resource_for_instance(
            context, instance, pci_reqs, request_groups, req_lvl_params)
        provider_mappings, resources = result

        try:
            pci_device = self._claim_pci_device_for_interface_attach(
                context, instance, pci_reqs)
        except exception.InterfaceAttachPciClaimFailed:
            with excutils.save_and_reraise_exception():
                if resources:
                    # TODO(gibi): Instead of giving up we could try another
                    # allocation candidate from _allocate_resources() if any
                    self._deallocate_port_resource_for_instance(
                        context, instance, port_id, resources)

        instance.pci_requests.requests.extend(pci_reqs.requests)

        network_info = self.network_api.allocate_for_instance(
            context,
            instance,
            requested_networks,
            bind_host_id=bind_host_id,
            resource_provider_mapping=provider_mappings,
        )

        if len(network_info) != 1:
            LOG.error('allocate_for_instance returned %(ports)s '
                      'ports', {'ports': len(network_info)})
            # TODO(elod.illes): an instance.interface_attach.error notification
            # should be sent here
            raise exception.InterfaceAttachFailed(
                    instance_uuid=instance.uuid)
        image_meta = objects.ImageMeta.from_instance(instance)

        try:
            self.driver.attach_interface(context, instance, image_meta,
                                         network_info[0])
        except exception.NovaException as ex:
            port_id = network_info[0].get('id')
            LOG.warning("attach interface failed , try to deallocate "
                        "port %(port_id)s, reason: %(msg)s",
                        {'port_id': port_id, 'msg': ex},
                        instance=instance)
            self._deallocate_port_for_instance(
                context, instance, port_id, pci_device=pci_device)

            compute_utils.notify_about_instance_action(
                context, instance, self.host,
                action=fields.NotificationAction.INTERFACE_ATTACH,
                phase=fields.NotificationPhase.ERROR,
                exception=ex)

            raise exception.InterfaceAttachFailed(
                instance_uuid=instance.uuid)

        if pci_device:
            # NOTE(gibi): The _claim_pci_device_for_interface_attach() call
            # found a pci device but it only marked the device as claimed. The
            # periodic update_available_resource would move the device to
            # allocated state. But as driver.attach_interface() has been
            # succeeded we now know that the interface is also allocated
            # (used by) to the instance. So make sure the pci tracker also
            # tracks this device as allocated. This way we can avoid a possible
            # race condition when a detach arrives for a device that is only
            # in claimed state.
            self.rt.allocate_pci_devices_for_instance(context, instance)

        instance.save()

        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.INTERFACE_ATTACH,
            phase=fields.NotificationPhase.END)

        return network_info[0]

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def detach_interface(self, context, instance, port_id):
        """Detach a network adapter from an instance."""
        lockname = 'interface-%s-%s' % (instance.uuid, port_id)

        @utils.synchronized(lockname)
        def do_detach_interface(context, instance, port_id):
            self._detach_interface(context, instance, port_id)

        do_detach_interface(context, instance, port_id)

    def _detach_interface(self, context, instance, port_id):
        # NOTE(aarents): we need to refresh info cache from DB here,
        # as previous detach/attach lock holder just updated it.
        compute_utils.refresh_info_cache_for_instance(context, instance)
        network_info = instance.info_cache.network_info
        condemned = None
        for vif in network_info:
            if vif['id'] == port_id:
                condemned = vif
                break
        if condemned is None:
            raise exception.PortNotFound(_("Port %s is not "
                                           "attached") % port_id)

        pci_req = pci_req_module.get_instance_pci_request_from_vif(
            context, instance, condemned)

        pci_device = None
        if pci_req:
            pci_devices = [pci_device
                           for pci_device in instance.pci_devices.objects
                           if pci_device.request_id == pci_req.request_id]

            if not pci_devices:
                LOG.warning(
                    "Detach interface failed, port_id=%(port_id)s, "
                    "reason: PCI device not found for PCI request %(pci_req)s",
                    {'port_id': port_id, 'pci_req': pci_req})
                raise exception.InterfaceDetachFailed(
                    instance_uuid=instance.uuid)

            pci_device = pci_devices[0]

        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.INTERFACE_DETACH,
            phase=fields.NotificationPhase.START)

        try:
            self.driver.detach_interface(context, instance, condemned)
        except exception.NovaException as ex:
            # If the instance was deleted before the interface was detached,
            # just log it at debug.
            log_level = (logging.DEBUG
                         if isinstance(ex, exception.InstanceNotFound)
                         else logging.WARNING)
            LOG.log(log_level,
                    "Detach interface failed, port_id=%(port_id)s, reason: "
                    "%(msg)s", {'port_id': port_id, 'msg': ex},
                    instance=instance)
            raise exception.InterfaceDetachFailed(instance_uuid=instance.uuid)
        else:
            self._deallocate_port_for_instance(
                context, instance, port_id, raise_on_failure=True,
                pci_device=pci_device)

        instance.save()

        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.INTERFACE_DETACH,
            phase=fields.NotificationPhase.END)

    def _get_compute_info(self, context, host):
        return objects.ComputeNode.get_first_node_by_host_for_old_compat(
            context, host)

    @wrap_exception()
    def check_instance_shared_storage(self, ctxt, data):
        """Check if the instance files are shared

        :param ctxt: security context
        :param data: result of driver.check_instance_shared_storage_local

        Returns True if instance disks located on shared storage and
        False otherwise.
        """
        return self.driver.check_instance_shared_storage_remote(ctxt, data)

    def _dest_can_numa_live_migrate(self, dest_check_data, migration):
        # TODO(artom) If we have a libvirt driver we expect it to set
        # dst_supports_numa_live_migration, but we have to remove it if we
        # did not get a migration from the conductor, indicating that it
        # cannot send RPC 5.3. This check can be removed in RPC 6.0.
        if ('dst_supports_numa_live_migration' in dest_check_data and
                dest_check_data.dst_supports_numa_live_migration and
                not migration):
            delattr(dest_check_data, 'dst_supports_numa_live_migration')
        return dest_check_data

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def check_can_live_migrate_destination(self, ctxt, instance,
                                           block_migration, disk_over_commit,
                                           migration, limits):
        """Check if it is possible to execute live migration.

        This runs checks on the destination host, and then calls
        back to the source host to check the results.

        :param context: security context
        :param instance: dict of instance data
        :param block_migration: if true, prepare for block migration
                                if None, calculate it in driver
        :param disk_over_commit: if true, allow disk over commit
                                 if None, ignore disk usage checking
        :param migration: objects.Migration object for this live migration.
        :param limits: objects.SchedulerLimits object for this live migration.
        :returns: a LiveMigrateData object (hypervisor-dependent)
        """

        # Error out if this host cannot accept the new instance due
        # to anti-affinity. This check at this moment is not very accurate, as
        # multiple requests may be happening concurrently and miss the lock,
        # but when it works it provides a better user experience by failing
        # earlier. Also, it should be safe to explode here, error becomes
        # NoValidHost and instance status remains ACTIVE.
        try:
            self._validate_instance_group_policy(ctxt, instance)
        except exception.RescheduledException as e:
            msg = ("Failed to validate instance group policy "
                   "due to: {}".format(e))
            raise exception.MigrationPreCheckError(reason=msg)

        src_compute_info = obj_base.obj_to_primitive(
            self._get_compute_info(ctxt, instance.host))
        dst_compute_info = obj_base.obj_to_primitive(
            self._get_compute_info(ctxt, self.host))
        dest_check_data = self.driver.check_can_live_migrate_destination(ctxt,
            instance, src_compute_info, dst_compute_info,
            block_migration, disk_over_commit)
        dest_check_data = self._dest_can_numa_live_migrate(dest_check_data,
                                                           migration)
        LOG.debug('destination check data is %s', dest_check_data)
        try:
            allocs = self.reportclient.get_allocations_for_consumer(
                ctxt, instance.uuid)
            migrate_data = self.compute_rpcapi.check_can_live_migrate_source(
                ctxt, instance, dest_check_data)
            if ('src_supports_numa_live_migration' in migrate_data and
                    migrate_data.src_supports_numa_live_migration):
                migrate_data = self._live_migration_claim(
                    ctxt, instance, migrate_data, migration, limits, allocs)
            elif 'dst_supports_numa_live_migration' in dest_check_data:
                LOG.info('Destination was ready for NUMA live migration, '
                         'but source is either too old, or is set to an '
                         'older upgrade level.', instance=instance)
            if self.network_api.has_port_binding_extension(ctxt):
                # Create migrate_data vifs if not provided by driver.
                if 'vifs' not in migrate_data:
                    migrate_data.vifs = (
                        migrate_data_obj.
                        VIFMigrateData.create_skeleton_migrate_vifs(
                            instance.get_network_info()))
                # Claim PCI devices for VIFs on destination (if needed)
                port_id_to_pci = self._claim_pci_for_instance_vifs(
                    ctxt, instance)
                # Update migrate VIFs with the newly claimed PCI devices
                self._update_migrate_vifs_profile_with_pci(
                    migrate_data.vifs, port_id_to_pci)
        finally:
            self.driver.cleanup_live_migration_destination_check(ctxt,
                    dest_check_data)
        return migrate_data

    def _live_migration_claim(self, ctxt, instance, migrate_data,
                              migration, limits, allocs):
        """Runs on the destination and does a resources claim, if necessary.
        Currently, only NUMA live migrations require it.

        :param ctxt: Request context
        :param instance: The Instance being live migrated
        :param migrate_data: The MigrateData object for this live migration
        :param migration: The Migration object for this live migration
        :param limits: The SchedulerLimits object for this live migration
        :returns: migrate_data with dst_numa_info set if necessary
        """
        try:
            # NOTE(artom) We might have gotten here from _find_destination() in
            # the conductor live migrate task. At that point,
            # migration.dest_node is not set yet (nor should it be, we're still
            # looking for a destination, after all). Therefore, we cannot use
            # migration.dest_node here and must use self._get_nodename().
            claim = self.rt.live_migration_claim(
                ctxt, instance, self._get_nodename(instance), migration,
                limits, allocs)
            LOG.debug('Created live migration claim.', instance=instance)
        except exception.ComputeResourcesUnavailable as e:
            raise exception.MigrationPreCheckError(
                reason=e.format_message())
        return self.driver.post_claim_migrate_data(ctxt, instance,
                                                   migrate_data, claim)

    def _source_can_numa_live_migrate(self, ctxt, dest_check_data,
                                      source_check_data):
        # TODO(artom) Our virt driver may have told us that it supports NUMA
        # live migration. However, the following other conditions must be met
        # for a NUMA live migration to happen:
        # 1. We got a True dst_supports_numa_live_migration in
        #    dest_check_data, indicating that the dest virt driver supports
        #    NUMA live migration and that the conductor can send RPC 5.3 and
        #    that the destination compute manager can receive it.
        # 2. Ourselves, the source, can send RPC 5.3. There's no
        #    sentinel/parameter for this, so we just ask our rpcapi directly.
        # If any of these are not met, we need to remove the
        # src_supports_numa_live_migration flag from source_check_data to avoid
        # incorrectly initiating a NUMA live migration.
        # All of this can be removed in RPC 6.0/objects 2.0.
        can_numa_live_migrate = (
            'dst_supports_numa_live_migration' in dest_check_data and
            dest_check_data.dst_supports_numa_live_migration and
            self.compute_rpcapi.supports_numa_live_migration(ctxt))
        if ('src_supports_numa_live_migration' in source_check_data and
                source_check_data.src_supports_numa_live_migration and
                not can_numa_live_migrate):
            delattr(source_check_data, 'src_supports_numa_live_migration')
        return source_check_data

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def check_can_live_migrate_source(self, ctxt, instance, dest_check_data):
        """Check if it is possible to execute live migration.

        This checks if the live migration can succeed, based on the
        results from check_can_live_migrate_destination.

        :param ctxt: security context
        :param instance: dict of instance data
        :param dest_check_data: result of check_can_live_migrate_destination
        :returns: a LiveMigrateData object
        """
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            ctxt, instance.uuid)
        is_volume_backed = compute_utils.is_volume_backed_instance(
            ctxt, instance, bdms)
        dest_check_data.is_volume_backed = is_volume_backed
        block_device_info = self._get_instance_block_device_info(
                            ctxt, instance, refresh_conn_info=False, bdms=bdms)
        result = self.driver.check_can_live_migrate_source(ctxt, instance,
                                                           dest_check_data,
                                                           block_device_info)
        result = self._source_can_numa_live_migrate(ctxt, dest_check_data,
                                                    result)
        LOG.debug('source check data is %s', result)
        return result

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def pre_live_migration(self, context, instance, disk, migrate_data):
        """Preparations for live migration at dest host.

        :param context: security context
        :param instance: dict of instance data
        :param disk: disk info of instance
        :param migrate_data: A dict or LiveMigrateData object holding data
                             required for live migration without shared
                             storage.
        :returns: migrate_data containing additional migration info
        """
        LOG.debug('pre_live_migration data is %s', migrate_data)

        # Error out if this host cannot accept the new instance due
        # to anti-affinity. At this point the migration is already in-progress,
        # so this is the definitive moment to abort due to the policy
        # violation. Also, it should be safe to explode here. The instance
        # status remains ACTIVE, migration status failed.
        self._validate_instance_group_policy(context, instance)

        migrate_data.old_vol_attachment_ids = {}
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            context, instance.uuid)
        network_info = self.network_api.get_instance_nw_info(context, instance)
        self._notify_about_instance_usage(
            context, instance, "live_migration.pre.start",
            network_info=network_info)
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_PRE,
            phase=fields.NotificationPhase.START, bdms=bdms)

        connector = None
        try:
            for bdm in bdms:
                if bdm.is_volume and bdm.attachment_id is not None:
                    # This bdm uses the new cinder v3.44 API.
                    # We will create a new attachment for this
                    # volume on this migration destination host. The old
                    # attachment will be deleted on the source host
                    # when the migration succeeds. The old attachment_id
                    # is stored in dict with the key being the bdm.volume_id
                    # so it can be restored on rollback.
                    #
                    # Also note that attachment_update is not needed as we
                    # are providing the connector in the create call.
                    if connector is None:
                        connector = self.driver.get_volume_connector(instance)
                    attach_ref = self.volume_api.attachment_create(
                        context, bdm.volume_id, bdm.instance_uuid,
                        connector=connector, mountpoint=bdm.device_name)

                    # save current attachment so we can detach it on success,
                    # or restore it on a rollback.
                    # NOTE(mdbooth): This data is no longer used by the source
                    # host since change Ibe9215c0. We can't remove it until we
                    # are sure the source host has been upgraded.
                    migrate_data.old_vol_attachment_ids[bdm.volume_id] = \
                        bdm.attachment_id

                    # update the bdm with the new attachment_id.
                    bdm.attachment_id = attach_ref['id']
                    bdm.save()

            block_device_info = self._get_instance_block_device_info(
                                context, instance, refresh_conn_info=True,
                                bdms=bdms)

            # The driver pre_live_migration will plug vifs on the host
            migrate_data = self.driver.pre_live_migration(context,
                                           instance,
                                           block_device_info,
                                           network_info,
                                           disk,
                                           migrate_data)
            LOG.debug('driver pre_live_migration data is %s', migrate_data)
            # driver.pre_live_migration is what plugs vifs on the destination
            # host so now we can set the wait_for_vif_plugged flag in the
            # migrate_data object which the source compute will use to
            # determine if it should wait for a 'network-vif-plugged' event
            # from neutron before starting the actual guest transfer in the
            # hypervisor
            using_multiple_port_bindings = (
                'vifs' in migrate_data and migrate_data.vifs)
            migrate_data.wait_for_vif_plugged = (
                CONF.compute.live_migration_wait_for_vif_plug and
                using_multiple_port_bindings
            )

            # NOTE(tr3buchet): setup networks on destination host
            self.network_api.setup_networks_on_host(context, instance,
                                                             self.host)

            # NOTE(lyarwood): The above call to driver.pre_live_migration
            # can result in the virt drivers attempting to stash additional
            # metadata into the connection_info of the underlying bdm.
            # Ensure this is saved to the database by calling .save() against
            # the driver BDMs we passed down via block_device_info.
            for driver_bdm in block_device_info['block_device_mapping']:
                driver_bdm.save()

        except Exception:
            # If we raise, migrate_data with the updated attachment ids
            # will not be returned to the source host for rollback.
            # So we need to rollback new attachments here.
            with excutils.save_and_reraise_exception():
                old_attachments = migrate_data.old_vol_attachment_ids
                for bdm in bdms:
                    if (bdm.is_volume and bdm.attachment_id is not None and
                            bdm.volume_id in old_attachments):
                        self.volume_api.attachment_delete(context,
                                                          bdm.attachment_id)
                        bdm.attachment_id = old_attachments[bdm.volume_id]
                        bdm.save()

        # Volume connections are complete, tell cinder that all the
        # attachments have completed.
        for bdm in bdms:
            if bdm.is_volume and bdm.attachment_id is not None:
                self.volume_api.attachment_complete(context,
                                                    bdm.attachment_id)

        self._notify_about_instance_usage(
                     context, instance, "live_migration.pre.end",
                     network_info=network_info)
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_PRE,
            phase=fields.NotificationPhase.END, bdms=bdms)

        LOG.debug('pre_live_migration result data is %s', migrate_data)
        return migrate_data

    @staticmethod
    def _neutron_failed_live_migration_callback(event_name, instance):
        msg = ('Neutron reported failure during live migration '
               'with %(event)s for instance %(uuid)s')
        msg_args = {'event': event_name, 'uuid': instance.uuid}
        if CONF.vif_plugging_is_fatal:
            raise exception.VirtualInterfacePlugException(msg % msg_args)
        LOG.error(msg, msg_args)

    @staticmethod
    def _get_neutron_events_for_live_migration(instance):
        # We don't generate events if CONF.vif_plugging_timeout=0
        # meaning that the operator disabled using them.
        if CONF.vif_plugging_timeout:
            return (instance.get_network_info()
                    .get_live_migration_plug_time_events())
        else:
            return []

    def _cleanup_pre_live_migration(self, context, dest, instance,
                                    migration, migrate_data, source_bdms):
        """Helper method for when pre_live_migration fails

        Sets the migration status to "error" and rolls back the live migration
        setup on the destination host.

        :param context: The user request context.
        :type context: nova.context.RequestContext
        :param dest: The live migration destination hostname.
        :type dest: str
        :param instance: The instance being live migrated.
        :type instance: nova.objects.Instance
        :param migration: The migration record tracking this live migration.
        :type migration: nova.objects.Migration
        :param migrate_data: Data about the live migration, populated from
                             the destination host.
        :type migrate_data: Subclass of nova.objects.LiveMigrateData
        :param source_bdms: BDMs prior to modification by the destination
                            compute host. Set by _do_live_migration and not
                            part of the callback interface, so this is never
                            None
        """
        self._set_migration_status(migration, 'error')
        # Make sure we set this for _rollback_live_migration()
        # so it can find it, as expected if it was called later
        migrate_data.migration = migration
        self._rollback_live_migration(context, instance, dest,
                                      migrate_data=migrate_data,
                                      source_bdms=source_bdms,
                                      pre_live_migration=True)

    def _do_pre_live_migration_from_source(self, context, dest, instance,
                                           block_migration, migration,
                                           migrate_data, source_bdms):
        """Prepares for pre-live-migration on the source host and calls dest

        Will setup a callback networking event handler (if configured) and
        then call the dest host's pre_live_migration method to prepare the
        dest host for live migration (plugs vifs, connect volumes, etc).

        _rollback_live_migration (on the source) will be called if
        pre_live_migration (on the dest) fails.

        :param context: nova auth request context for this operation
        :param dest: name of the destination compute service host
        :param instance: Instance object being live migrated
        :param block_migration: If true, prepare for block migration.
        :param migration: Migration object tracking this operation
        :param migrate_data: MigrateData object for this operation populated
            by the destination host compute driver as part of the
            check_can_live_migrate_destination call.
        :param source_bdms: BlockDeviceMappingList of BDMs currently attached
            to the instance from the source host.
        :returns: MigrateData object which is a modified version of the
            ``migrate_data`` argument from the compute driver on the dest
            host during the ``pre_live_migration`` call.
        :raises: MigrationError if waiting for the network-vif-plugged event
            timed out and is fatal.
        """
        class _BreakWaitForInstanceEvent(Exception):
            """Used as a signal to stop waiting for the network-vif-plugged
            event when we discover that
            [compute]/live_migration_wait_for_vif_plug is not set on the
            destination.
            """
            pass

        events = self._get_neutron_events_for_live_migration(instance)
        try:
            if ('block_migration' in migrate_data and
                    migrate_data.block_migration):
                block_device_info = self._get_instance_block_device_info(
                    context, instance, bdms=source_bdms)
                disk = self.driver.get_instance_disk_info(
                    instance, block_device_info=block_device_info)
            else:
                disk = None

            deadline = CONF.vif_plugging_timeout
            error_cb = self._neutron_failed_live_migration_callback
            # In order to avoid a race with the vif plugging that the virt
            # driver does on the destination host, we register our events
            # to wait for before calling pre_live_migration. Then if the
            # dest host reports back that we shouldn't wait, we can break
            # out of the context manager using _BreakWaitForInstanceEvent.
            with self.virtapi.wait_for_instance_event(
                    instance, events, deadline=deadline,
                    error_callback=error_cb):
                with timeutils.StopWatch() as timer:
                    # TODO(mriedem): The "block_migration" parameter passed
                    # here is not actually used in pre_live_migration but it
                    # is not optional in the RPC interface either.
                    migrate_data = self.compute_rpcapi.pre_live_migration(
                        context, instance,
                        block_migration, disk, dest, migrate_data)
                LOG.info('Took %0.2f seconds for pre_live_migration on '
                         'destination host %s.',
                         timer.elapsed(), dest, instance=instance)
                wait_for_vif_plugged = (
                    'wait_for_vif_plugged' in migrate_data and
                    migrate_data.wait_for_vif_plugged)
                if events and not wait_for_vif_plugged:
                    raise _BreakWaitForInstanceEvent
        except _BreakWaitForInstanceEvent:
            if events:
                LOG.debug('Not waiting for events after pre_live_migration: '
                          '%s. ', events, instance=instance)
        except exception.VirtualInterfacePlugException:
            with excutils.save_and_reraise_exception():
                LOG.exception('Failed waiting for network virtual interfaces '
                              'to be plugged on the destination host %s.',
                              dest, instance=instance)
                self._cleanup_pre_live_migration(
                    context, dest, instance, migration, migrate_data,
                    source_bdms)
        except eventlet.timeout.Timeout:
            # We only get here if wait_for_vif_plugged is True which means
            # live_migration_wait_for_vif_plug=True on the destination host.
            msg = (
                'Timed out waiting for events: %(events)s. If these timeouts '
                'are a persistent issue it could mean the networking backend '
                'on host %(dest)s does not support sending these events '
                'unless there are port binding host changes which does not '
                'happen at this point in the live migration process. You may '
                'need to disable the live_migration_wait_for_vif_plug option '
                'on host %(dest)s.')
            subs = {'events': events, 'dest': dest}
            LOG.warning(msg, subs, instance=instance)
            if CONF.vif_plugging_is_fatal:
                self._cleanup_pre_live_migration(
                    context, dest, instance, migration, migrate_data,
                    source_bdms)
                raise exception.MigrationError(reason=msg % subs)
        except Exception:
            with excutils.save_and_reraise_exception():
                LOG.exception('Pre live migration failed at %s',
                              dest, instance=instance)
                self._cleanup_pre_live_migration(
                    context, dest, instance, migration, migrate_data,
                    source_bdms)
        return migrate_data

    def _do_live_migration(self, context, dest, instance, block_migration,
                           migration, migrate_data):
        # NOTE(danms): We should enhance the RT to account for migrations
        # and use the status field to denote when the accounting has been
        # done on source/destination. For now, this is just here for status
        # reporting
        self._set_migration_status(migration, 'preparing')
        source_bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)

        migrate_data = self._do_pre_live_migration_from_source(
            context, dest, instance, block_migration, migration, migrate_data,
            source_bdms)

        # Set migrate_data.migration because that is how _post_live_migration
        # and _rollback_live_migration get the migration object for cleanup.
        # Yes this is gross but changing the _post_live_migration and
        # _rollback_live_migration interfaces would also mean changing how the
        # virt drivers call them from the driver.live_migration method, i.e.
        # we would have to pass the migration object through the driver (or
        # consider using a partial but some do not like that pattern).
        migrate_data.migration = migration

        # NOTE(Kevin_Zheng): Pop the migration from the waiting queue
        # if it exist in the queue, then we are good to moving on, if
        # not, some other process must have aborted it, then we should
        # rollback.
        try:
            self._waiting_live_migrations.pop(instance.uuid)
        except KeyError:
            LOG.debug('Migration %s aborted by another process, rollback.',
                      migration.uuid, instance=instance)
            self._rollback_live_migration(context, instance, dest,
                                          migrate_data, 'cancelled',
                                          source_bdms=source_bdms)
            self._notify_live_migrate_abort_end(context, instance)
            return

        self._set_migration_status(migration, 'running')

        # NOTE(mdbooth): pre_live_migration will update connection_info and
        # attachment_id on all volume BDMS to reflect the new destination
        # host attachment. We fetch BDMs before that to retain connection_info
        # and attachment_id relating to the source host for post migration
        # cleanup.
        post_live_migration = functools.partial(
            self._post_live_migration_update_host, source_bdms=source_bdms
            )
        rollback_live_migration = functools.partial(
            self._rollback_live_migration, source_bdms=source_bdms)

        LOG.debug('live_migration data is %s', migrate_data)
        try:
            self.driver.live_migration(context, instance, dest,
                                       post_live_migration,
                                       rollback_live_migration,
                                       block_migration, migrate_data)
        except Exception:
            LOG.exception('Live migration failed.', instance=instance)
            with excutils.save_and_reraise_exception():
                # Put instance and migration into error state,
                # as its almost certainly too late to rollback
                self._set_migration_status(migration, 'error')
                # first refresh instance as it may have got updated by
                # post_live_migration_at_destination
                instance.refresh()
                self._set_instance_obj_error_state(instance,
                                                   clean_task_state=True)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @errors_out_migration
    @wrap_instance_fault
    def live_migration(self, context, dest, instance, block_migration,
                       migration, migrate_data):
        """Executing live migration.

        :param context: security context
        :param dest: destination host
        :param instance: a nova.objects.instance.Instance object
        :param block_migration: if true, prepare for block migration
        :param migration: an nova.objects.Migration object
        :param migrate_data: implementation specific params

        """
        self._set_migration_status(migration, 'queued')
        # NOTE(Kevin_Zheng): Submit the live_migration job to the pool and
        # put the returned Future object into dict mapped with migration.uuid
        # in order to be able to track and abort it in the future.
        self._waiting_live_migrations[instance.uuid] = (None, None)
        try:
            future = self._live_migration_executor.submit(
                self._do_live_migration, context, dest, instance,
                block_migration, migration, migrate_data)
            self._waiting_live_migrations[instance.uuid] = (migration, future)
        except RuntimeError:
            # GreenThreadPoolExecutor.submit will raise RuntimeError if the
            # pool is shutdown, which happens in
            # _cleanup_live_migrations_in_pool.
            LOG.info('Migration %s failed to submit as the compute service '
                     'is shutting down.', migration.uuid, instance=instance)
            raise exception.LiveMigrationNotSubmitted(
                migration_uuid=migration.uuid, instance_uuid=instance.uuid)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def live_migration_force_complete(self, context, instance):
        """Force live migration to complete.

        :param context: Security context
        :param instance: The instance that is being migrated
        """

        self._notify_about_instance_usage(
            context, instance, 'live.migration.force.complete.start')
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_FORCE_COMPLETE,
            phase=fields.NotificationPhase.START)
        self.driver.live_migration_force_complete(instance)
        self._notify_about_instance_usage(
            context, instance, 'live.migration.force.complete.end')
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_FORCE_COMPLETE,
            phase=fields.NotificationPhase.END)

    def _notify_live_migrate_abort_end(self, context, instance):
        self._notify_about_instance_usage(
            context, instance, 'live.migration.abort.end')
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_ABORT,
            phase=fields.NotificationPhase.END)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def live_migration_abort(self, context, instance, migration_id):
        """Abort an in-progress live migration.

        :param context: Security context
        :param instance: The instance that is being migrated
        :param migration_id: ID of in-progress live migration

        """
        self._notify_about_instance_usage(
            context, instance, 'live.migration.abort.start')
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_ABORT,
            phase=fields.NotificationPhase.START)
        # NOTE(Kevin_Zheng): Pop the migration out from the queue, this might
        # lead to 3 scenarios:
        # 1. The selected migration is still in queue, and the future.cancel()
        #    succeed, then the abort action is succeed, mark the migration
        #    status to 'cancelled'.
        # 2. The selected migration is still in queue, but the future.cancel()
        #    failed, then the _do_live_migration() has started executing, and
        #    the migration status is 'preparing', then we just pop it from the
        #    queue, and the migration process will handle it later. And the
        #    migration status couldn't be 'running' in this scenario because
        #    if _do_live_migration has started executing and we've already
        #    popped it from the queue and set the migration status to
        #    'running' at this point, popping it here will raise KeyError at
        #    which point we check if it's running and if so, we abort the old
        #    way.
        # 3. The selected migration is not in the queue, then the migration
        #    status is 'running', let the driver handle it.
        try:
            migration, future = (
                self._waiting_live_migrations.pop(instance.uuid))
            if future and future.cancel():
                # If we got here, we've successfully dropped a queued
                # migration from the queue, so _do_live_migration won't run
                # and we only need to revert minor changes introduced by Nova
                # control plane (port bindings, resource allocations and
                # instance's PCI devices), restore VM's state, set the
                # migration's status to cancelled and send the notification.
                # If Future.cancel() fails, it means _do_live_migration is
                # running and the migration status is preparing, and
                # _do_live_migration() itself will attempt to pop the queued
                # migration, hit a KeyError, and rollback, set the migration
                # to cancelled and send the live.migration.abort.end
                # notification.
                self._revert_allocation(context, instance, migration)
                try:
                    # This call will delete any inactive destination host
                    # port bindings.
                    self.network_api.setup_networks_on_host(
                        context, instance, host=migration.dest_compute,
                        teardown=True)
                except exception.PortBindingDeletionFailed as e:
                    # Removing the inactive port bindings from the destination
                    # host is not critical so just log an error but don't fail.
                    LOG.error(
                        'Network cleanup failed for destination host %s '
                        'during live migration rollback. You may need to '
                        'manually clean up resources in the network service. '
                        'Error: %s', migration.dest_compute, str(e))
                except Exception:
                    with excutils.save_and_reraise_exception():
                        LOG.exception(
                            'An error occurred while cleaning up networking '
                            'during live migration rollback.',
                            instance=instance)
                instance.task_state = None
                instance.save(expected_task_state=[task_states.MIGRATING])
                self._set_migration_status(migration, 'cancelled')
        except KeyError:
            migration = objects.Migration.get_by_id(context, migration_id)
            if migration.status != 'running':
                raise exception.InvalidMigrationState(
                    migration_id=migration_id, instance_uuid=instance.uuid,
                    state=migration.status, method='abort live migration')
            self.driver.live_migration_abort(instance)
        self._notify_live_migrate_abort_end(context, instance)

    def _live_migration_cleanup_flags(self, migrate_data, migr_ctxt=None):
        """Determine whether disks, instance path or other resources
        need to be cleaned up after live migration (at source on success,
        at destination on rollback)

        Block migration needs empty image at destination host before migration
        starts, so if any failure occurs, any empty images has to be deleted.

        Also Volume backed live migration w/o shared storage needs to delete
        newly created instance-xxx dir on the destination as a part of its
        rollback process

        There may be other resources which need cleanup; currently this is
        limited to vPMEM devices with the libvirt driver.

        :param migrate_data: implementation specific data
        :param migr_ctxt: specific resources stored in migration_context
        :returns: (bool, bool) -- do_cleanup, destroy_disks
        """
        # NOTE(pkoniszewski): block migration specific params are set inside
        # migrate_data objects for drivers that expose block live migration
        # information (i.e. Libvirt, HyperV). For other drivers cleanup is not
        # needed.
        do_cleanup = False
        destroy_disks = False
        if isinstance(migrate_data, migrate_data_obj.LibvirtLiveMigrateData):
            has_vpmem = False
            if migr_ctxt and migr_ctxt.old_resources:
                for resource in migr_ctxt.old_resources:
                    if ('metadata' in resource and
                        isinstance(resource.metadata,
                                   objects.LibvirtVPMEMDevice)):
                        has_vpmem = True
                        break
            # No instance booting at source host, but instance dir
            # must be deleted for preparing next block migration
            # must be deleted for preparing next live migration w/o shared
            # storage
            # vpmem must be cleaned
            do_cleanup = not migrate_data.is_shared_instance_path or has_vpmem
            destroy_disks = not migrate_data.is_shared_block_storage
        elif isinstance(migrate_data, migrate_data_obj.HyperVLiveMigrateData):
            # NOTE(claudiub): We need to cleanup any zombie Planned VM.
            do_cleanup = True
            destroy_disks = not migrate_data.is_shared_instance_path

        return (do_cleanup, destroy_disks)

    def _post_live_migration_remove_source_vol_connections(
            self, context, instance, source_bdms):
        """Disconnect volume connections from the source host during
        _post_live_migration.

        :param context: nova auth RequestContext
        :param instance: Instance object being live migrated
        :param source_bdms: BlockDeviceMappingList representing the attached
            volumes with connection_info set for the source host
        """
        # Detaching volumes.
        connector = None
        for bdm in source_bdms:
            if bdm.is_volume:
                # Detaching volumes is a call to an external API that can fail.
                # If it does, we need to handle it gracefully so that the call
                # to post_live_migration_at_destination - where we set instance
                # host and task state - still happens. We need to rethink the
                # current approach of setting instance host and task state
                # AFTER a whole bunch of things that could fail in unhandled
                # ways, but that is left as a TODO(artom).
                try:
                    if bdm.attachment_id is None:
                        # Prior to cinder v3.44:
                        # We don't want to actually mark the volume detached,
                        # or delete the bdm, just remove the connection from
                        # this host.
                        #
                        # remove the volume connection without detaching from
                        # hypervisor because the instance is not running
                        # anymore on the current host
                        if connector is None:
                            connector = self.driver.get_volume_connector(
                                instance)
                        self.volume_api.terminate_connection(context,
                                                             bdm.volume_id,
                                                             connector)
                    else:
                        # cinder v3.44 api flow - delete the old attachment
                        # for the source host
                        self.volume_api.attachment_delete(context,
                                                          bdm.attachment_id)

                except Exception as e:
                    if bdm.attachment_id is None:
                        LOG.error('Connection for volume %s not terminated on '
                                  'source host %s during post_live_migration: '
                                  '%s', bdm.volume_id, self.host,
                                  str(e), instance=instance)
                    else:
                        LOG.error('Volume attachment %s not deleted on source '
                                  'host %s during post_live_migration: %s',
                                  bdm.attachment_id, self.host,
                                  str(e), instance=instance)

    # TODO(sean-k-mooney): add typing
    def _post_live_migration_update_host(
        self, ctxt, instance, dest, block_migration=False,
        migrate_data=None, source_bdms=None
    ):
        try:
            self._post_live_migration(
                ctxt, instance, dest, block_migration, migrate_data,
                source_bdms)
        except Exception:
            # Restore the instance object
            node_name = None
            try:
                # get node name of compute, where instance will be
                # running after migration, that is destination host
                compute_node = self._get_compute_info(ctxt, dest)
                node_name = compute_node.hypervisor_hostname
            except exception.ComputeHostNotFound:
                LOG.exception('Failed to get compute_info for %s', dest)

            # we can never rollback from post live migration and we can only
            # get here if the instance is running on the dest so we ensure
            # the instance.host is set correctly and reraise the original
            # exception unmodified.
            if instance.host != dest:
                # apply saves the new fields while drop actually removes the
                # migration context from the instance, so migration persists.
                instance.apply_migration_context()
                instance.drop_migration_context()
                instance.host = dest
                instance.task_state = None
                instance.node = node_name
                instance.progress = 0
                instance.save()
            raise

    @wrap_exception()
    @wrap_instance_fault
    def _post_live_migration(self, ctxt, instance, dest,
                             block_migration=False, migrate_data=None,
                             source_bdms=None):
        """Post operations for live migration.

        This method is called from live_migration
        and mainly updating database record.

        :param ctxt: security context
        :param instance: instance object
        :param dest: destination host
        :param block_migration: if true, prepare for block migration
        :param migrate_data: if not None, it is a dict which has data
        :param source_bdms: BDMs prior to modification by the destination
                            compute host. Set by _do_live_migration and not
                            part of the callback interface, so this is never
                            None
        required for live migration without shared storage

        """
        LOG.info('_post_live_migration() is started..',
                 instance=instance)

        # Cleanup source host post live-migration
        block_device_info = self._get_instance_block_device_info(
                            ctxt, instance, bdms=source_bdms)
        self.driver.post_live_migration(ctxt, instance, block_device_info,
                                        migrate_data)

        # Disconnect volumes from this (the source) host.
        self._post_live_migration_remove_source_vol_connections(
            ctxt, instance, source_bdms)

        # NOTE(artom) At this point in time we have not bound the ports to the
        # destination host yet (this happens in migrate_instance_start()
        # below). Therefore, the "old" source network info that's still in the
        # instance info cache is safe to use here, since it'll be used below
        # during driver.post_live_migration_at_source() to unplug the VIFs on
        # the source.
        network_info = instance.get_network_info()

        self._notify_about_instance_usage(ctxt, instance,
                                          "live_migration._post.start",
                                          network_info=network_info)
        compute_utils.notify_about_instance_action(
            ctxt, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_POST,
            phase=fields.NotificationPhase.START)

        migration = objects.Migration(
            source_compute=self.host, dest_compute=dest,
        )
        # For neutron, migrate_instance_start will activate the destination
        # host port bindings, if there are any created by conductor before live
        # migration started.
        self.network_api.migrate_instance_start(ctxt,
                                                instance,
                                                migration)

        destroy_vifs = False
        try:
            # It's possible that the vif type changed on the destination
            # host and is already bound and active, so we need to use the
            # stashed source vifs in migrate_data.vifs (if present) to unplug
            # on the source host.
            unplug_nw_info = network_info
            if migrate_data and 'vifs' in migrate_data:
                nw_info = []
                for migrate_vif in migrate_data.vifs:
                    nw_info.append(migrate_vif.source_vif)
                unplug_nw_info = network_model.NetworkInfo.hydrate(nw_info)
                LOG.debug('Calling driver.post_live_migration_at_source '
                          'with original source VIFs from migrate_data: %s',
                          unplug_nw_info, instance=instance)
            self.driver.post_live_migration_at_source(ctxt, instance,
                                                      unplug_nw_info)
        except NotImplementedError as ex:
            LOG.debug(ex, instance=instance)
            # For all hypervisors other than libvirt, there is a possibility
            # they are unplugging networks from source node in the cleanup
            # method
            destroy_vifs = True

        # Free instance allocations on source before claims are allocated on
        # destination node
        self.rt.free_pci_device_allocations_for_instance(ctxt, instance)
        # NOTE(danms): Save source node before calling post method on
        # destination, which will update it
        source_node = instance.node

        do_cleanup, destroy_disks = self._live_migration_cleanup_flags(
            migrate_data, migr_ctxt=instance.migration_context)

        if do_cleanup:
            LOG.debug('Calling driver.cleanup from _post_live_migration',
                      instance=instance)
            self.driver.cleanup(ctxt, instance, unplug_nw_info,
                                destroy_disks=destroy_disks,
                                migrate_data=migrate_data,
                                destroy_vifs=destroy_vifs)

        # Define domain at destination host, without doing it,
        # pause/suspend/terminate do not work.
        post_at_dest_success = True
        try:
            self.compute_rpcapi.post_live_migration_at_destination(ctxt,
                    instance, block_migration, dest)
        except Exception as error:
            post_at_dest_success = False
            # We don't want to break _post_live_migration() if
            # post_live_migration_at_destination() fails as it should never
            # affect cleaning up source node.
            LOG.exception("Post live migration at destination %s failed",
                          dest, instance=instance, error=error)

        self.instance_events.clear_events_for_instance(instance)

        # NOTE(timello): make sure we update available resources on source
        # host even before next periodic task.
        self.update_available_resource(ctxt)

        self._update_scheduler_instance_info(ctxt, instance)
        self._notify_about_instance_usage(ctxt, instance,
                                          "live_migration._post.end",
                                          network_info=network_info)
        compute_utils.notify_about_instance_action(
            ctxt, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_POST,
            phase=fields.NotificationPhase.END)
        if post_at_dest_success:
            LOG.info('Migrating instance to %s finished successfully.',
                     dest, instance=instance)

        self._clean_instance_console_tokens(ctxt, instance)
        if migrate_data and migrate_data.obj_attr_is_set('migration'):
            migrate_data.migration.status = 'completed'
            migrate_data.migration.save()
            self._delete_allocation_after_move(ctxt,
                                               instance,
                                               migrate_data.migration)
        else:
            # We didn't have data on a migration, which means we can't
            # look up to see if we had new-style migration-based
            # allocations. This should really only happen in cases of
            # a buggy virt driver. Log a warning so we know it happened.
            LOG.warning('Live migration ended with no migrate_data '
                        'record. Unable to clean up migration-based '
                        'allocations for node %s which is almost certainly '
                        'not an expected situation.', source_node,
                        instance=instance)

    def _consoles_enabled(self):
        """Returns whether a console is enable."""
        return (CONF.vnc.enabled or CONF.spice.enabled or
                CONF.rdp.enabled or CONF.serial_console.enabled or
                CONF.mks.enabled)

    def _clean_instance_console_tokens(self, ctxt, instance):
        """Clean console tokens stored for an instance."""
        # If the database backend isn't in use, don't bother trying to clean
        # tokens.
        if self._consoles_enabled():
            objects.ConsoleAuthToken.\
                clean_console_auths_for_instance(ctxt, instance.uuid)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def post_live_migration_at_destination(self, context, instance,
                                           block_migration):
        """Post operations for live migration .

        :param context: security context
        :param instance: Instance dict
        :param block_migration: if true, prepare for block migration

        """
        LOG.info('Post operation of migration started',
                 instance=instance)

        # NOTE(tr3buchet): setup networks on destination host
        #                  this is called a second time because
        #                  multi_host does not create the bridge in
        #                  plug_vifs
        # NOTE(mriedem): This is a no-op for neutron.
        self.network_api.setup_networks_on_host(context, instance,
                                                         self.host)
        migration = objects.Migration(
            source_compute=instance.host,
            dest_compute=self.host,
            migration_type=fields.MigrationType.LIVE_MIGRATION)
        self.network_api.migrate_instance_finish(
            context, instance, migration, provider_mappings=None)

        network_info = self.network_api.get_instance_nw_info(context, instance)
        self._notify_about_instance_usage(
                     context, instance, "live_migration.post.dest.start",
                     network_info=network_info)
        compute_utils.notify_about_instance_action(context, instance,
                self.host,
                action=fields.NotificationAction.LIVE_MIGRATION_POST_DEST,
                phase=fields.NotificationPhase.START)
        block_device_info = self._get_instance_block_device_info(context,
                                                                 instance)
        # Allocate the claimed PCI resources at destination.
        self.rt.allocate_pci_devices_for_instance(context, instance)

        try:
            self.driver.post_live_migration_at_destination(
                context, instance, network_info, block_migration,
                block_device_info)
        except Exception:
            with excutils.save_and_reraise_exception():
                instance.vm_state = vm_states.ERROR
                LOG.error('Unexpected error during post live migration at '
                          'destination host.', instance=instance)
        finally:
            # Restore instance state and update host
            current_power_state = self._get_power_state(instance)
            node_name = None
            prev_host = instance.host
            try:
                compute_node = self._get_compute_info(context, self.host)
                node_name = compute_node.hypervisor_hostname
            except exception.ComputeHostNotFound:
                LOG.exception('Failed to get compute_info for %s', self.host)
            finally:
                # NOTE(artom) We need to apply the migration context here
                # regardless of whether the driver's
                # post_live_migration_at_destination succeeded or not: the
                # instance is on the destination, potentially with a new NUMA
                # topology and resource usage. We need to persist that.
                # NOTE(artom) Apply followed by drop looks weird, but apply
                # just saves the new fields while drop actually removes the
                # migration context from the instance.
                instance.apply_migration_context()
                instance.drop_migration_context()
                instance.host = self.host
                instance.power_state = current_power_state
                instance.task_state = None
                instance.node = node_name
                instance.progress = 0
                instance.save(expected_task_state=task_states.MIGRATING)

        # NOTE(tr3buchet): tear down networks on source host (nova-net)
        # NOTE(mriedem): For neutron, this will delete any inactive source
        # host port bindings.
        try:
            self.network_api.setup_networks_on_host(context, instance,
                                                    prev_host, teardown=True)
        except exception.PortBindingDeletionFailed as e:
            # Removing the inactive port bindings from the source host is not
            # critical so just log an error but don't fail.
            LOG.error('Network cleanup failed for source host %s during post '
                      'live migration. You may need to manually clean up '
                      'resources in the network service. Error: %s',
                      prev_host, str(e))
        # NOTE(vish): this is necessary to update dhcp for nova-network
        # NOTE(mriedem): This is a no-op for neutron.
        self.network_api.setup_networks_on_host(context, instance, self.host)
        self._notify_about_instance_usage(
                     context, instance, "live_migration.post.dest.end",
                     network_info=network_info)
        compute_utils.notify_about_instance_action(context, instance,
                self.host,
                action=fields.NotificationAction.LIVE_MIGRATION_POST_DEST,
                phase=fields.NotificationPhase.END)

    def _remove_remote_volume_connections(self, context, dest, bdms, instance):
        """Rollback remote volume connections on the dest"""
        for bdm in bdms:
            try:
                # remove the connection on the destination host
                # NOTE(lyarwood): This actually calls the cinderv2
                # os-terminate_connection API if required.
                self.compute_rpcapi.remove_volume_connection(
                        context, instance, bdm.volume_id, dest)
            except Exception:
                LOG.warning("Ignoring exception while attempting "
                            "to rollback volume connections for "
                            "volume %s on host %s.", bdm.volume_id,
                            dest, instance=instance)

    def _rollback_volume_bdms(self, context, bdms, original_bdms, instance):
        """Rollback the connection_info and attachment_id for each bdm"""
        original_bdms_by_volid = {bdm.volume_id: bdm for bdm in original_bdms
                                  if bdm.is_volume}
        for bdm in bdms:
            try:
                original_bdm = original_bdms_by_volid[bdm.volume_id]
                # NOTE(lyarwood): Only delete the referenced attachment if it
                # is different to the original in order to avoid accidentally
                # removing the source host volume attachment after it has
                # already been rolled back by a failure in pre_live_migration.
                if (bdm.attachment_id and original_bdm.attachment_id and
                    bdm.attachment_id != original_bdm.attachment_id):
                    # NOTE(lyarwood): 3.44 cinder api flow. Delete the
                    # attachment used by the bdm and reset it to that of
                    # the original bdm.
                    self.volume_api.attachment_delete(context,
                                                      bdm.attachment_id)
                    bdm.attachment_id = original_bdm.attachment_id
                # NOTE(lyarwood): Reset the connection_info to the original
                bdm.connection_info = original_bdm.connection_info
                bdm.save()
            except cinder_exception.ClientException:
                LOG.warning("Ignoring cinderclient exception when "
                            "attempting to delete attachment %s for volume "
                            "%s while rolling back volume bdms.",
                            bdm.attachment_id, bdm.volume_id,
                            instance=instance)
            except Exception:
                with excutils.save_and_reraise_exception():
                    LOG.exception("Exception while attempting to rollback "
                                  "BDM for volume %s.", bdm.volume_id,
                                  instance=instance)

    @wrap_exception()
    @wrap_instance_fault
    def _rollback_live_migration(self, context, instance,
                                 dest, migrate_data=None,
                                 migration_status='failed',
                                 source_bdms=None,
                                 pre_live_migration=False):
        """Recovers Instance/volume state from migrating -> running.

        :param context: security context
        :param instance: nova.objects.instance.Instance object
        :param dest:
            This method is called from live migration src host.
            This param specifies destination host.
        :param migrate_data:
            if not none, contains implementation specific data.
        :param migration_status:
            Contains the status we want to set for the migration object
        :param source_bdms: BDMs prior to modification by the destination
                            compute host. Set by _do_live_migration and not
                            part of the callback interface, so this is never
                            None

        """
        # NOTE(gibi): We need to refresh pci_requests of the instance as it
        # might be changed by the conductor during scheduling based on the
        # selected destination host. If the instance has SRIOV ports with
        # resource request then the LiveMigrationTask._find_destination call
        # updated the instance.pci_requests.requests[].spec with the SRIOV PF
        # device name to be used on the destination host. As the migration is
        # rolling back to the source host now we don't want to persist the
        # destination host related changes in the DB.
        instance.pci_requests = \
            objects.InstancePCIRequests.get_by_instance_uuid(
                context, instance.uuid)

        if (isinstance(migrate_data, migrate_data_obj.LiveMigrateData) and
              migrate_data.obj_attr_is_set('migration')):
            migration = migrate_data.migration
        else:
            migration = None

        if migration:
            # Remove allocations created in Placement for the dest node.
            # If migration is None, the virt driver didn't pass it which is
            # a bug.
            self._revert_allocation(context, instance, migration)
        else:
            LOG.error('Unable to revert allocations during live migration '
                      'rollback; compute driver did not provide migrate_data',
                      instance=instance)

        # NOTE(tr3buchet): setup networks on source host (really it's re-setup
        #                  for nova-network)
        # NOTE(mriedem): This is a no-op for neutron.
        self.network_api.setup_networks_on_host(context, instance, self.host)

        # NOTE(erlon): We should make sure that rollback_live_migration_at_src
        # is not called in the pre_live_migration rollback as that will trigger
        # the src host to re-attach interfaces which were not detached
        # previously.
        if not pre_live_migration:
            self.driver.rollback_live_migration_at_source(context, instance,
                                                          migrate_data)

        # NOTE(lyarwood): Fetch the current list of BDMs, disconnect any
        # connected volumes from the dest and delete any volume attachments
        # used by the destination host before rolling back to the original
        # still valid source host volume attachments.
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)
        # TODO(lyarwood): Turn the following into a lookup method within
        # BlockDeviceMappingList.
        vol_bdms = [bdm for bdm in bdms if bdm.is_volume]
        self._remove_remote_volume_connections(context, dest, vol_bdms,
                                               instance)
        self._rollback_volume_bdms(context, vol_bdms, source_bdms, instance)

        self._notify_about_instance_usage(context, instance,
                                          "live_migration._rollback.start")
        compute_utils.notify_about_instance_action(context, instance,
                self.host,
                action=fields.NotificationAction.LIVE_MIGRATION_ROLLBACK,
                phase=fields.NotificationPhase.START,
                bdms=bdms)

        do_cleanup, destroy_disks = self._live_migration_cleanup_flags(
                migrate_data, migr_ctxt=instance.migration_context)

        if do_cleanup:
            self.compute_rpcapi.rollback_live_migration_at_destination(
                    context, instance, dest, destroy_disks=destroy_disks,
                    migrate_data=migrate_data)
        else:
            # The port binding profiles need to be cleaned up.
            with errors_out_migration_ctxt(migration):
                try:
                    # This call will delete any inactive destination host
                    # port bindings.
                    self.network_api.setup_networks_on_host(
                        context, instance, host=dest, teardown=True)
                except exception.PortBindingDeletionFailed as e:
                    # Removing the inactive port bindings from the destination
                    # host is not critical so just log an error but don't fail.
                    LOG.error(
                        'Network cleanup failed for destination host %s '
                        'during live migration rollback. You may need to '
                        'manually clean up resources in the network service. '
                        'Error: %s', dest, str(e))
                except Exception:
                    with excutils.save_and_reraise_exception():
                        LOG.exception(
                            'An error occurred while cleaning up networking '
                            'during live migration rollback.',
                            instance=instance)

        # NOTE(luyao): We drop move_claim and migration_context after cleanup
        # is complete, to ensure the specific resources claimed on destination
        # are released safely.
        # TODO(artom) drop_move_claim_at_destination() is new in RPC 5.3, only
        # call it if we performed a NUMA-aware live migration (which implies us
        # being able to send RPC 5.3). To check this, we can use the
        # src_supports_numa_live_migration flag, as it will be set if and only
        # if:
        # - dst_supports_numa_live_migration made its way to the source
        #   (meaning both dest and source are new and conductor can speak
        #   RPC 5.3)
        # - src_supports_numa_live_migration was set by the source driver and
        #   passed the send-RPC-5.3 check.
        # This check can be removed in RPC 6.0.
        if ('src_supports_numa_live_migration' in migrate_data and
                migrate_data.src_supports_numa_live_migration):
            LOG.debug('Calling destination to drop move claim.',
                      instance=instance)
            self.compute_rpcapi.drop_move_claim_at_destination(context,
                                                               instance, dest)

        # NOTE(luyao): We only update instance info after rollback operations
        # are complete
        instance.task_state = None
        instance.progress = 0
        instance.drop_migration_context()
        instance.save(expected_task_state=[task_states.MIGRATING])

        self._notify_about_instance_usage(context, instance,
                                          "live_migration._rollback.end")
        compute_utils.notify_about_instance_action(context, instance,
                self.host,
                action=fields.NotificationAction.LIVE_MIGRATION_ROLLBACK,
                phase=fields.NotificationPhase.END,
                bdms=bdms)

        # NOTE(luyao): we have cleanup everything and get instance
        # back to normal status, now set migration status to 'failed'
        self._set_migration_status(migration, migration_status)

    @wrap_exception()
    @wrap_instance_fault
    def drop_move_claim_at_destination(self, context, instance):
        """Called by the source of a live migration during rollback to ask the
        destination to drop the MoveClaim object that was created for the live
        migration on the destination.
        """
        nodename = self._get_nodename(instance)
        LOG.debug('Dropping live migration resource claim on destination '
                  'node %s', nodename, instance=instance)
        self.rt.drop_move_claim(
            context, instance, nodename, flavor=instance.flavor)

    @wrap_exception()
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def rollback_live_migration_at_destination(self, context, instance,
                                               destroy_disks,
                                               migrate_data):
        """Cleaning up image directory that is created pre_live_migration.

        :param context: security context
        :param instance: a nova.objects.instance.Instance object sent over rpc
        :param destroy_disks: whether to destroy volumes or not
        :param migrate_data: contains migration info
        """
        network_info = self.network_api.get_instance_nw_info(context, instance)
        self._notify_about_instance_usage(
                      context, instance, "live_migration.rollback.dest.start",
                      network_info=network_info)
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_ROLLBACK_DEST,
            phase=fields.NotificationPhase.START)
        try:
            # NOTE(tr3buchet): tear down networks on dest host (nova-net)
            # NOTE(mriedem): For neutron, this call will delete any
            # destination host port bindings.
            # TODO(mriedem): We should eventually remove this call from
            # this method (rollback_live_migration_at_destination) since this
            # method is only called conditionally based on whether or not the
            # instance is running on shared storage. _rollback_live_migration
            # already calls this method for neutron if we are running on
            # shared storage.
            self.network_api.setup_networks_on_host(context, instance,
                                                    self.host, teardown=True)
        except exception.PortBindingDeletionFailed as e:
            # Removing the inactive port bindings from the destination
            # host is not critical so just log an error but don't fail.
            LOG.error(
                'Network cleanup failed for destination host %s '
                'during live migration rollback. You may need to '
                'manually clean up resources in the network service. '
                'Error: %s', self.host, str(e))
        except Exception:
            with excutils.save_and_reraise_exception():
                # NOTE(tdurakov): even if teardown networks fails driver
                # should try to rollback live migration on destination.
                LOG.exception('An error occurred while deallocating network.',
                              instance=instance)
        finally:
            # always run this even if setup_networks_on_host fails
            # NOTE(vish): The mapping is passed in so the driver can disconnect
            #             from remote volumes if necessary
            block_device_info = self._get_instance_block_device_info(context,
                                                                     instance)
            # free any instance PCI claims done on destination during
            # check_can_live_migrate_destination()
            self.rt.free_pci_device_claims_for_instance(context, instance)

            # NOTE(luyao): Apply migration_context temporarily since it's
            # on destination host, we rely on instance object to cleanup
            # specific resources like vpmem
            with instance.mutated_migration_context():
                self.driver.rollback_live_migration_at_destination(
                    context, instance, network_info, block_device_info,
                    destroy_disks=destroy_disks, migrate_data=migrate_data)

        self._notify_about_instance_usage(
                        context, instance, "live_migration.rollback.dest.end",
                        network_info=network_info)
        compute_utils.notify_about_instance_action(
            context, instance, self.host,
            action=fields.NotificationAction.LIVE_MIGRATION_ROLLBACK_DEST,
            phase=fields.NotificationPhase.END)

    def _require_nw_info_update(self, context, instance):
        """Detect whether there is a mismatch in binding:host_id, or
        binding_failed or unbound binding:vif_type for any of the instances
        ports.
        """
        # Only update port bindings if compute manager does manage port
        # bindings instead of the compute driver. For example IronicDriver
        # manages the port binding for baremetal instance ports, hence,
        # external intervention with the binding is not desired.
        if self.driver.manages_network_binding_host_id():
            return False

        search_opts = {'device_id': instance.uuid,
                       'fields': ['binding:host_id', 'binding:vif_type']}
        ports = self.network_api.list_ports(context, **search_opts)
        for p in ports['ports']:
            if p.get('binding:host_id') != self.host:
                return True
            vif_type = p.get('binding:vif_type')
            if (vif_type == network_model.VIF_TYPE_UNBOUND or
                    vif_type == network_model.VIF_TYPE_BINDING_FAILED):
                return True
        return False

    @periodic_task.periodic_task(
        spacing=CONF.heal_instance_info_cache_interval)
    def _heal_instance_info_cache(self, context):
        """Called periodically.  On every call, try to update the
        info_cache's network information for another instance by
        calling to the network manager.

        This is implemented by keeping a cache of uuids of instances
        that live on this host.  On each call, we pop one off of a
        list, pull the DB record, and try the call to the network API.
        If anything errors don't fail, as it's possible the instance
        has been deleted, etc.
        """
        heal_interval = CONF.heal_instance_info_cache_interval
        if not heal_interval:
            return

        instance_uuids = getattr(self, '_instance_uuids_to_heal', [])
        instance = None

        LOG.debug('Starting heal instance info cache')

        if not instance_uuids:
            # The list of instances to heal is empty so rebuild it
            LOG.debug('Rebuilding the list of instances to heal')
            db_instances = objects.InstanceList.get_by_host(
                context, self.host, expected_attrs=[], use_slave=True)
            for inst in db_instances:
                # We don't want to refresh the cache for instances
                # which are building or deleting so don't put them
                # in the list. If they are building they will get
                # added to the list next time we build it.
                if (inst.vm_state == vm_states.BUILDING):
                    LOG.debug('Skipping network cache update for instance '
                              'because it is Building.', instance=inst)
                    continue
                if (inst.task_state == task_states.DELETING):
                    LOG.debug('Skipping network cache update for instance '
                              'because it is being deleted.', instance=inst)
                    continue

                if not instance:
                    # Save the first one we find so we don't
                    # have to get it again
                    instance = inst
                else:
                    instance_uuids.append(inst['uuid'])

            self._instance_uuids_to_heal = instance_uuids
        else:
            # Find the next valid instance on the list
            while instance_uuids:
                try:
                    inst = objects.Instance.get_by_uuid(
                            context, instance_uuids.pop(0),
                            expected_attrs=['system_metadata', 'info_cache',
                                            'flavor'],
                            use_slave=True)
                except exception.InstanceNotFound:
                    # Instance is gone.  Try to grab another.
                    continue

                # Check the instance hasn't been migrated
                if inst.host != self.host:
                    LOG.debug('Skipping network cache update for instance '
                              'because it has been migrated to another '
                              'host.', instance=inst)
                # Check the instance isn't being deleting
                elif inst.task_state == task_states.DELETING:
                    LOG.debug('Skipping network cache update for instance '
                              'because it is being deleted.', instance=inst)
                else:
                    instance = inst
                    break

        if instance:
            # We have an instance now to refresh
            try:
                # Fix potential mismatch in port binding if evacuation failed
                # after reassigning the port binding to the dest host but
                # before the instance host is changed.
                # Do this only when instance has no pending task.
                if instance.task_state is None and \
                        self._require_nw_info_update(context, instance):
                    LOG.info("Updating ports in neutron", instance=instance)
                    self.network_api.setup_instance_network_on_host(
                        context, instance, self.host)
                # Call to network API to get instance info.. this will
                # force an update to the instance's info_cache
                self.network_api.get_instance_nw_info(
                    context, instance, force_refresh=True)
                LOG.debug('Updated the network info_cache for instance',
                          instance=instance)
            except exception.InstanceNotFound:
                # Instance is gone.
                LOG.debug('Instance no longer exists. Unable to refresh',
                          instance=instance)
                return
            except exception.InstanceInfoCacheNotFound:
                # InstanceInfoCache is gone.
                LOG.debug('InstanceInfoCache no longer exists. '
                          'Unable to refresh', instance=instance)
            except Exception:
                LOG.error('An error occurred while refreshing the network '
                          'cache.', instance=instance, exc_info=True)
        else:
            LOG.debug("Didn't find any instances for network info cache "
                      "update.")

    @periodic_task.periodic_task
    def _poll_rebooting_instances(self, context):
        if CONF.reboot_timeout > 0:
            filters = {'task_state':
                       [task_states.REBOOTING,
                        task_states.REBOOT_STARTED,
                        task_states.REBOOT_PENDING],
                       'host': self.host}
            rebooting = objects.InstanceList.get_by_filters(
                context, filters, expected_attrs=[], use_slave=True)

            to_poll = []
            for instance in rebooting:
                if timeutils.is_older_than(instance.updated_at,
                                           CONF.reboot_timeout):
                    to_poll.append(instance)

            self.driver.poll_rebooting_instances(CONF.reboot_timeout, to_poll)

    @periodic_task.periodic_task
    def _poll_rescued_instances(self, context):
        if CONF.rescue_timeout > 0:
            filters = {'vm_state': vm_states.RESCUED,
                       'host': self.host}
            rescued_instances = objects.InstanceList.get_by_filters(
                context, filters, expected_attrs=["system_metadata"],
                use_slave=True)

            to_unrescue = []
            for instance in rescued_instances:
                if timeutils.is_older_than(instance.launched_at,
                                           CONF.rescue_timeout):
                    to_unrescue.append(instance)

            for instance in to_unrescue:
                self.compute_api.unrescue(context, instance)

    @periodic_task.periodic_task
    def _poll_unconfirmed_resizes(self, context):
        if CONF.resize_confirm_window == 0:
            return

        migrations = objects.MigrationList.get_unconfirmed_by_dest_compute(
                context, CONF.resize_confirm_window, self.host,
                use_slave=True)

        migrations_info = dict(migration_count=len(migrations),
                confirm_window=CONF.resize_confirm_window)

        if migrations_info["migration_count"] > 0:
            LOG.info("Found %(migration_count)d unconfirmed migrations "
                     "older than %(confirm_window)d seconds",
                     migrations_info)

        def _set_migration_to_error(migration, reason, **kwargs):
            LOG.warning("Setting migration %(migration_id)s to error: "
                        "%(reason)s",
                        {'migration_id': migration.id, 'reason': reason},
                        **kwargs)
            migration.status = 'error'
            migration.save()

        for migration in migrations:
            instance_uuid = migration.instance_uuid
            LOG.info("Automatically confirming migration "
                     "%(migration_id)s for instance %(instance_uuid)s",
                     {'migration_id': migration.id,
                      'instance_uuid': instance_uuid})
            expected_attrs = ['metadata', 'system_metadata']
            try:
                instance = objects.Instance.get_by_uuid(context,
                            instance_uuid, expected_attrs=expected_attrs,
                            use_slave=True)
            except exception.InstanceNotFound:
                reason = (_("Instance %s not found") %
                          instance_uuid)
                _set_migration_to_error(migration, reason)
                continue
            if instance.vm_state == vm_states.ERROR:
                reason = _("In ERROR state")
                _set_migration_to_error(migration, reason,
                                        instance=instance)
                continue
            # race condition: The instance in DELETING state should not be
            # set the migration state to error, otherwise the instance in
            # to be deleted which is in RESIZED state
            # will not be able to confirm resize
            if instance.task_state in [task_states.DELETING,
                                       task_states.SOFT_DELETING]:
                msg = ("Instance being deleted or soft deleted during resize "
                       "confirmation. Skipping.")
                LOG.debug(msg, instance=instance)
                continue

            # race condition: This condition is hit when this method is
            # called between the save of the migration record with a status of
            # finished and the save of the instance object with a state of
            # RESIZED. The migration record should not be set to error.
            if instance.task_state == task_states.RESIZE_FINISH:
                msg = ("Instance still resizing during resize "
                       "confirmation. Skipping.")
                LOG.debug(msg, instance=instance)
                continue

            vm_state = instance.vm_state
            task_state = instance.task_state
            if vm_state != vm_states.RESIZED or task_state is not None:
                reason = (_("In states %(vm_state)s/%(task_state)s, not "
                           "RESIZED/None") %
                          {'vm_state': vm_state,
                           'task_state': task_state})
                _set_migration_to_error(migration, reason,
                                        instance=instance)
                continue
            try:
                self.compute_api.confirm_resize(context, instance,
                                                migration=migration)
            except Exception as e:
                LOG.info("Error auto-confirming resize: %s. "
                         "Will retry later.", e, instance=instance)

    @periodic_task.periodic_task(spacing=CONF.shelved_poll_interval)
    def _poll_shelved_instances(self, context):

        if CONF.shelved_offload_time <= 0:
            return

        filters = {'vm_state': vm_states.SHELVED,
                   'task_state': None,
                   'host': self.host}
        shelved_instances = objects.InstanceList.get_by_filters(
            context, filters=filters, expected_attrs=['system_metadata'],
            use_slave=True)

        to_gc = []
        for instance in shelved_instances:
            sys_meta = instance.system_metadata
            shelved_at = timeutils.parse_strtime(sys_meta['shelved_at'])
            if timeutils.is_older_than(shelved_at, CONF.shelved_offload_time):
                to_gc.append(instance)

        cyclient = cyborg.get_client(context)
        for instance in to_gc:
            try:
                instance.task_state = task_states.SHELVING_OFFLOADING
                instance.save(expected_task_state=(None,))
                accel_uuids = []
                if instance.flavor.extra_specs.get('accel:device_profile'):
                    # TODO(brinzhang): After cyborg support batch query ARQs
                    # for more than one instances, we will improve efficiency
                    # with this implemention.
                    accel_uuids = cyclient.get_arq_uuids_for_instance(instance)
                self.shelve_offload_instance(
                    context, instance, clean_shutdown=False,
                    accel_uuids=accel_uuids)
            except Exception:
                LOG.exception('Periodic task failed to offload instance.',
                              instance=instance)

    @periodic_task.periodic_task
    def _instance_usage_audit(self, context):
        if not CONF.instance_usage_audit:
            return

        begin, end = utils.last_completed_audit_period()
        if objects.TaskLog.get(context, 'instance_usage_audit', begin, end,
                               self.host):
            return

        instances = objects.InstanceList.get_active_by_window_joined(
            context, begin, end, host=self.host,
            expected_attrs=['system_metadata', 'info_cache', 'metadata',
                            'flavor'],
            use_slave=True)
        num_instances = len(instances)
        errors = 0
        successes = 0
        LOG.info("Running instance usage audit for host %(host)s "
                 "from %(begin_time)s to %(end_time)s. "
                 "%(number_instances)s instances.",
                 {'host': self.host,
                  'begin_time': begin,
                  'end_time': end,
                  'number_instances': num_instances})
        start_time = time.time()
        task_log = objects.TaskLog(context)
        task_log.task_name = 'instance_usage_audit'
        task_log.period_beginning = begin
        task_log.period_ending = end
        task_log.host = self.host
        task_log.task_items = num_instances
        task_log.message = 'Instance usage audit started...'
        task_log.begin_task()
        for instance in instances:
            try:
                compute_utils.notify_usage_exists(
                    self.notifier, context, instance, self.host,
                    ignore_missing_network_data=False)
                successes += 1
            except Exception:
                LOG.exception('Failed to generate usage '
                              'audit for instance '
                              'on host %s', self.host,
                              instance=instance)
                errors += 1
        task_log.errors = errors
        task_log.message = (
            'Instance usage audit ran for host %s, %s instances in %s seconds.'
            % (self.host, num_instances, time.time() - start_time))
        task_log.end_task()

    def _get_host_volume_bdms(self, context, use_slave=False):
        """Return all block device mappings on a compute host."""
        compute_host_bdms = []
        instances = objects.InstanceList.get_by_host(context, self.host,
            use_slave=use_slave)
        for instance in instances:
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                    context, instance.uuid, use_slave=use_slave)
            instance_bdms = [bdm for bdm in bdms if bdm.is_volume]
            compute_host_bdms.append(dict(instance=instance,
                                          instance_bdms=instance_bdms))

        return compute_host_bdms

    def _update_volume_usage_cache(self, context, vol_usages):
        """Updates the volume usage cache table with a list of stats."""
        for usage in vol_usages:
            # Allow switching of greenthreads between queries.
            greenthread.sleep(0)
            vol_usage = objects.VolumeUsage(context)
            vol_usage.volume_id = usage['volume']
            vol_usage.instance_uuid = usage['instance'].uuid
            vol_usage.project_id = usage['instance'].project_id
            vol_usage.user_id = usage['instance'].user_id
            vol_usage.availability_zone = usage['instance'].availability_zone
            vol_usage.curr_reads = usage['rd_req']
            vol_usage.curr_read_bytes = usage['rd_bytes']
            vol_usage.curr_writes = usage['wr_req']
            vol_usage.curr_write_bytes = usage['wr_bytes']
            vol_usage.save()
            self.notifier.info(context, 'volume.usage', vol_usage.to_dict())
            compute_utils.notify_about_volume_usage(context, vol_usage,
                                                    self.host)

    @periodic_task.periodic_task(spacing=CONF.volume_usage_poll_interval)
    def _poll_volume_usage(self, context):
        if CONF.volume_usage_poll_interval == 0:
            return

        compute_host_bdms = self._get_host_volume_bdms(context,
                                                       use_slave=True)
        if not compute_host_bdms:
            return

        LOG.debug("Updating volume usage cache")
        try:
            vol_usages = self.driver.get_all_volume_usage(context,
                                                          compute_host_bdms)
        except NotImplementedError:
            return

        self._update_volume_usage_cache(context, vol_usages)

    @periodic_task.periodic_task(spacing=CONF.sync_power_state_interval,
                                 run_immediately=True)
    def _sync_power_states(self, context):
        """Align power states between the database and the hypervisor.

        To sync power state data we make a DB call to get the number of
        virtual machines known by the hypervisor and if the number matches the
        number of virtual machines known by the database, we proceed in a lazy
        loop, one database record at a time, checking if the hypervisor has the
        same power state as is in the database.
        """
        db_instances = objects.InstanceList.get_by_host(context, self.host,
                                                        expected_attrs=[],
                                                        use_slave=True)

        try:
            num_vm_instances = self.driver.get_num_instances()
        except exception.VirtDriverNotReady as e:
            # If the virt driver is not ready, like ironic-api not being up
            # yet in the case of ironic, just log it and exit.
            LOG.info('Skipping _sync_power_states periodic task due to: %s', e)
            return

        num_db_instances = len(db_instances)

        if num_vm_instances != num_db_instances:
            LOG.warning("While synchronizing instance power states, found "
                        "%(num_db_instances)s instances in the database "
                        "and %(num_vm_instances)s instances on the "
                        "hypervisor.",
                        {'num_db_instances': num_db_instances,
                         'num_vm_instances': num_vm_instances})

        def _sync(db_instance):
            # NOTE(melwitt): This must be synchronized as we query state from
            #                two separate sources, the driver and the database.
            #                They are set (in stop_instance) and read, in sync.
            @utils.synchronized(db_instance.uuid)
            def query_driver_power_state_and_sync():
                self._query_driver_power_state_and_sync(context, db_instance)

            try:
                query_driver_power_state_and_sync()
            except Exception:
                LOG.exception("Periodic sync_power_state task had an "
                              "error while processing an instance.",
                              instance=db_instance)

            self._syncs_in_progress.pop(db_instance.uuid)

        for db_instance in db_instances:
            # process syncs asynchronously - don't want instance locking to
            # block entire periodic task thread
            uuid = db_instance.uuid
            if uuid in self._syncs_in_progress:
                LOG.debug('Sync already in progress for %s', uuid)
            else:
                LOG.debug('Triggering sync for uuid %s', uuid)
                self._syncs_in_progress[uuid] = True
                self._sync_power_pool.spawn_n(_sync, db_instance)

    def _query_driver_power_state_and_sync(self, context, db_instance):
        if db_instance.task_state is not None:
            LOG.info("During sync_power_state the instance has a "
                     "pending task (%(task)s). Skip.",
                     {'task': db_instance.task_state}, instance=db_instance)
            return
        # No pending tasks. Now try to figure out the real vm_power_state.
        try:
            vm_instance = self.driver.get_info(db_instance)
            vm_power_state = vm_instance.state
        except exception.InstanceNotFound:
            vm_power_state = power_state.NOSTATE
        # Note(maoy): the above get_info call might take a long time,
        # for example, because of a broken libvirt driver.
        try:
            self._sync_instance_power_state(context,
                                            db_instance,
                                            vm_power_state,
                                            use_slave=True)
        except exception.InstanceNotFound:
            # NOTE(hanlind): If the instance gets deleted during sync,
            # silently ignore.
            pass

    def _stop_unexpected_shutdown_instance(self, context, vm_state,
                                           db_instance, orig_db_power_state):
        # this is an exceptional case; make sure our data is up
        # to date before slamming through a power off
        vm_instance = self.driver.get_info(db_instance,
                                           use_cache=False)
        vm_power_state = vm_instance.state

        # if it still looks off, go ahead and call stop()
        if vm_power_state in (power_state.SHUTDOWN,
                              power_state.CRASHED):

            LOG.warning("Instance shutdown by itself. Calling the "
                        "stop API. Current vm_state: %(vm_state)s, "
                        "current task_state: %(task_state)s, "
                        "original DB power_state: %(db_power_state)s, "
                        "current VM power_state: %(vm_power_state)s",
                        {'vm_state': vm_state,
                         'task_state': db_instance.task_state,
                         'db_power_state': orig_db_power_state,
                         'vm_power_state': vm_power_state},
                        instance=db_instance)
            try:
                # Note(maoy): here we call the API instead of
                # brutally updating the vm_state in the database
                # to allow all the hooks and checks to be performed.
                if db_instance.shutdown_terminate:
                    self.compute_api.delete(context, db_instance)
                else:
                    self.compute_api.stop(context, db_instance)
            except Exception:
                # Note(maoy): there is no need to propagate the error
                # because the same power_state will be retrieved next
                # time and retried.
                # For example, there might be another task scheduled.
                LOG.exception("error during stop() in sync_power_state.",
                              instance=db_instance)

    def _sync_instance_power_state(self, context, db_instance, vm_power_state,
                                   use_slave=False):
        """Align instance power state between the database and hypervisor.

        If the instance is not found on the hypervisor, but is in the database,
        then a stop() API will be called on the instance.
        """

        # We re-query the DB to get the latest instance info to minimize
        # (not eliminate) race condition.
        db_instance.refresh(use_slave=use_slave)
        db_power_state = db_instance.power_state
        vm_state = db_instance.vm_state

        if self.host != db_instance.host:
            # on the sending end of nova-compute _sync_power_state
            # may have yielded to the greenthread performing a live
            # migration; this in turn has changed the resident-host
            # for the VM; However, the instance is still active, it
            # is just in the process of migrating to another host.
            # This implies that the compute source must relinquish
            # control to the compute destination.
            LOG.info("During the sync_power process the "
                     "instance has moved from "
                     "host %(src)s to host %(dst)s",
                     {'src': db_instance.host,
                      'dst': self.host},
                     instance=db_instance)
            return
        elif db_instance.task_state is not None:
            # on the receiving end of nova-compute, it could happen
            # that the DB instance already report the new resident
            # but the actual VM has not showed up on the hypervisor
            # yet. In this case, let's allow the loop to continue
            # and run the state sync in a later round
            LOG.info("During sync_power_state the instance has a "
                     "pending task (%(task)s). Skip.",
                     {'task': db_instance.task_state},
                     instance=db_instance)
            return

        orig_db_power_state = db_power_state
        if vm_power_state != db_power_state:
            LOG.info('During _sync_instance_power_state the DB '
                     'power_state (%(db_power_state)s) does not match '
                     'the vm_power_state from the hypervisor '
                     '(%(vm_power_state)s). Updating power_state in the '
                     'DB to match the hypervisor.',
                     {'db_power_state': db_power_state,
                      'vm_power_state': vm_power_state},
                     instance=db_instance)
            # power_state is always updated from hypervisor to db
            db_instance.power_state = vm_power_state
            db_instance.save()
            db_power_state = vm_power_state

        # Note(maoy): Now resolve the discrepancy between vm_state and
        # vm_power_state. We go through all possible vm_states.
        if vm_state in (vm_states.BUILDING,
                        vm_states.RESCUED,
                        vm_states.RESIZED,
                        vm_states.SUSPENDED,
                        vm_states.ERROR):
            # TODO(maoy): we ignore these vm_state for now.
            pass
        elif vm_state == vm_states.ACTIVE:
            # The only rational power state should be RUNNING
            if vm_power_state in (power_state.SHUTDOWN,
                                  power_state.CRASHED):
                self._stop_unexpected_shutdown_instance(
                    context, vm_state, db_instance, orig_db_power_state)
            elif vm_power_state == power_state.SUSPENDED:
                LOG.warning("Instance is suspended unexpectedly. Calling "
                            "the stop API.", instance=db_instance)
                try:
                    self.compute_api.stop(context, db_instance)
                except Exception:
                    LOG.exception("error during stop() in sync_power_state.",
                                  instance=db_instance)
            elif vm_power_state == power_state.PAUSED:
                # Note(maoy): a VM may get into the paused state not only
                # because the user request via API calls, but also
                # due to (temporary) external instrumentations.
                # Before the virt layer can reliably report the reason,
                # we simply ignore the state discrepancy. In many cases,
                # the VM state will go back to running after the external
                # instrumentation is done. See bug 1097806 for details.
                LOG.warning("Instance is paused unexpectedly. Ignore.",
                            instance=db_instance)
            elif vm_power_state == power_state.NOSTATE:
                # Occasionally, depending on the status of the hypervisor,
                # which could be restarting for example, an instance may
                # not be found.  Therefore just log the condition.
                LOG.warning("Instance is unexpectedly not found. Ignore.",
                            instance=db_instance)
        elif vm_state == vm_states.STOPPED:
            if vm_power_state not in (power_state.NOSTATE,
                                      power_state.SHUTDOWN,
                                      power_state.CRASHED):
                LOG.warning("Instance is not stopped. Calling "
                            "the stop API. Current vm_state: %(vm_state)s,"
                            " current task_state: %(task_state)s, "
                            "original DB power_state: %(db_power_state)s, "
                            "current VM power_state: %(vm_power_state)s",
                            {'vm_state': vm_state,
                             'task_state': db_instance.task_state,
                             'db_power_state': orig_db_power_state,
                             'vm_power_state': vm_power_state},
                            instance=db_instance)
                try:
                    # NOTE(russellb) Force the stop, because normally the
                    # compute API would not allow an attempt to stop a stopped
                    # instance.
                    self.compute_api.force_stop(context, db_instance)
                except Exception:
                    LOG.exception("error during stop() in sync_power_state.",
                                  instance=db_instance)
        elif vm_state == vm_states.PAUSED:
            if vm_power_state in (power_state.SHUTDOWN,
                                  power_state.CRASHED):
                LOG.warning("Paused instance shutdown by itself. Calling "
                            "the stop API.", instance=db_instance)
                try:
                    self.compute_api.force_stop(context, db_instance)
                except Exception:
                    LOG.exception("error during stop() in sync_power_state.",
                                  instance=db_instance)
        elif vm_state in (vm_states.SOFT_DELETED,
                          vm_states.DELETED):
            if vm_power_state not in (power_state.NOSTATE,
                                      power_state.SHUTDOWN):
                # Note(maoy): this should be taken care of periodically in
                # _cleanup_running_deleted_instances().
                LOG.warning("Instance is not (soft-)deleted.",
                            instance=db_instance)

    @periodic_task.periodic_task
    def _reclaim_queued_deletes(self, context):
        """Reclaim instances that are queued for deletion."""
        interval = CONF.reclaim_instance_interval
        if interval <= 0:
            LOG.debug("CONF.reclaim_instance_interval <= 0, skipping...")
            return

        filters = {'vm_state': vm_states.SOFT_DELETED,
                   'task_state': None,
                   'host': self.host}
        instances = objects.InstanceList.get_by_filters(
            context, filters,
            expected_attrs=objects.instance.INSTANCE_DEFAULT_FIELDS,
            use_slave=True)
        for instance in instances:
            if self._deleted_old_enough(instance, interval):
                bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                        context, instance.uuid)
                LOG.info('Reclaiming deleted instance', instance=instance)
                try:
                    self._delete_instance(context, instance, bdms)
                except Exception as e:
                    LOG.warning("Periodic reclaim failed to delete "
                                "instance: %s",
                                e, instance=instance)

    def _get_nodename(self, instance, refresh=False):
        """Helper method to get the name of the first available node
        on this host. This method should not be used with any operations
        on ironic instances since it does not handle multiple nodes.
        """
        node = self.driver.get_available_nodes(refresh=refresh)[0]
        LOG.debug("No node specified, defaulting to %s", node,
                  instance=instance)
        return node

    def _update_available_resource_for_node(self, context, nodename,
                                            startup=False):

        try:
            self.rt.update_available_resource(context, nodename,
                                              startup=startup)
        except exception.ComputeHostNotFound:
            LOG.warning("Compute node '%s' not found in "
                        "update_available_resource.", nodename)
        except exception.ReshapeFailed:
            # We're only supposed to get here on startup, if a reshape was
            # needed, was attempted, and failed. We want to kill the service.
            with excutils.save_and_reraise_exception():
                LOG.critical("Resource provider data migration failed "
                             "fatally during startup for node %s.", nodename)
        except exception.ReshapeNeeded:
            # This exception should only find its way here if the virt driver's
            # update_provider_tree raised it incorrectly: either
            # a) After the resource tracker already caught it once and
            # reinvoked update_provider_tree with allocations. At this point
            # the driver is just supposed to *do* the reshape, so if it raises
            # ReshapeNeeded, it's a bug, and we want to kill the compute
            # service.
            # b) On periodic rather than startup (we only allow reshapes to
            # happen on startup). In this case we'll just make the logs red and
            # go again at the next periodic interval, where the same thing may
            # or may not happen again. Depending on the previous and intended
            # shape of the providers/inventories, this may not actually cause
            # any immediately visible symptoms (in terms of scheduling, etc.)
            # If this becomes a problem, we may wish to make it pop immediately
            # (e.g. disable the service).
            with excutils.save_and_reraise_exception():
                LOG.exception("ReshapeNeeded exception is unexpected here!")
        except exception.PlacementPciException:
            # If we are at startup and the Placement PCI inventory handling
            # failed then probably there is a configuration error. Propagate
            # the error up to kill the service.
            if startup:
                raise
            # If we are not at startup then we can assume that the
            # configuration was correct at startup so the error is probably
            # transient. Anyhow we cannot kill the service any more so just
            # log the error and continue.
            LOG.exception(
                "Error updating PCI resources for node %(node)s.",
                {'node': nodename})
        except Exception:
            LOG.exception("Error updating resources for node %(node)s.",
                          {'node': nodename})

    @periodic_task.periodic_task(spacing=CONF.update_resources_interval)
    def update_available_resource(self, context, startup=False):
        """See driver.get_available_resource()

        Periodic process that keeps that the compute host's understanding of
        resource availability and usage in sync with the underlying hypervisor.

        :param context: security context
        :param startup: True if this is being called when the nova-compute
            service is starting, False otherwise.
        """
        try:
            nodenames = set(self.driver.get_available_nodes())
        except exception.VirtDriverNotReady:
            LOG.warning("Virt driver is not ready.")
            return

        compute_nodes_in_db = self._get_compute_nodes_in_db(context,
                                                            nodenames,
                                                            use_slave=True,
                                                            startup=startup)

        self.rt.clean_compute_node_cache(compute_nodes_in_db)

        # Delete orphan compute node not reported by driver but still in db
        for cn in compute_nodes_in_db:
            if cn.hypervisor_hostname not in nodenames:
                LOG.info("Deleting orphan compute node %(id)s "
                         "hypervisor host is %(hh)s, "
                         "nodes are %(nodes)s",
                         {'id': cn.id, 'hh': cn.hypervisor_hostname,
                          'nodes': nodenames})
                try:
                    cn.destroy()
                except exception.ObjectActionError:
                    # NOTE(mgoddard): it's possible that another compute
                    # service took ownership of this compute node since we
                    # queried it due to a rebalance, and this will cause the
                    # deletion to fail. Ignore the error in that case.
                    LOG.info("Ignoring failure to delete orphan compute node "
                             "%(id)s on hypervisor host %(hh)s due to "
                             "possible node rebalance",
                             {'id': cn.id, 'hh': cn.hypervisor_hostname})
                    self.rt.remove_node(cn.hypervisor_hostname)
                    self.reportclient.invalidate_resource_provider(cn.uuid)
                else:
                    self.rt.remove_node(cn.hypervisor_hostname)
                    # Delete the corresponding resource provider in placement,
                    # along with any associated allocations.
                    try:
                        self.reportclient.delete_resource_provider(
                            context, cn, cascade=True)
                    except keystone_exception.ClientException as e:
                        LOG.error(
                            "Failed to delete compute node resource provider "
                            "for compute node %s: %s", cn.uuid, str(e))

        for nodename in nodenames:
            self._update_available_resource_for_node(context, nodename,
                                                     startup=startup)

    def _get_compute_nodes_in_db(self, context, nodenames, use_slave=False,
                                 startup=False):
        try:
            return objects.ComputeNodeList.get_all_by_host(context, self.host,
                                                           use_slave=use_slave)
        except exception.NotFound:
            # If the driver is not reporting any nodenames we should not
            # expect there to be compute nodes so we just return in that case.
            # For example, this could be an ironic compute and it is not
            # managing any nodes yet.
            if nodenames:
                if startup:
                    LOG.warning(
                        "No compute node record found for host %s. If this is "
                        "the first time this service is starting on this "
                        "host, then you can ignore this warning.", self.host)
                else:
                    LOG.error("No compute node record for host %s", self.host)
            return []

    @periodic_task.periodic_task(
        spacing=CONF.running_deleted_instance_poll_interval,
        run_immediately=True)
    def _cleanup_running_deleted_instances(self, context):
        """Cleanup any instances which are erroneously still running after
        having been deleted.

        Valid actions to take are:

            1. noop - do nothing
            2. log - log which instances are erroneously running
            3. reap - shutdown and cleanup any erroneously running instances
            4. shutdown - power off *and disable* any erroneously running
                          instances

        The use-case for this cleanup task is: for various reasons, it may be
        possible for the database to show an instance as deleted but for that
        instance to still be running on a host machine (see bug
        https://bugs.launchpad.net/nova/+bug/911366).

        This cleanup task is a cross-hypervisor utility for finding these
        zombied instances and either logging the discrepancy (likely what you
        should do in production), or automatically reaping the instances (more
        appropriate for dev environments).
        """
        action = CONF.running_deleted_instance_action

        if action == "noop":
            return

        # NOTE(sirp): admin contexts don't ordinarily return deleted records
        with utils.temporary_mutation(context, read_deleted="yes"):

            try:
                instances = self._running_deleted_instances(context)
            except exception.VirtDriverNotReady:
                # Since this task runs immediately on startup, if the
                # hypervisor is not yet ready handle it gracefully.
                LOG.debug('Unable to check for running deleted instances '
                          'at this time since the hypervisor is not ready.')
                return

            for instance in instances:
                if action == "log":
                    LOG.warning("Detected instance with name label "
                                "'%s' which is marked as "
                                "DELETED but still present on host.",
                                instance.name, instance=instance)

                elif action == 'shutdown':
                    LOG.info("Powering off instance with name label "
                             "'%s' which is marked as "
                             "DELETED but still present on host.",
                             instance.name, instance=instance)
                    try:
                        self.driver.power_off(instance)
                    except Exception:
                        LOG.warning("Failed to power off instance",
                                    instance=instance, exc_info=True)

                elif action == 'reap':
                    LOG.info("Destroying instance with name label "
                             "'%s' which is marked as "
                             "DELETED but still present on host.",
                             instance.name, instance=instance)
                    bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                        context, instance.uuid, use_slave=True)
                    self.instance_events.clear_events_for_instance(instance)
                    try:
                        self._shutdown_instance(context, instance, bdms,
                                                notify=False)
                        self._cleanup_volumes(context, instance, bdms,
                                              detach=False)
                    except Exception as e:
                        LOG.warning("Periodic cleanup failed to delete "
                                    "instance: %s",
                                    e, instance=instance)
                else:
                    raise Exception(_("Unrecognized value '%s'"
                                      " for CONF.running_deleted_"
                                      "instance_action") % action)

    def _running_deleted_instances(self, context):
        """Returns a list of instances nova thinks is deleted,
        but the hypervisor thinks is still running.
        """
        timeout = CONF.running_deleted_instance_timeout
        filters = {'deleted': True,
                   'soft_deleted': False}
        instances = self._get_instances_on_driver(context, filters)
        return [i for i in instances if self._deleted_old_enough(i, timeout)]

    def _deleted_old_enough(self, instance, timeout):
        deleted_at = instance.deleted_at
        if deleted_at:
            deleted_at = deleted_at.replace(tzinfo=None)
        return (not deleted_at or timeutils.is_older_than(deleted_at, timeout))

    @contextlib.contextmanager
    def _error_out_instance_on_exception(self, context, instance,
                                         instance_state=vm_states.ACTIVE):
        """Context manager to set instance.vm_state after some operation raises

        Used to handle NotImplementedError and InstanceFaultRollback errors
        and reset the instance vm_state and task_state. The vm_state is set
        to the $instance_state parameter and task_state is set to None.
        For all other types of exceptions, the vm_state is set to ERROR and
        the task_state is left unchanged (although most callers will have the
        @reverts_task_state decorator which will set the task_state to None).

        Re-raises the original exception *except* in the case of
        InstanceFaultRollback in which case the wrapped `inner_exception` is
        re-raised.

        :param context: The nova auth request context for the operation.
        :param instance: The instance to update. The vm_state will be set by
            this context manager when an exception is raised.
        :param instance_state: For NotImplementedError and
            InstanceFaultRollback this is the vm_state to set the instance to
            when handling one of those types of exceptions. By default the
            instance will be set to ACTIVE, but the caller should control this
            in case there have been no changes to the running state of the
            instance. For example, resizing a stopped server where prep_resize
            fails early and does not change the power state of the guest should
            not set the instance status to ACTIVE but remain STOPPED.
            This parameter is ignored for all other types of exceptions and the
            instance vm_state is set to ERROR.
        """
        # NOTE(mriedem): Why doesn't this method just save off the
        # original instance.vm_state here rather than use a parameter? Or use
        # instance_state=None as an override but default to the current
        # vm_state when rolling back.
        instance_uuid = instance.uuid
        try:
            yield
        except (NotImplementedError, exception.InstanceFaultRollback) as error:
            # Use reraise=False to determine if we want to raise the original
            # exception or something else.
            with excutils.save_and_reraise_exception(reraise=False) as ctxt:
                LOG.info("Setting instance back to %(state)s after: %(error)s",
                         {'state': instance_state, 'error': error},
                         instance_uuid=instance_uuid)
                self._instance_update(context, instance,
                                      vm_state=instance_state,
                                      task_state=None)
                if isinstance(error, exception.InstanceFaultRollback):
                    # Raise the wrapped exception.
                    raise error.inner_exception
                # Else re-raise the NotImplementedError.
                ctxt.reraise = True
        except Exception:
            LOG.exception('Setting instance vm_state to ERROR',
                          instance_uuid=instance_uuid)
            with excutils.save_and_reraise_exception():
                # NOTE(mriedem): Why don't we pass clean_task_state=True here?
                self._set_instance_obj_error_state(instance)

    def _process_instance_event(self, instance, event):
        _event = self.instance_events.pop_instance_event(instance, event)
        if _event:
            LOG.debug('Processing event %(event)s',
                      {'event': event.key}, instance=instance)
            _event.send(event)
        else:
            # If it's a network-vif-unplugged event and the instance is being
            # deleted or live migrated then we don't need to make this a
            # warning as it's expected. There are other expected things which
            # could trigger this event like detaching an interface, but we
            # don't have a task state for that.
            # TODO(mriedem): We have other move operations and things like
            # hard reboot (probably rebuild as well) which trigger this event
            # but nothing listens for network-vif-unplugged. We should either
            # handle those other known cases or consider just not logging a
            # warning if we get this event and the instance is undergoing some
            # task state transition.
            if (event.name == 'network-vif-unplugged' and
                    instance.task_state in (
                        task_states.DELETING, task_states.MIGRATING)):
                LOG.debug('Received event %s for instance with task_state %s.',
                          event.key, instance.task_state, instance=instance)
            else:
                LOG.warning('Received unexpected event %(event)s for '
                            'instance with vm_state %(vm_state)s and '
                            'task_state %(task_state)s.',
                            {'event': event.key,
                             'vm_state': instance.vm_state,
                             'task_state': instance.task_state},
                            instance=instance)

    def _process_instance_vif_deleted_event(self, context, instance,
                                            deleted_vif_id):
        # If an attached port is deleted by neutron, it needs to
        # be detached from the instance.
        # And info cache needs to be updated.
        network_info = instance.info_cache.network_info
        for index, vif in enumerate(network_info):
            if vif['id'] == deleted_vif_id:
                LOG.info('Neutron deleted interface %(intf)s; '
                         'detaching it from the instance and '
                         'deleting it from the info cache',
                         {'intf': vif['id']},
                         instance=instance)
                profile = vif.get('profile', {}) or {}  # profile can be None
                rps = profile.get('allocation')
                if rps:
                    if isinstance(rps, dict):
                        # if extended resource request extension is enabled
                        # then we have a dict of providers, flatten it for the
                        # log.
                        rps = ','.join(rps.values())
                    LOG.error(
                        'The bound port %(port_id)s is deleted in Neutron but '
                        'the resource allocation on the resource providers '
                        '%(rp_uuid)s are leaked until the server '
                        '%(server_uuid)s is deleted.',
                        {'port_id': vif['id'],
                         'rp_uuid': rps,
                         'server_uuid': instance.uuid})

                del network_info[index]
                neutron.update_instance_cache_with_nw_info(
                    self.network_api, context, instance, nw_info=network_info)
                try:
                    self.driver.detach_interface(context, instance, vif)
                except NotImplementedError:
                    # Not all virt drivers support attach/detach of interfaces
                    # yet (like Ironic), so just ignore this.
                    pass
                except exception.NovaException as ex:
                    # If the instance was deleted before the interface was
                    # detached, just log it at debug.
                    log_level = (logging.DEBUG
                                 if isinstance(ex, exception.InstanceNotFound)
                                 else logging.WARNING)
                    LOG.log(log_level,
                            "Detach interface failed, "
                            "port_id=%(port_id)s, reason: %(msg)s",
                            {'port_id': deleted_vif_id, 'msg': ex},
                            instance=instance)
                break

    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def extend_volume(self, context, instance, extended_volume_id):

        # If an attached volume is extended by cinder, it needs to
        # be extended by virt driver so host can detect its new size.
        # And bdm needs to be updated.
        LOG.debug('Handling volume-extended event for volume %(vol)s',
                  {'vol': extended_volume_id}, instance=instance)

        try:
            bdm = objects.BlockDeviceMapping.get_by_volume_and_instance(
                   context, extended_volume_id, instance.uuid)
        except exception.NotFound:
            LOG.warning('Extend volume failed, '
                        'volume %(vol)s is not attached to instance.',
                        {'vol': extended_volume_id},
                        instance=instance)
            return

        LOG.info('Cinder extended volume %(vol)s; '
                 'extending it to detect new size',
                 {'vol': extended_volume_id},
                 instance=instance)
        volume = self.volume_api.get(context, bdm.volume_id)

        if bdm.connection_info is None:
            LOG.warning('Extend volume failed, '
                        'attached volume %(vol)s has no connection_info',
                        {'vol': extended_volume_id},
                        instance=instance)
            return

        connection_info = jsonutils.loads(bdm.connection_info)
        bdm.volume_size = volume['size']
        bdm.save()

        if not self.driver.capabilities.get('supports_extend_volume', False):
            raise exception.ExtendVolumeNotSupported()

        try:
            self.driver.extend_volume(context, connection_info, instance,
                                      bdm.volume_size * units.Gi)
        except Exception as ex:
            LOG.warning('Extend volume failed, '
                        'volume_id=%(volume_id)s, reason: %(msg)s',
                        {'volume_id': extended_volume_id, 'msg': ex},
                        instance=instance)
            raise

    @staticmethod
    def _is_state_valid_for_power_update_event(instance, target_power_state):
        """Check if the current state of the instance allows it to be
        a candidate for the power-update event.

        :param instance: The nova instance object.
        :param target_power_state: The desired target power state; this should
                                   either be "POWER_ON" or "POWER_OFF".
        :returns Boolean: True if the instance can be subjected to the
                          power-update event.
        """
        if ((target_power_state == external_event_obj.POWER_ON and
                instance.task_state is None and
                instance.vm_state == vm_states.STOPPED and
                instance.power_state == power_state.SHUTDOWN) or
            (target_power_state == external_event_obj.POWER_OFF and
                instance.task_state is None and
                instance.vm_state == vm_states.ACTIVE and
                instance.power_state == power_state.RUNNING)):
            return True
        return False

    @wrap_exception()
    @reverts_task_state
    @wrap_instance_event(prefix='compute')
    @wrap_instance_fault
    def power_update(self, context, instance, target_power_state):
        """Power update of an instance prompted by an external event.
        :param context: The API request context.
        :param instance: The nova instance object.
        :param target_power_state: The desired target power state;
                                   this should either be "POWER_ON" or
                                   "POWER_OFF".
        """

        @utils.synchronized(instance.uuid)
        def do_power_update():
            LOG.debug('Handling power-update event with target_power_state %s '
                      'for instance', target_power_state, instance=instance)
            if not self._is_state_valid_for_power_update_event(
                    instance, target_power_state):
                pow_state = fields.InstancePowerState.from_index(
                    instance.power_state)
                LOG.info('The power-update %(tag)s event for instance '
                         '%(uuid)s is a no-op since the instance is in '
                         'vm_state %(vm_state)s, task_state '
                         '%(task_state)s and power_state '
                         '%(power_state)s.',
                         {'tag': target_power_state, 'uuid': instance.uuid,
                         'vm_state': instance.vm_state,
                         'task_state': instance.task_state,
                         'power_state': pow_state})
                return
            LOG.debug("Trying to %s instance",
                      target_power_state, instance=instance)
            if target_power_state == external_event_obj.POWER_ON:
                action = fields.NotificationAction.POWER_ON
                notification_name = "power_on."
                instance.task_state = task_states.POWERING_ON
            else:
                # It's POWER_OFF
                action = fields.NotificationAction.POWER_OFF
                notification_name = "power_off."
                instance.task_state = task_states.POWERING_OFF
                instance.progress = 0

            try:
                # Note that the task_state is set here rather than the API
                # because this is a best effort operation and deferring
                # updating the task_state until we get to the compute service
                # avoids error handling in the API and needing to account for
                # older compute services during rolling upgrades from Stein.
                # If we lose a race, UnexpectedTaskStateError is handled
                # below.
                instance.save(expected_task_state=[None])
                self._notify_about_instance_usage(context, instance,
                                                  notification_name + "start")
                compute_utils.notify_about_instance_action(context, instance,
                    self.host, action=action,
                    phase=fields.NotificationPhase.START)
                # UnexpectedTaskStateError raised from the driver will be
                # handled below and not result in a fault, error notification
                # or failure of the instance action. Other driver errors like
                # NotImplementedError will be record a fault, send an error
                # notification and mark the instance action as failed.
                self.driver.power_update_event(instance, target_power_state)
                self._notify_about_instance_usage(context, instance,
                                                  notification_name + "end")
                compute_utils.notify_about_instance_action(context, instance,
                    self.host, action=action,
                    phase=fields.NotificationPhase.END)
            except exception.UnexpectedTaskStateError as e:
                # Handling the power-update event is best effort and if we lost
                # a race with some other action happening to the instance we
                # just log it and return rather than fail the action.
                LOG.info("The power-update event was possibly preempted: %s ",
                         e.format_message(), instance=instance)
                return
        do_power_update()

    @wrap_exception()
    def external_instance_event(self, context, instances, events):
        # NOTE(danms): Some event types are handled by the manager, such
        # as when we're asked to update the instance's info_cache. If it's
        # not one of those, look for some thread(s) waiting for the event and
        # unblock them if so.
        for event in events:
            instance = [inst for inst in instances
                        if inst.uuid == event.instance_uuid][0]
            LOG.debug('Received event %(event)s',
                      {'event': event.key},
                      instance=instance)
            if event.name == 'network-changed':
                try:
                    LOG.debug('Refreshing instance network info cache due to '
                              'event %s.', event.key, instance=instance)
                    self.network_api.get_instance_nw_info(
                        context, instance, refresh_vif_id=event.tag)
                except exception.NotFound as e:
                    LOG.info('Failed to process external instance event '
                             '%(event)s due to: %(error)s',
                             {'event': event.key, 'error': str(e)},
                             instance=instance)
            elif event.name == 'network-vif-deleted':
                try:
                    self._process_instance_vif_deleted_event(context,
                                                             instance,
                                                             event.tag)
                except exception.NotFound as e:
                    LOG.info('Failed to process external instance event '
                             '%(event)s due to: %(error)s',
                             {'event': event.key, 'error': str(e)},
                             instance=instance)
            elif event.name == 'volume-extended':
                self.extend_volume(context, instance, event.tag)
            elif event.name == 'power-update':
                self.power_update(context, instance, event.tag)
            else:
                self._process_instance_event(instance, event)

    @periodic_task.periodic_task(spacing=CONF.image_cache.manager_interval,
                                 external_process_ok=True)
    def _run_image_cache_manager_pass(self, context):
        """Run a single pass of the image cache manager."""

        if not self.driver.capabilities.get("has_imagecache", False):
            return

        # Determine what other nodes use this storage
        storage_users.register_storage_use(CONF.instances_path, CONF.host)
        nodes = storage_users.get_storage_users(CONF.instances_path)

        # Filter all_instances to only include those nodes which share this
        # storage path.
        # TODO(mikal): this should be further refactored so that the cache
        # cleanup code doesn't know what those instances are, just a remote
        # count, and then this logic should be pushed up the stack.
        filters = {'deleted': False,
                   'soft_deleted': True,
                   'host': nodes}
        filtered_instances = objects.InstanceList.get_by_filters(context,
                                 filters, expected_attrs=[], use_slave=True)

        self.driver.manage_image_cache(context, filtered_instances)

    def cache_images(self, context, image_ids):
        """Ask the virt driver to pre-cache a set of base images.

        :param context: The RequestContext
        :param image_ids: The image IDs to be cached
        :return: A dict, keyed by image-id where the values are one of:
                 'cached' if the image was downloaded,
                 'existing' if the image was already in the cache,
                 'unsupported' if the virt driver does not support caching,
                 'error' if the virt driver raised an exception.
        """

        results = {}

        LOG.info('Caching %i image(s) by request', len(image_ids))
        for image_id in image_ids:
            try:
                cached = self.driver.cache_image(context, image_id)
                if cached:
                    results[image_id] = 'cached'
                else:
                    results[image_id] = 'existing'
            except NotImplementedError:
                LOG.warning('Virt driver does not support image pre-caching;'
                            ' ignoring request')
                # NOTE(danms): Yes, technically we could short-circuit here to
                # avoid trying the rest of the images, but it's very cheap to
                # just keep hitting the NotImplementedError to keep the logic
                # clean.
                results[image_id] = 'unsupported'
            except Exception as e:
                results[image_id] = 'error'
                LOG.error('Failed to cache image %(image_id)s: %(err)s',
                          {'image_id': image_id,
                           'err': e})

        return results

    @periodic_task.periodic_task(spacing=CONF.instance_delete_interval)
    def _run_pending_deletes(self, context):
        """Retry any pending instance file deletes."""
        LOG.debug('Cleaning up deleted instances')
        filters = {'deleted': True,
                   'soft_deleted': False,
                   'host': CONF.host,
                   'cleaned': False}
        attrs = ['system_metadata']
        with utils.temporary_mutation(context, read_deleted='yes'):
            instances = objects.InstanceList.get_by_filters(
                context, filters, expected_attrs=attrs, use_slave=True)
        LOG.debug('There are %d instances to clean', len(instances))

        for instance in instances:
            attempts = int(instance.system_metadata.get('clean_attempts', '0'))
            LOG.debug('Instance has had %(attempts)s of %(max)s '
                      'cleanup attempts',
                      {'attempts': attempts,
                       'max': CONF.maximum_instance_delete_attempts},
                      instance=instance)
            if attempts < CONF.maximum_instance_delete_attempts:
                success = self.driver.delete_instance_files(instance)

                instance.system_metadata['clean_attempts'] = str(attempts + 1)
                if success:
                    instance.cleaned = True
                with utils.temporary_mutation(context, read_deleted='yes'):
                    instance.save()

    @periodic_task.periodic_task(spacing=CONF.instance_delete_interval)
    def _cleanup_incomplete_migrations(self, context):
        """Cleanup on failed resize/revert-resize operation and
        failed rollback live migration operation.

        During resize/revert-resize operation, or after a failed rollback
        live migration operation, if that instance gets deleted then instance
        files might remain either on source or destination compute node and
        other specific resources might not be cleaned up because of the race
        condition.
        """
        LOG.debug('Cleaning up deleted instances with incomplete migration ')
        migration_filters = {'host': CONF.host,
                             'status': 'error'}
        migrations = objects.MigrationList.get_by_filters(context,
                                                          migration_filters)

        if not migrations:
            return

        inst_uuid_from_migrations = set([migration.instance_uuid for migration
                                         in migrations])

        inst_filters = {'deleted': True, 'soft_deleted': False,
                        'uuid': inst_uuid_from_migrations}
        attrs = ['info_cache', 'security_groups', 'system_metadata']
        with utils.temporary_mutation(context, read_deleted='yes'):
            instances = objects.InstanceList.get_by_filters(
                context, inst_filters, expected_attrs=attrs, use_slave=True)

        for instance in instances:
            if instance.host == CONF.host:
                continue
            for migration in migrations:
                if instance.uuid != migration.instance_uuid:
                    continue
                self.driver.delete_instance_files(instance)
                # we are not sure whether the migration_context is applied
                # during incompleted migrating, we need to apply/revert
                # migration_context to get instance object content matching
                # current host.
                revert = (True if migration.source_compute == CONF.host
                          else False)
                with instance.mutated_migration_context(revert=revert):
                    self.driver.cleanup_lingering_instance_resources(instance)

                try:
                    migration.status = 'failed'
                    migration.save()
                except exception.MigrationNotFound:
                    LOG.warning("Migration %s is not found.",
                                migration.id,
                                instance=instance)
                break

    @messaging.expected_exceptions(exception.InstanceQuiesceNotSupported,
                                   exception.QemuGuestAgentNotEnabled,
                                   exception.NovaException,
                                   NotImplementedError)
    @wrap_exception()
    def quiesce_instance(self, context, instance):
        """Quiesce an instance on this host."""
        context = context.elevated()
        image_meta = objects.ImageMeta.from_instance(instance)
        self.driver.quiesce(context, instance, image_meta)

    def _wait_for_snapshots_completion(self, context, mapping):
        for mapping_dict in mapping:
            if mapping_dict.get('source_type') == 'snapshot':

                def _wait_snapshot():
                    snapshot = self.volume_api.get_snapshot(
                        context, mapping_dict['snapshot_id'])
                    if snapshot.get('status') != 'creating':
                        raise loopingcall.LoopingCallDone()

                timer = loopingcall.FixedIntervalLoopingCall(_wait_snapshot)
                timer.start(interval=0.5).wait()

    @messaging.expected_exceptions(exception.InstanceQuiesceNotSupported,
                                   exception.QemuGuestAgentNotEnabled,
                                   exception.NovaException,
                                   NotImplementedError)
    @wrap_exception()
    def unquiesce_instance(self, context, instance, mapping=None):
        """Unquiesce an instance on this host.

        If snapshots' image mapping is provided, it waits until snapshots are
        completed before unqueiscing.
        """
        context = context.elevated()
        if mapping:
            try:
                self._wait_for_snapshots_completion(context, mapping)
            except Exception as error:
                LOG.exception("Exception while waiting completion of "
                              "volume snapshots: %s",
                              error, instance=instance)
        image_meta = objects.ImageMeta.from_instance(instance)
        self.driver.unquiesce(context, instance, image_meta)

    @periodic_task.periodic_task(spacing=CONF.instance_delete_interval)
    def _cleanup_expired_console_auth_tokens(self, context):
        """Remove all expired console auth tokens.

        Console authorization tokens and their connection data are stored
        in the database when a user asks for a console connection to an
        instance. After a time they expire. We periodically remove any expired
        tokens from the database.
        """
        objects.ConsoleAuthToken.clean_expired_console_auths(context)

    def _claim_pci_for_instance_vifs(self, ctxt, instance):
        """Claim PCI devices for the instance's VIFs on the compute node

        :param ctxt: Context
        :param instance: Instance object
        :return: <port ID: PciDevice> mapping for the VIFs that yielded a
                PCI claim on the compute node
        """
        pci_req_id_to_port_id = {}
        pci_reqs = []
        port_id_to_pci_dev = {}

        for vif in instance.get_network_info():
            pci_req = pci_req_module.get_instance_pci_request_from_vif(
                ctxt,
                instance,
                vif)
            if pci_req:
                pci_req_id_to_port_id[pci_req.request_id] = vif['id']
                pci_reqs.append(pci_req)

        if pci_reqs:
            # Create PCI requests and claim against PCI resource tracker
            # NOTE(adrianc): We claim against the same requests as on the
            # source node.
            vif_pci_requests = objects.InstancePCIRequests(
                requests=pci_reqs,
                instance_uuid=instance.uuid)

            # if we are called during the live migration with NUMA topology
            # support the PCI claim needs to consider the destination NUMA
            # topology that is then stored in the migration_context
            dest_topo = None
            if instance.migration_context:
                dest_topo = instance.migration_context.new_numa_topology

            claimed_pci_devices_objs = self.rt.claim_pci_devices(
                ctxt, vif_pci_requests, dest_topo)

            # Update VIFMigrateData profile with the newly claimed PCI
            # device
            for pci_dev in claimed_pci_devices_objs:
                LOG.debug("PCI device: %s Claimed on destination node",
                          pci_dev.address)
                port_id = pci_req_id_to_port_id[pci_dev.request_id]
                port_id_to_pci_dev[port_id] = pci_dev

        return port_id_to_pci_dev

    def _update_migrate_vifs_profile_with_pci(self,
                                              migrate_vifs,
                                              port_id_to_pci_dev):
        """Update migrate vifs profile with the claimed PCI devices

        :param migrate_vifs: list of VIFMigrateData objects
        :param port_id_to_pci_dev: a <port_id: PciDevice> mapping
        :return: None.
        """
        for mig_vif in migrate_vifs:
            port_id = mig_vif.port_id
            if port_id not in port_id_to_pci_dev:
                continue

            pci_dev = port_id_to_pci_dev[port_id]
            profile = copy.deepcopy(mig_vif.source_vif['profile'])
            profile['pci_slot'] = pci_dev.address
            profile['pci_vendor_info'] = ':'.join([pci_dev.vendor_id,
                                                   pci_dev.product_id])
            if profile.get('card_serial_number'):
                # Assume it is there since Nova makes sure that PCI devices
                # tagged as remote-managed have a serial in PCI VPD.
                profile['card_serial_number'] = pci_dev.card_serial_number
            if profile.get('pf_mac_address'):
                profile['pf_mac_address'] = pci_dev.sriov_cap['pf_mac_address']
            if profile.get('vf_num'):
                profile['vf_num'] = pci_dev.sriov_cap['vf_num']

            if pci_dev.mac_address:
                profile['device_mac_address'] = pci_dev.mac_address

            mig_vif.profile = profile
            LOG.debug("Updating migrate VIF profile for port %(port_id)s:"
                      "%(profile)s", {'port_id': port_id,
                                      'profile': profile})


# TODO(sbauza): Remove this proxy class in the X release once we drop the 5.x
# support.
# NOTE(sbauza): This proxy class will support the existing <=5.13 RPC calls
# from any RPC client but will also make sure that the new 6.0 RPC calls will
# be supported.
class _ComputeV5Proxy(object):

    target = messaging.Target(version='5.13')

    def __init__(self, manager):
        self.manager = manager

    def __getattr__(self, name):
        # NOTE(sbauza): Proxying all the other methods but the V5 ones.
        return getattr(self.manager, name)

    # 5.0 support for block_migration argument
    def pre_live_migration(self, context, instance, block_migration, disk,
                           migrate_data):
        return self.manager.pre_live_migration(context, instance, disk,
                                               migrate_data)

    # 5.1 support for legacy request_spec argument
    def prep_resize(self, context, image, instance, instance_type,
                    request_spec, filter_properties, node,
                    clean_shutdown, migration, host_list):
        if not isinstance(request_spec, objects.RequestSpec):
            # Prior to compute RPC API 5.1 conductor would pass a legacy dict
            # version of the request spec to compute and since Stein compute
            # could be sending that back to conductor on reschedule, so if we
            # got a dict convert it to an object.
            # TODO(mriedem): We can drop this compat code when we only support
            # compute RPC API >=6.0.
            request_spec = objects.RequestSpec.from_primitives(
                context, request_spec, filter_properties)
            # We don't have to set the new flavor on the request spec because
            # if we got here it was due to a reschedule from the compute and
            # the request spec would already have the new flavor in it from the
            # else block below.
        self.manager.prep_resize(context, image, instance, instance_type,
                                 request_spec, filter_properties, node,
                                 clean_shutdown, migration, host_list)

    # 5.2 support for optional request_spec argument
    def resize_instance(self, context, instance, image,
                        migration, instance_type, clean_shutdown,
                        request_spec=None):
        self.manager.resize_instance(context, instance, image,
                                    migration, instance_type, clean_shutdown,
                                    request_spec)

    # 5.2 support for optional request_spec argument
    def finish_resize(self, context, disk_info, image, instance,
                      migration, request_spec=None):
        self.manager.finish_resize(context, disk_info, image, instance,
                                   migration, request_spec)

    # 5.2 support for optional request_spec argument
    def revert_resize(self, context, instance, migration, request_spec=None):
        self.manager.revert_resize(context, instance, migration, request_spec)

    # 5.2 support for optional request_spec argument
    def finish_revert_resize(
            self, context, instance, migration, request_spec=None):
        self.manager.finish_revert_resize(context, instance, migration,
                                          request_spec)

    # 5.2 support for optional request_spec argument
    # 5.13 support for optional accel_uuids argument
    def unshelve_instance(self, context, instance, image, filter_properties,
                          node, request_spec=None, accel_uuids=None):
        self.manager.unshelve_instance(context, instance, image,
                                       filter_properties, node, request_spec,
                                       accel_uuids or [])

    # 5.3 support for optional migration and limits arguments
    def check_can_live_migrate_destination(self, ctxt, instance,
                                           block_migration, disk_over_commit,
                                           migration=None, limits=None):
        return self.manager.check_can_live_migrate_destination(
            ctxt, instance, block_migration, disk_over_commit,
            migration, limits)

    # 5.11 support for optional accel_uuids argument
    def build_and_run_instance(self, context, instance, image, request_spec,
                     filter_properties, admin_password=None,
                     injected_files=None, requested_networks=None,
                     security_groups=None, block_device_mapping=None,
                     node=None, limits=None, host_list=None, accel_uuids=None):
        self.manager.build_and_run_instance(
            context, instance, image, request_spec,
            filter_properties, accel_uuids, admin_password,
            injected_files, requested_networks,
            security_groups, block_device_mapping,
            node, limits, host_list)

    # 5.12 support for optional accel_uuids argument
    def rebuild_instance(self, context, instance, orig_image_ref, image_ref,
                         injected_files, new_pass, orig_sys_metadata,
                         bdms, recreate, on_shared_storage,
                         preserve_ephemeral, migration,
                         scheduled_node, limits, request_spec,
                         accel_uuids=None):
        self.manager.rebuild_instance(
            context, instance, orig_image_ref, image_ref,
            injected_files, new_pass, orig_sys_metadata,
            bdms, recreate, on_shared_storage,
            preserve_ephemeral, migration,
            scheduled_node, limits, request_spec,
            accel_uuids, False)

    # 5.13 support for optional accel_uuids argument
    def shelve_instance(self, context, instance, image_id,
                        clean_shutdown, accel_uuids=None):
        self.manager.shelve_instance(context, instance, image_id,
                                     clean_shutdown, accel_uuids)

    # 5.13 support for optional accel_uuids argument
    def shelve_offload_instance(self, context, instance, clean_shutdown,
            accel_uuids=None):
        self.manager.shelve_offload_instance(
            context, instance, clean_shutdown, accel_uuids)

    # 6.0 drop unused request_spec argument
    def prep_snapshot_based_resize_at_dest(
            self, ctxt, instance, flavor, nodename, migration, limits,
            request_spec):
        return self.manager.prep_snapshot_based_resize_at_dest(
            ctxt, instance, flavor, nodename, migration, limits)

    # 6.0 drop unused request_spec argument
    def finish_snapshot_based_resize_at_dest(
            self, ctxt, instance, migration, snapshot_id, request_spec):
        self.manager.finish_snapshot_based_resize_at_dest(
            ctxt, instance, migration, snapshot_id)

    # 6.0 drop unused instance argument
    def check_instance_shared_storage(self, ctxt, instance, data):
        return self.manager.check_instance_shared_storage(ctxt, data)