diff options
Diffstat (limited to 'nova')
55 files changed, 1891 insertions, 179 deletions
diff --git a/nova/api/openstack/compute/rest_api_version_history.rst b/nova/api/openstack/compute/rest_api_version_history.rst index b34510be5c..c7a2777d3a 100644 --- a/nova/api/openstack/compute/rest_api_version_history.rst +++ b/nova/api/openstack/compute/rest_api_version_history.rst @@ -1231,17 +1231,19 @@ image similar to the result of rebuilding an ephemeral disk. 2.94 ---------------------- +---- The ``hostname`` parameter to the ``POST /servers`` (create server), ``PUT /servers/{id}`` (update server) and ``POST /servers/{server_id}/action (rebuild)`` (rebuild server) APIs is now allowed to be a Fully Qualified Domain Name (FQDN). +.. _microversion 2.95: -2.95 ---------------------- +2.95 (Maximum in 2023.1 Antelope) +--------------------------------- Any evacuated instances will be now stopped at destination. This -requires minimun compute version 27.0.0 (antelope 2023.1). Operators -can still use previous microversion for older behavior. +requires minimun nova release 27.0.0, OpenStack release 2023.1 +Antelope. Operators can still use previous microversion for older +behavior. diff --git a/nova/compute/manager.py b/nova/compute/manager.py index 0bf6fdfec5..5ea71827fc 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1473,7 +1473,7 @@ class ComputeManager(manager.Manager): node. """ try: - node_uuids = self.driver.get_available_node_uuids() + node_ids = self.driver.get_nodenames_by_uuid() except exception.VirtDriverNotReady: LOG.warning( "Virt driver is not ready. If this is the first time this " @@ -1481,7 +1481,8 @@ class ComputeManager(manager.Manager): "this warning.") return {} - nodes = objects.ComputeNodeList.get_all_by_uuids(context, node_uuids) + nodes = objects.ComputeNodeList.get_all_by_uuids(context, + list(node_ids.keys())) if not nodes: # NOTE(danms): This should only happen if the compute_id is # pre-provisioned on a host that has never started. @@ -1489,9 +1490,18 @@ class ComputeManager(manager.Manager): 'database. If this is the first time this service is ' 'starting on this host, then you can ignore this ' 'warning.', - node_uuids, self.host) + list(node_ids.keys()), self.host) return {} + for node in nodes: + if node.hypervisor_hostname != node_ids.get(node.uuid): + raise exception.InvalidConfiguration( + ('My compute node %s has hypervisor_hostname %s ' + 'but virt driver reports it should be %s. Possible ' + 'rename detected, refusing to start!') % ( + node.uuid, node.hypervisor_hostname, + node_ids.get(node.uuid))) + return {n.uuid: n for n in nodes} def _ensure_existing_node_identity(self, service_ref): @@ -1499,15 +1509,21 @@ class ComputeManager(manager.Manager): to write our node identity uuid (if not already done) based on nodes assigned to us in the database. """ - if service_ref.version >= service_obj.NODE_IDENTITY_VERSION: - # Already new enough, nothing to do here - return - if 'ironic' in CONF.compute_driver.lower(): # We do not persist a single local node identity for # ironic return + if service_ref.version >= service_obj.NODE_IDENTITY_VERSION: + # Already new enough, nothing to do here, but make sure that we + # have a UUID file already, as this is not our first time starting. + if nova.virt.node.read_local_node_uuid() is None: + raise exception.InvalidConfiguration( + ('No local node identity found, but this is not our ' + 'first startup on this host. Refusing to start after ' + 'potentially having lost that state!')) + return + if nova.virt.node.read_local_node_uuid(): # We already have a local node identity, no migration needed return @@ -1594,12 +1610,6 @@ class ComputeManager(manager.Manager): # 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 context = nova.context.get_admin_context() nodes_by_uuid = self._get_nodes(context) @@ -1617,6 +1627,12 @@ class ComputeManager(manager.Manager): self._validate_pinning_configuration(instances) self._validate_vtpm_configuration(instances) + # 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 try: # checking that instance was not already evacuated to other host evacuated_instances = self._destroy_evacuated_instances( @@ -6858,6 +6874,9 @@ class ComputeManager(manager.Manager): current_power_state = self._get_power_state(instance) network_info = self.network_api.get_instance_nw_info(context, instance) + ports_id = [vif['id'] for vif in network_info] + self.network_api.unbind_ports(context, ports_id, detach=False) + block_device_info = self._get_instance_block_device_info(context, instance, bdms=bdms) @@ -8897,7 +8916,8 @@ class ComputeManager(manager.Manager): # 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( + future = nova.utils.pass_context( + 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) @@ -10181,7 +10201,9 @@ class ComputeManager(manager.Manager): else: LOG.debug('Triggering sync for uuid %s', uuid) self._syncs_in_progress[uuid] = True - self._sync_power_pool.spawn_n(_sync, db_instance) + nova.utils.pass_context(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: @@ -10464,6 +10486,14 @@ class ComputeManager(manager.Manager): LOG.exception( "Error updating PCI resources for node %(node)s.", {'node': nodename}) + except exception.InvalidConfiguration as e: + if startup: + # If this happens during startup, we need to let it raise to + # abort our service startup. + raise + else: + LOG.error("Error updating resources for node %s: %s", + nodename, e) except Exception: LOG.exception("Error updating resources for node %(node)s.", {'node': nodename}) diff --git a/nova/compute/resource_tracker.py b/nova/compute/resource_tracker.py index 70c56fd2e3..3f911f3708 100644 --- a/nova/compute/resource_tracker.py +++ b/nova/compute/resource_tracker.py @@ -728,7 +728,13 @@ class ResourceTracker(object): cn = objects.ComputeNode(context) cn.host = self.host self._copy_resources(cn, resources, initial=True) - cn.create() + try: + cn.create() + except exception.DuplicateRecord: + raise exception.InvalidConfiguration( + 'Duplicate compute node record found for host %s node %s' % ( + cn.host, cn.hypervisor_hostname)) + # Only map the ComputeNode into compute_nodes if create() was OK # because if create() fails, on the next run through here nodename # would be in compute_nodes and we won't try to create again (because diff --git a/nova/compute/rpcapi.py b/nova/compute/rpcapi.py index 4a97a90807..efc06300db 100644 --- a/nova/compute/rpcapi.py +++ b/nova/compute/rpcapi.py @@ -425,7 +425,7 @@ class ComputeAPI(object): 'xena': '6.0', 'yoga': '6.0', 'zed': '6.1', - 'antilope': '6.2', + 'antelope': '6.2', } @property diff --git a/nova/conductor/manager.py b/nova/conductor/manager.py index 8177519331..4b34b8339c 100644 --- a/nova/conductor/manager.py +++ b/nova/conductor/manager.py @@ -2096,8 +2096,8 @@ class ComputeTaskManager: skipped_host(target_ctxt, host, image_ids) continue - fetch_pool.spawn_n(wrap_cache_images, target_ctxt, host, - image_ids) + utils.pass_context(fetch_pool.spawn_n, wrap_cache_images, + target_ctxt, host, image_ids) # Wait until all those things finish fetch_pool.waitall() diff --git a/nova/conf/libvirt.py b/nova/conf/libvirt.py index 16a3f63090..204fe5c4b8 100644 --- a/nova/conf/libvirt.py +++ b/nova/conf/libvirt.py @@ -1478,6 +1478,23 @@ Related options: """), ] +libvirt_cpu_mgmt_opts = [ + cfg.BoolOpt('cpu_power_management', + default=False, + help='Use libvirt to manage CPU cores performance.'), + cfg.StrOpt('cpu_power_management_strategy', + choices=['cpu_state', 'governor'], + default='cpu_state', + help='Tuning strategy to reduce CPU power consumption when ' + 'unused'), + cfg.StrOpt('cpu_power_governor_low', + default='powersave', + help='Governor to use in order ' + 'to reduce CPU power consumption'), + cfg.StrOpt('cpu_power_governor_high', + default='performance', + help='Governor to use in order to have best CPU performance'), +] ALL_OPTS = list(itertools.chain( libvirt_general_opts, @@ -1499,6 +1516,7 @@ ALL_OPTS = list(itertools.chain( libvirt_volume_nvmeof_opts, libvirt_pmem_opts, libvirt_vtpm_opts, + libvirt_cpu_mgmt_opts, )) diff --git a/nova/conf/spice.py b/nova/conf/spice.py index 59ed4e80a0..e5854946f1 100644 --- a/nova/conf/spice.py +++ b/nova/conf/spice.py @@ -85,6 +85,59 @@ Agent. With the Spice agent installed the following features are enabled: needing to click inside the console or press keys to release it. The performance of mouse movement is also improved. """), + cfg.StrOpt('image_compression', + advanced=True, + choices=[ + ('auto_glz', 'enable image compression mode to choose between glz ' + 'and quic algorithm, based on image properties'), + ('auto_lz', 'enable image compression mode to choose between lz ' + 'and quic algorithm, based on image properties'), + ('quic', 'enable image compression based on the SFALIC algorithm'), + ('glz', 'enable image compression using lz with history based ' + 'global dictionary'), + ('lz', 'enable image compression with the Lempel-Ziv algorithm'), + ('off', 'disable image compression') + ], + help=""" +Configure the SPICE image compression (lossless). +"""), + cfg.StrOpt('jpeg_compression', + advanced=True, + choices=[ + ('auto', 'enable JPEG image compression automatically'), + ('never', 'disable JPEG image compression'), + ('always', 'enable JPEG image compression') + ], + help=""" +Configure the SPICE wan image compression (lossy for slow links). +"""), + cfg.StrOpt('zlib_compression', + advanced=True, + choices=[ + ('auto', 'enable zlib image compression automatically'), + ('never', 'disable zlib image compression'), + ('always', 'enable zlib image compression') + ], + help=""" +Configure the SPICE wan image compression (lossless for slow links). +"""), + cfg.BoolOpt('playback_compression', + advanced=True, + help=""" +Enable the SPICE audio stream compression (using celt). +"""), + cfg.StrOpt('streaming_mode', + advanced=True, + choices=[ + ('filter', 'SPICE server adds additional filters to decide if ' + 'video streaming should be activated'), + ('all', 'any fast-refreshing window can be encoded into a video ' + 'stream'), + ('off', 'no video detection and (lossy) compression is performed') + ], + help=""" +Configure the SPICE video stream detection and (lossy) compression. +"""), cfg.URIOpt('html5proxy_base_url', default='http://127.0.0.1:6082/spice_auto.html', help=""" diff --git a/nova/exception.py b/nova/exception.py index 20c112b628..0c0ffa85a1 100644 --- a/nova/exception.py +++ b/nova/exception.py @@ -2512,8 +2512,12 @@ class InvalidNodeConfiguration(NovaException): msg_fmt = _('Invalid node identity configuration: %(reason)s') +class DuplicateRecord(NovaException): + msg_fmt = _('Unable to create duplicate record for %(target)s') + + class NotSupportedComputeForEvacuateV295(NotSupported): - msg_fmt = _("Starting to microversion 2.95, evacuate API will stop " + msg_fmt = _("Starting with microversion 2.95, evacuate API will stop " "instance on destination. To evacuate before upgrades are " "complete please use an older microversion. Required version " "for compute %(expected), current version %(currently)s") diff --git a/nova/filesystem.py b/nova/filesystem.py new file mode 100644 index 0000000000..5394d2d835 --- /dev/null +++ b/nova/filesystem.py @@ -0,0 +1,59 @@ +# 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. + +"""Functions to address filesystem calls, particularly sysfs.""" + +import os + +from oslo_log import log as logging + +from nova import exception + +LOG = logging.getLogger(__name__) + + +SYS = '/sys' + + +# NOTE(bauzas): this method is deliberately not wrapped in a privsep entrypoint +def read_sys(path: str) -> str: + """Reads the content of a file in the sys filesystem. + + :param path: relative or absolute. If relative, will be prefixed by /sys. + :returns: contents of that file. + :raises: nova.exception.FileNotFound if we can't read that file. + """ + try: + # The path can be absolute with a /sys prefix but that's fine. + with open(os.path.join(SYS, path), mode='r') as data: + return data.read() + except (OSError, ValueError) as exc: + raise exception.FileNotFound(file_path=path) from exc + + +# NOTE(bauzas): this method is deliberately not wrapped in a privsep entrypoint +# In order to correctly use it, you need to decorate the caller with a specific +# privsep entrypoint. +def write_sys(path: str, data: str) -> None: + """Writes the content of a file in the sys filesystem with data. + + :param path: relative or absolute. If relative, will be prefixed by /sys. + :param data: the data to write. + :returns: contents of that file. + :raises: nova.exception.FileNotFound if we can't write that file. + """ + try: + # The path can be absolute with a /sys prefix but that's fine. + with open(os.path.join(SYS, path), mode='w') as fd: + fd.write(data) + except (OSError, ValueError) as exc: + raise exception.FileNotFound(file_path=path) from exc diff --git a/nova/network/neutron.py b/nova/network/neutron.py index 27e7d06455..6c9f19f010 100644 --- a/nova/network/neutron.py +++ b/nova/network/neutron.py @@ -612,10 +612,22 @@ class API: raise exception.ExternalNetworkAttachForbidden( network_uuid=net['id']) + def unbind_ports(self, context, ports, detach=True): + """Unbind and detach the given ports by clearing their + device_owner and dns_name. + The device_id will also be cleaned if detach=True. + + :param context: The request context. + :param ports: list of port IDs. + """ + neutron = get_client(context) + self._unbind_ports(context, ports, neutron, detach=detach) + def _unbind_ports(self, context, ports, - neutron, port_client=None): - """Unbind the given ports by clearing their device_id, + neutron, port_client=None, detach=True): + """Unbind and detach the given ports by clearing their device_owner and dns_name. + The device_id will also be cleaned if detach=True. :param context: The request context. :param ports: list of port IDs. @@ -638,11 +650,12 @@ class API: port_req_body: ty.Dict[str, ty.Any] = { 'port': { - 'device_id': '', - 'device_owner': '', constants.BINDING_HOST_ID: None, } } + if detach: + port_req_body['port']['device_id'] = '' + port_req_body['port']['device_owner'] = '' try: port = self._show_port( context, port_id, neutron_client=neutron, diff --git a/nova/objects/compute_node.py b/nova/objects/compute_node.py index 528cfc0776..dfc1b2ae28 100644 --- a/nova/objects/compute_node.py +++ b/nova/objects/compute_node.py @@ -12,6 +12,7 @@ # License for the specific language governing permissions and limitations # under the License. +from oslo_db import exception as db_exc from oslo_serialization import jsonutils from oslo_utils import uuidutils from oslo_utils import versionutils @@ -339,7 +340,12 @@ class ComputeNode(base.NovaPersistentObject, base.NovaObject): self._convert_supported_instances_to_db_format(updates) self._convert_pci_stats_to_db_format(updates) - db_compute = db.compute_node_create(self._context, updates) + try: + db_compute = db.compute_node_create(self._context, updates) + except db_exc.DBDuplicateEntry: + target = 'compute node %s:%s' % (updates['hypervisor_hostname'], + updates['uuid']) + raise exception.DuplicateRecord(target=target) self._from_db_object(self._context, self, db_compute) @base.remotable diff --git a/nova/objects/service.py b/nova/objects/service.py index 0ed443ef17..1a4629cc84 100644 --- a/nova/objects/service.py +++ b/nova/objects/service.py @@ -237,15 +237,30 @@ SERVICE_VERSION_HISTORY = ( # local node identity for single-node systems. NODE_IDENTITY_VERSION = 65 -# This is used to raise an error at service startup if older than N-1 computes -# are detected. Update this at the beginning of every release cycle to point to -# the smallest service version that was added in N-1. -OLDEST_SUPPORTED_SERVICE_VERSION = 'Yoga' +# This is used to raise an error at service startup if older than supported +# computes are detected. +# NOTE(sbauza) : Please modify it this way : +# * At the beginning of a non-SLURP release (eg. 2023.2 Bobcat) (or just after +# the previous SLURP release RC1, like 2023.1 Antelope), please bump +# OLDEST_SUPPORTED_SERVICE_VERSION to the previous SLURP release (in that +# example, Antelope) +# * At the beginning of a SLURP release (eg. 2024.1 C) (or just after the +# previous non-SLURP release RC1, like 2023.2 Bobcat), please keep the +# OLDEST_SUPPORTED_SERVICE_VERSION value using the previous SLURP release +# (in that example, Antelope) +# * At the end of any release (SLURP or non-SLURP), please modify +# SERVICE_VERSION_ALIASES to add a key/value with key being the release name +# and value be the latest service version that the release supports (for +# example, before Bobcat RC1, please add 'Bobcat': XX where X is the latest +# servion version that was added) +OLDEST_SUPPORTED_SERVICE_VERSION = 'Antelope' SERVICE_VERSION_ALIASES = { 'Victoria': 52, 'Wallaby': 54, 'Xena': 57, 'Yoga': 61, + 'Zed': 64, + 'Antelope': 66, } diff --git a/nova/pci/request.py b/nova/pci/request.py index 27ada6c045..8ae2385549 100644 --- a/nova/pci/request.py +++ b/nova/pci/request.py @@ -168,7 +168,7 @@ def _get_alias_from_config() -> Alias: def _translate_alias_to_requests( - alias_spec: str, affinity_policy: str = None, + alias_spec: str, affinity_policy: ty.Optional[str] = None, ) -> ty.List['objects.InstancePCIRequest']: """Generate complete pci requests from pci aliases in extra_spec.""" pci_aliases = _get_alias_from_config() @@ -255,7 +255,7 @@ def get_instance_pci_request_from_vif( def get_pci_requests_from_flavor( - flavor: 'objects.Flavor', affinity_policy: str = None, + flavor: 'objects.Flavor', affinity_policy: ty.Optional[str] = None, ) -> 'objects.InstancePCIRequests': """Validate and return PCI requests. diff --git a/nova/pci/stats.py b/nova/pci/stats.py index 5c5f7c669c..c6e4844b34 100644 --- a/nova/pci/stats.py +++ b/nova/pci/stats.py @@ -82,7 +82,7 @@ class PciDeviceStats(object): self, numa_topology: 'objects.NUMATopology', stats: 'objects.PCIDevicePoolList' = None, - dev_filter: whitelist.Whitelist = None, + dev_filter: ty.Optional[whitelist.Whitelist] = None, ) -> None: self.numa_topology = numa_topology self.pools = ( diff --git a/nova/pci/whitelist.py b/nova/pci/whitelist.py index 8862a0ef4f..152cc29ca6 100644 --- a/nova/pci/whitelist.py +++ b/nova/pci/whitelist.py @@ -33,7 +33,7 @@ class Whitelist(object): assignable. """ - def __init__(self, whitelist_spec: str = None) -> None: + def __init__(self, whitelist_spec: ty.Optional[str] = None) -> None: """White list constructor For example, the following json string specifies that devices whose diff --git a/nova/scheduler/client/report.py b/nova/scheduler/client/report.py index 1242752be1..7c14f3d7ef 100644 --- a/nova/scheduler/client/report.py +++ b/nova/scheduler/client/report.py @@ -1047,7 +1047,7 @@ class SchedulerReportClient(object): context: nova_context.RequestContext, rp_uuid: str, traits: ty.Iterable[str], - generation: int = None + generation: ty.Optional[int] = None ): """Replace a provider's traits with those specified. diff --git a/nova/scheduler/filters/__init__.py b/nova/scheduler/filters/__init__.py index 74e24b7bc3..785a13279e 100644 --- a/nova/scheduler/filters/__init__.py +++ b/nova/scheduler/filters/__init__.py @@ -16,8 +16,12 @@ """ Scheduler host filters """ +from oslo_log import log as logging + from nova import filters +LOG = logging.getLogger(__name__) + class BaseHostFilter(filters.BaseFilter): """Base class for host filters.""" @@ -53,6 +57,43 @@ class BaseHostFilter(filters.BaseFilter): raise NotImplementedError() +class CandidateFilterMixin: + """Mixing that helps to implement a Filter that needs to filter host by + Placement allocation candidates. + """ + + def filter_candidates(self, host_state, filter_func): + """Checks still viable allocation candidates by the filter_func and + keep only those that are passing it. + + :param host_state: HostState object holding the list of still viable + allocation candidates + :param filter_func: A callable that takes an allocation candidate and + returns a True like object if the candidate passed the filter or a + False like object if it doesn't. + """ + good_candidates = [] + for candidate in host_state.allocation_candidates: + LOG.debug( + f'{self.__class__.__name__} tries allocation candidate: ' + f'{candidate}', + ) + if filter_func(candidate): + LOG.debug( + f'{self.__class__.__name__} accepted allocation ' + f'candidate: {candidate}', + ) + good_candidates.append(candidate) + else: + LOG.debug( + f'{self.__class__.__name__} rejected allocation ' + f'candidate: {candidate}', + ) + + host_state.allocation_candidates = good_candidates + return good_candidates + + class HostFilterHandler(filters.BaseFilterHandler): def __init__(self): super(HostFilterHandler, self).__init__(BaseHostFilter) diff --git a/nova/scheduler/filters/numa_topology_filter.py b/nova/scheduler/filters/numa_topology_filter.py index 7ec9ca5648..ae50db90e5 100644 --- a/nova/scheduler/filters/numa_topology_filter.py +++ b/nova/scheduler/filters/numa_topology_filter.py @@ -20,7 +20,10 @@ from nova.virt import hardware LOG = logging.getLogger(__name__) -class NUMATopologyFilter(filters.BaseHostFilter): +class NUMATopologyFilter( + filters.BaseHostFilter, + filters.CandidateFilterMixin, +): """Filter on requested NUMA topology.""" # NOTE(sean-k-mooney): In change I0322d872bdff68936033a6f5a54e8296a6fb343 @@ -97,34 +100,19 @@ class NUMATopologyFilter(filters.BaseHostFilter): if network_metadata: limits.network_metadata = network_metadata - good_candidates = [] - for candidate in host_state.allocation_candidates: - LOG.debug( - 'NUMATopologyFilter tries allocation candidate: %s, %s', - candidate, requested_topology - ) - instance_topology = (hardware.numa_fit_instance_to_host( - host_topology, requested_topology, + good_candidates = self.filter_candidates( + host_state, + lambda candidate: hardware.numa_fit_instance_to_host( + host_topology, + requested_topology, limits=limits, pci_requests=pci_requests, pci_stats=host_state.pci_stats, - provider_mapping=candidate['mappings'], - )) - if instance_topology: - LOG.debug( - 'NUMATopologyFilter accepted allocation candidate: %s', - candidate - ) - good_candidates.append(candidate) - else: - LOG.debug( - 'NUMATopologyFilter rejected allocation candidate: %s', - candidate - ) - - host_state.allocation_candidates = good_candidates - - if not host_state.allocation_candidates: + provider_mapping=candidate["mappings"], + ), + ) + + if not good_candidates: LOG.debug("%(host)s, %(node)s fails NUMA topology " "requirements. The instance does not fit on this " "host.", {'host': host_state.host, diff --git a/nova/scheduler/filters/pci_passthrough_filter.py b/nova/scheduler/filters/pci_passthrough_filter.py index 36f0b5901c..992879072a 100644 --- a/nova/scheduler/filters/pci_passthrough_filter.py +++ b/nova/scheduler/filters/pci_passthrough_filter.py @@ -20,7 +20,10 @@ from nova.scheduler import filters LOG = logging.getLogger(__name__) -class PciPassthroughFilter(filters.BaseHostFilter): +class PciPassthroughFilter( + filters.BaseHostFilter, + filters.CandidateFilterMixin, +): """Pci Passthrough Filter based on PCI request Filter that schedules instances on a host if the host has devices @@ -54,28 +57,12 @@ class PciPassthroughFilter(filters.BaseHostFilter): {'host_state': host_state, 'requests': pci_requests}) return False - good_candidates = [] - for candidate in host_state.allocation_candidates: - LOG.debug( - 'PciPassthroughFilter tries allocation candidate: %s', - candidate - ) - if host_state.pci_stats.support_requests( - pci_requests.requests, - provider_mapping=candidate['mappings'] - ): - LOG.debug( - 'PciPassthroughFilter accepted allocation candidate: %s', - candidate - ) - good_candidates.append(candidate) - else: - LOG.debug( - 'PciPassthroughFilter rejected allocation candidate: %s', - candidate - ) - - host_state.allocation_candidates = good_candidates + good_candidates = self.filter_candidates( + host_state, + lambda candidate: host_state.pci_stats.support_requests( + pci_requests.requests, provider_mapping=candidate["mappings"] + ), + ) if not good_candidates: LOG.debug("%(host_state)s doesn't have the required PCI devices" diff --git a/nova/scheduler/manager.py b/nova/scheduler/manager.py index 11581c4f2d..620519d403 100644 --- a/nova/scheduler/manager.py +++ b/nova/scheduler/manager.py @@ -23,6 +23,7 @@ import collections import copy import random +from keystoneauth1 import exceptions as ks_exc from oslo_log import log as logging import oslo_messaging as messaging from oslo_serialization import jsonutils @@ -67,10 +68,42 @@ class SchedulerManager(manager.Manager): self.host_manager = host_manager.HostManager() self.servicegroup_api = servicegroup.API() self.notifier = rpc.get_notifier('scheduler') - self.placement_client = report.report_client_singleton() + self._placement_client = None + + try: + # Test our placement client during initialization + self.placement_client + except (ks_exc.EndpointNotFound, + ks_exc.DiscoveryFailure, + ks_exc.RequestTimeout, + ks_exc.GatewayTimeout, + ks_exc.ConnectFailure) as e: + # Non-fatal, likely transient (although not definitely); + # continue startup but log the warning so that when things + # fail later, it will be clear why we can not do certain + # things. + LOG.warning('Unable to initialize placement client (%s); ' + 'Continuing with startup, but scheduling ' + 'will not be possible.', e) + except (ks_exc.MissingAuthPlugin, + ks_exc.Unauthorized) as e: + # This is almost definitely fatal mis-configuration. The + # Unauthorized error might be transient, but it is + # probably reasonable to consider it fatal. + LOG.error('Fatal error initializing placement client; ' + 'config is incorrect or incomplete: %s', e) + raise + except Exception as e: + # Unknown/unexpected errors here are fatal + LOG.error('Fatal error initializing placement client: %s', e) + raise super().__init__(service_name='scheduler', *args, **kwargs) + @property + def placement_client(self): + return report.report_client_singleton() + @periodic_task.periodic_task( spacing=CONF.scheduler.discover_hosts_in_cells_interval, run_immediately=True) diff --git a/nova/scheduler/utils.py b/nova/scheduler/utils.py index c7e6ffed97..02c44093bd 100644 --- a/nova/scheduler/utils.py +++ b/nova/scheduler/utils.py @@ -1080,6 +1080,17 @@ _SUPPORTS_SOFT_AFFINITY = None _SUPPORTS_SOFT_ANTI_AFFINITY = None +def reset_globals(): + global _SUPPORTS_AFFINITY + _SUPPORTS_AFFINITY = None + global _SUPPORTS_ANTI_AFFINITY + _SUPPORTS_ANTI_AFFINITY = None + global _SUPPORTS_SOFT_AFFINITY + _SUPPORTS_SOFT_AFFINITY = None + global _SUPPORTS_SOFT_ANTI_AFFINITY + _SUPPORTS_SOFT_ANTI_AFFINITY = None + + def _get_group_details(context, instance_uuid, user_group_hosts=None): """Provide group_hosts and group_policies sets related to instances if those instances are belonging to a group and if corresponding filters are diff --git a/nova/test.py b/nova/test.py index 562bd2516e..e37967b06d 100644 --- a/nova/test.py +++ b/nova/test.py @@ -62,6 +62,7 @@ from nova import objects from nova.objects import base as objects_base from nova import quota from nova.scheduler.client import report +from nova.scheduler import utils as scheduler_utils from nova.tests import fixtures as nova_fixtures from nova.tests.unit import matchers from nova import utils @@ -171,6 +172,12 @@ class TestCase(base.BaseTestCase): # base class when USES_DB is True. NUMBER_OF_CELLS = 1 + # The stable compute id stuff is intentionally singleton-ish, which makes + # it a nightmare for testing multiple host/node combinations in tests like + # we do. So, mock it out by default, unless the test is specifically + # designed to handle it. + STUB_COMPUTE_ID = True + def setUp(self): """Run before each test method to initialize test environment.""" # Ensure BaseTestCase's ConfigureLogging fixture is disabled since @@ -301,7 +308,14 @@ class TestCase(base.BaseTestCase): # Reset our local node uuid cache (and avoid writing to the # local filesystem when we generate a new one). - self.useFixture(nova_fixtures.ComputeNodeIdFixture()) + if self.STUB_COMPUTE_ID: + self.useFixture(nova_fixtures.ComputeNodeIdFixture()) + + # Reset globals indicating affinity filter support. Some tests may set + # self.flags(enabled_filters=...) which could make the affinity filter + # support globals get set to a non-default configuration which affects + # all other tests. + scheduler_utils.reset_globals() def _setup_cells(self): """Setup a normal cellsv2 environment. diff --git a/nova/tests/fixtures/__init__.py b/nova/tests/fixtures/__init__.py index df254608fd..9ff4a2a601 100644 --- a/nova/tests/fixtures/__init__.py +++ b/nova/tests/fixtures/__init__.py @@ -16,6 +16,8 @@ from .cast_as_call import CastAsCallFixture # noqa: F401 from .cinder import CinderFixture # noqa: F401 from .conf import ConfFixture # noqa: F401, F403 from .cyborg import CyborgFixture # noqa: F401 +from .filesystem import SysFileSystemFixture # noqa: F401 +from .filesystem import TempFileSystemFixture # noqa: F401 from .glance import GlanceFixture # noqa: F401 from .libvirt import LibvirtFixture # noqa: F401 from .libvirt_imagebackend import LibvirtImageBackendFixture # noqa: F401 diff --git a/nova/tests/fixtures/filesystem.py b/nova/tests/fixtures/filesystem.py new file mode 100644 index 0000000000..932d42fe27 --- /dev/null +++ b/nova/tests/fixtures/filesystem.py @@ -0,0 +1,81 @@ +# 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. + +import os +import shutil +import tempfile +from unittest import mock + +import fixtures + +from nova import filesystem +from nova.virt.libvirt.cpu import core + + +SYS = 'sys' + + +class TempFileSystemFixture(fixtures.Fixture): + """Creates a fake / filesystem""" + + def _setUp(self): + self.temp_dir = tempfile.TemporaryDirectory(prefix='fake_fs') + # NOTE(sbauza): I/O disk errors may raise an exception here, as we + # don't ignore them. If that's causing a problem in our CI jobs, the + # recommended solution is to use shutil.rmtree instead of cleanup() + # with ignore_errors parameter set to True (or wait for the minimum + # python version to be 3.10 as TemporaryDirectory will provide + # ignore_cleanup_errors parameter) + self.addCleanup(self.temp_dir.cleanup) + + +class SysFileSystemFixture(TempFileSystemFixture): + """Creates a fake /sys filesystem""" + + def __init__(self, cpus_supported=None): + self.cpus_supported = cpus_supported or 10 + + def _setUp(self): + super()._setUp() + self.sys_path = os.path.join(self.temp_dir.name, SYS) + self.addCleanup(shutil.rmtree, self.sys_path, ignore_errors=True) + + sys_patcher = mock.patch( + 'nova.filesystem.SYS', + new_callable=mock.PropertyMock(return_value=self.sys_path)) + self.sys_mock = sys_patcher.start() + self.addCleanup(sys_patcher.stop) + + avail_path_patcher = mock.patch( + 'nova.virt.libvirt.cpu.core.AVAILABLE_PATH', + new_callable=mock.PropertyMock( + return_value=os.path.join(self.sys_path, + 'devices/system/cpu/present'))) + self.avail_path_mock = avail_path_patcher.start() + self.addCleanup(avail_path_patcher.stop) + + cpu_path_patcher = mock.patch( + 'nova.virt.libvirt.cpu.core.CPU_PATH_TEMPLATE', + new_callable=mock.PropertyMock( + return_value=os.path.join(self.sys_path, + 'devices/system/cpu/cpu%(core)s'))) + self.cpu_path_mock = cpu_path_patcher.start() + self.addCleanup(cpu_path_patcher.stop) + + for cpu_nr in range(self.cpus_supported): + cpu_dir = os.path.join(self.cpu_path_mock % {'core': cpu_nr}) + os.makedirs(os.path.join(cpu_dir, 'cpufreq')) + filesystem.write_sys( + os.path.join(cpu_dir, 'cpufreq/scaling_governor'), + data='powersave') + filesystem.write_sys(core.AVAILABLE_PATH, + f'0-{self.cpus_supported - 1}') diff --git a/nova/tests/fixtures/nova.py b/nova/tests/fixtures/nova.py index 76dc63755d..5fd893e7dc 100644 --- a/nova/tests/fixtures/nova.py +++ b/nova/tests/fixtures/nova.py @@ -1863,3 +1863,7 @@ class ComputeNodeIdFixture(fixtures.Fixture): self.useFixture(fixtures.MockPatch( 'nova.virt.node.write_local_node_uuid', lambda uuid: None)) + self.useFixture(fixtures.MockPatch( + 'nova.compute.manager.ComputeManager.' + '_ensure_existing_node_identity', + mock.DEFAULT)) diff --git a/nova/tests/functional/libvirt/test_pci_sriov_servers.py b/nova/tests/functional/libvirt/test_pci_sriov_servers.py index 6b8b254af9..098a0e857b 100644 --- a/nova/tests/functional/libvirt/test_pci_sriov_servers.py +++ b/nova/tests/functional/libvirt/test_pci_sriov_servers.py @@ -1549,7 +1549,11 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): 'not supported for instance with vDPA ports', ex.response.text) + # NOTE(sbauza): Now we're post-Antelope release, we don't need to support + # this test def test_attach_interface_service_version_61(self): + self.flags(disable_compute_service_check_for_ffu=True, + group='workarounds') with mock.patch( "nova.objects.service.get_minimum_version_all_cells", return_value=61 @@ -1578,7 +1582,11 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): self.assertEqual(hostname, port['binding:host_id']) self.assertEqual(server['id'], port['device_id']) + # NOTE(sbauza): Now we're post-Antelope release, we don't need to support + # this test def test_detach_interface_service_version_61(self): + self.flags(disable_compute_service_check_for_ffu=True, + group='workarounds') with mock.patch( "nova.objects.service.get_minimum_version_all_cells", return_value=61 @@ -1612,10 +1620,7 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): # to any host but should still be owned by the vm port = self.neutron.show_port(vdpa_port['id'])['port'] self.assertEqual(server['id'], port['device_id']) - # FIXME(sean-k-mooney): we should be unbinding the port from - # the host when we shelve offload but we don't today. - # This is unrelated to vdpa port and is a general issue. - self.assertEqual(hostname, port['binding:host_id']) + self.assertIsNone(port['binding:host_id']) self.assertIn('binding:profile', port) self.assertIsNone(server['OS-EXT-SRV-ATTR:hypervisor_hostname']) self.assertIsNone(server['OS-EXT-SRV-ATTR:host']) @@ -1637,9 +1642,7 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): self.assertPCIDeviceCounts(hostname, total=num_pci, free=num_pci) self.assertIsNone(server['OS-EXT-SRV-ATTR:hypervisor_hostname']) port = self.neutron.show_port(vdpa_port['id'])['port'] - # FIXME(sean-k-mooney): shelve offload should unbind the port - # self.assertEqual('', port['binding:host_id']) - self.assertEqual(hostname, port['binding:host_id']) + self.assertIsNone(port['binding:host_id']) server = self._unshelve_server(server) self.assertPCIDeviceCounts(hostname, total=num_pci, free=num_pci - 2) @@ -1670,9 +1673,7 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): self.assertPCIDeviceCounts(source, total=num_pci, free=num_pci) self.assertIsNone(server['OS-EXT-SRV-ATTR:hypervisor_hostname']) port = self.neutron.show_port(vdpa_port['id'])['port'] - # FIXME(sean-k-mooney): shelve should unbind the port - # self.assertEqual('', port['binding:host_id']) - self.assertEqual(source, port['binding:host_id']) + self.assertIsNone(port['binding:host_id']) # force the unshelve to the other host self.api.put_service( @@ -1871,7 +1872,11 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): self.assertEqual( dest, server['OS-EXT-SRV-ATTR:hypervisor_hostname']) + # NOTE(sbauza): Now we're post-Antelope release, we don't need to support + # this test def test_suspend_and_resume_service_version_62(self): + self.flags(disable_compute_service_check_for_ffu=True, + group='workarounds') with mock.patch( "nova.objects.service.get_minimum_version_all_cells", return_value=62 @@ -1890,7 +1895,11 @@ class VDPAServersTest(_PCIServersWithMigrationTestBase): self.assertPCIDeviceCounts(source, total=num_pci, free=num_pci - 2) self.assertEqual('ACTIVE', server['status']) + # NOTE(sbauza): Now we're post-Antelope release, we don't need to support + # this test def test_live_migrate_service_version_62(self): + self.flags(disable_compute_service_check_for_ffu=True, + group='workarounds') with mock.patch( "nova.objects.service.get_minimum_version_all_cells", return_value=62 @@ -3926,17 +3935,7 @@ class RemoteManagedServersTest(_PCIServersWithMigrationTestBase): port = self.neutron.show_port(uuids.dpu_tunnel_port)['port'] self.assertIn('binding:profile', port) - self.assertEqual( - { - 'pci_vendor_info': '15b3:101e', - 'pci_slot': '0000:82:00.4', - 'physical_network': None, - 'pf_mac_address': '52:54:00:1e:59:02', - 'vf_num': 3, - 'card_serial_number': 'MT0000X00002', - }, - port['binding:profile'], - ) + self.assertEqual({}, port['binding:profile']) def test_suspend(self): self.start_compute() diff --git a/nova/tests/functional/libvirt/test_power_manage.py b/nova/tests/functional/libvirt/test_power_manage.py new file mode 100644 index 0000000000..9f80446bd6 --- /dev/null +++ b/nova/tests/functional/libvirt/test_power_manage.py @@ -0,0 +1,270 @@ +# 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. + +from unittest import mock + +import fixtures + +from nova import context as nova_context +from nova import exception +from nova import objects +from nova.tests import fixtures as nova_fixtures +from nova.tests.fixtures import libvirt as fakelibvirt +from nova.tests.functional.libvirt import base +from nova.virt import hardware +from nova.virt.libvirt.cpu import api as cpu_api + + +class PowerManagementTestsBase(base.ServersTestBase): + + ADDITIONAL_FILTERS = ['NUMATopologyFilter'] + + ADMIN_API = True + + def setUp(self): + super(PowerManagementTestsBase, self).setUp() + + self.ctxt = nova_context.get_admin_context() + + # Mock the 'NUMATopologyFilter' filter, as most tests need to inspect + # this + host_manager = self.scheduler.manager.host_manager + numa_filter_class = host_manager.filter_cls_map['NUMATopologyFilter'] + host_pass_mock = mock.Mock(wraps=numa_filter_class().host_passes) + _p = mock.patch('nova.scheduler.filters' + '.numa_topology_filter.NUMATopologyFilter.host_passes', + side_effect=host_pass_mock) + self.mock_filter = _p.start() + self.addCleanup(_p.stop) + + # for the sake of resizing, we need to patch the two methods below + self.useFixture(fixtures.MockPatch( + 'nova.virt.libvirt.LibvirtDriver._get_instance_disk_info', + return_value=[])) + self.useFixture(fixtures.MockPatch('os.rename')) + + self.useFixture(nova_fixtures.PrivsepFixture()) + + # Defining the main flavor for 4 vCPUs all pinned + self.extra_spec = { + 'hw:cpu_policy': 'dedicated', + 'hw:cpu_thread_policy': 'prefer', + } + self.pcpu_flavor_id = self._create_flavor( + vcpu=4, extra_spec=self.extra_spec) + + def _assert_server_cpus_state(self, server, expected='online'): + inst = objects.Instance.get_by_uuid(self.ctxt, server['id']) + if not inst.numa_topology: + self.fail('Instance should have a NUMA topology in order to know ' + 'its physical CPUs') + instance_pcpus = inst.numa_topology.cpu_pinning + self._assert_cpu_set_state(instance_pcpus, expected=expected) + return instance_pcpus + + def _assert_cpu_set_state(self, cpu_set, expected='online'): + for i in cpu_set: + core = cpu_api.Core(i) + if expected == 'online': + self.assertTrue(core.online, f'{i} is not online') + elif expected == 'offline': + self.assertFalse(core.online, f'{i} is online') + elif expected == 'powersave': + self.assertEqual('powersave', core.governor) + elif expected == 'performance': + self.assertEqual('performance', core.governor) + + +class PowerManagementTests(PowerManagementTestsBase): + """Test suite for a single host with 9 dedicated cores and 1 used for OS""" + + def setUp(self): + super(PowerManagementTests, self).setUp() + + self.useFixture(nova_fixtures.SysFileSystemFixture()) + + # Definining the CPUs to be pinned. + self.flags(cpu_dedicated_set='1-9', cpu_shared_set=None, + group='compute') + self.flags(vcpu_pin_set=None) + self.flags(cpu_power_management=True, group='libvirt') + + self.flags(allow_resize_to_same_host=True) + self.host_info = fakelibvirt.HostInfo(cpu_nodes=1, cpu_sockets=1, + cpu_cores=5, cpu_threads=2) + self.compute1 = self.start_compute(host_info=self.host_info, + hostname='compute1') + + # All cores are shutdown at startup, let's check. + cpu_dedicated_set = hardware.get_cpu_dedicated_set() + self._assert_cpu_set_state(cpu_dedicated_set, expected='offline') + + def test_hardstop_compute_service_if_wrong_opt(self): + self.flags(cpu_dedicated_set=None, cpu_shared_set=None, + group='compute') + self.flags(vcpu_pin_set=None) + self.flags(cpu_power_management=True, group='libvirt') + self.assertRaises(exception.InvalidConfiguration, + self.start_compute, host_info=self.host_info, + hostname='compute2') + + def test_create_server(self): + server = self._create_server( + flavor_id=self.pcpu_flavor_id, + expected_state='ACTIVE') + # Let's verify that the pinned CPUs are now online + self._assert_server_cpus_state(server, expected='online') + + # Verify that the unused CPUs are still offline + inst = objects.Instance.get_by_uuid(self.ctxt, server['id']) + instance_pcpus = inst.numa_topology.cpu_pinning + cpu_dedicated_set = hardware.get_cpu_dedicated_set() + unused_cpus = cpu_dedicated_set - instance_pcpus + self._assert_cpu_set_state(unused_cpus, expected='offline') + + def test_stop_start_server(self): + server = self._create_server( + flavor_id=self.pcpu_flavor_id, + expected_state='ACTIVE') + + server = self._stop_server(server) + # Let's verify that the pinned CPUs are now stopped... + self._assert_server_cpus_state(server, expected='offline') + + server = self._start_server(server) + # ...and now, they should be back. + self._assert_server_cpus_state(server, expected='online') + + def test_resize(self): + server = self._create_server( + flavor_id=self.pcpu_flavor_id, + expected_state='ACTIVE') + server_pcpus = self._assert_server_cpus_state(server, + expected='online') + + new_flavor_id = self._create_flavor( + vcpu=5, extra_spec=self.extra_spec) + self._resize_server(server, new_flavor_id) + server2_pcpus = self._assert_server_cpus_state(server, + expected='online') + # Even if the resize is not confirmed yet, the original guest is now + # destroyed so the cores are now offline. + self._assert_cpu_set_state(server_pcpus, expected='offline') + + # let's revert the resize + self._revert_resize(server) + # So now the original CPUs will be online again, while the previous + # cores should be back offline. + self._assert_cpu_set_state(server_pcpus, expected='online') + self._assert_cpu_set_state(server2_pcpus, expected='offline') + + def test_changing_strategy_fails(self): + # As a reminder, all cores have been shutdown before. + # Now we want to change the strategy and then we restart the service + self.flags(cpu_power_management_strategy='governor', group='libvirt') + # See, this is not possible as we would have offline CPUs. + self.assertRaises(exception.InvalidConfiguration, + self.restart_compute_service, hostname='compute1') + + +class PowerManagementTestsGovernor(PowerManagementTestsBase): + """Test suite for speific governor usage (same 10-core host)""" + + def setUp(self): + super(PowerManagementTestsGovernor, self).setUp() + + self.useFixture(nova_fixtures.SysFileSystemFixture()) + + # Definining the CPUs to be pinned. + self.flags(cpu_dedicated_set='1-9', cpu_shared_set=None, + group='compute') + self.flags(vcpu_pin_set=None) + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_power_management_strategy='governor', group='libvirt') + + self.flags(allow_resize_to_same_host=True) + self.host_info = fakelibvirt.HostInfo(cpu_nodes=1, cpu_sockets=1, + cpu_cores=5, cpu_threads=2) + self.compute1 = self.start_compute(host_info=self.host_info, + hostname='compute1') + + def test_create(self): + cpu_dedicated_set = hardware.get_cpu_dedicated_set() + # With the governor strategy, cores are still online but run with a + # powersave governor. + self._assert_cpu_set_state(cpu_dedicated_set, expected='powersave') + + # Now, start an instance + server = self._create_server( + flavor_id=self.pcpu_flavor_id, + expected_state='ACTIVE') + # When pinned cores are run, the governor state is now performance + self._assert_server_cpus_state(server, expected='performance') + + def test_changing_strategy_fails(self): + # Arbitratly set a core governor strategy to be performance + cpu_api.Core(1).set_high_governor() + # and then forget about it while changing the strategy. + self.flags(cpu_power_management_strategy='cpu_state', group='libvirt') + # This time, this wouldn't be acceptable as some core would have a + # difference performance while Nova would only online/offline it. + self.assertRaises(exception.InvalidConfiguration, + self.restart_compute_service, hostname='compute1') + + +class PowerManagementMixedInstances(PowerManagementTestsBase): + """Test suite for a single host with 6 dedicated cores, 3 shared and one + OS-restricted. + """ + + def setUp(self): + super(PowerManagementMixedInstances, self).setUp() + + self.useFixture(nova_fixtures.SysFileSystemFixture()) + + # Definining 6 CPUs to be dedicated, not all of them in a series. + self.flags(cpu_dedicated_set='1-3,5-7', cpu_shared_set='4,8-9', + group='compute') + self.flags(vcpu_pin_set=None) + self.flags(cpu_power_management=True, group='libvirt') + + self.host_info = fakelibvirt.HostInfo(cpu_nodes=1, cpu_sockets=1, + cpu_cores=5, cpu_threads=2) + self.compute1 = self.start_compute(host_info=self.host_info, + hostname='compute1') + + # Make sure only 6 are offline now + cpu_dedicated_set = hardware.get_cpu_dedicated_set() + self._assert_cpu_set_state(cpu_dedicated_set, expected='offline') + + # cores 4 and 8-9 should be online + self._assert_cpu_set_state({4, 8, 9}, expected='online') + + def test_standard_server_works_and_passes(self): + + std_flavor_id = self._create_flavor(vcpu=2) + self._create_server(flavor_id=std_flavor_id, expected_state='ACTIVE') + + # Since this is an instance with floating vCPUs on the shared set, we + # can only lookup the host CPUs and see they haven't changed state. + cpu_dedicated_set = hardware.get_cpu_dedicated_set() + self._assert_cpu_set_state(cpu_dedicated_set, expected='offline') + self._assert_cpu_set_state({4, 8, 9}, expected='online') + + # We can now try to boot an instance with pinned CPUs to test the mix + pinned_server = self._create_server( + flavor_id=self.pcpu_flavor_id, + expected_state='ACTIVE') + # We'll see that its CPUs are now online + self._assert_server_cpus_state(pinned_server, expected='online') + # but it doesn't change the shared set + self._assert_cpu_set_state({4, 8, 9}, expected='online') diff --git a/nova/tests/functional/test_servers.py b/nova/tests/functional/test_servers.py index 43208aa812..5887c99081 100644 --- a/nova/tests/functional/test_servers.py +++ b/nova/tests/functional/test_servers.py @@ -6526,3 +6526,41 @@ class PortAndFlavorAccelsServerCreateTest(AcceleratorServerBase): binding_profile = neutronapi.get_binding_profile(updated_port) self.assertNotIn('arq_uuid', binding_profile) self.assertNotIn('pci_slot', binding_profile) + + +class PortBindingShelvedServerTest(integrated_helpers._IntegratedTestBase): + """Tests for servers with ports.""" + + compute_driver = 'fake.SmallFakeDriver' + + def setUp(self): + super(PortBindingShelvedServerTest, self).setUp() + self.flavor_id = self._create_flavor( + disk=10, ephemeral=20, swap=5 * 1024) + + def test_shelve_offload_with_port(self): + # Do not wait before offloading + self.flags(shelved_offload_time=0) + + server = self._create_server( + flavor_id=self.flavor_id, + networks=[{'port': self.neutron.port_1['id']}]) + + port = self.neutron.show_port(self.neutron.port_1['id'])['port'] + + # Assert that the port is actually associated to the instance + self.assertEqual(port['device_id'], server['id']) + self.assertEqual(port['binding:host_id'], 'compute') + self.assertEqual(port['binding:status'], 'ACTIVE') + + # Do shelve + server = self._shelve_server(server, 'SHELVED_OFFLOADED') + + # Retrieve the updated port + port = self.neutron.show_port(self.neutron.port_1['id'])['port'] + + # Assert that the port is still associated to the instance + # but the binding is not on the compute anymore + self.assertEqual(port['device_id'], server['id']) + self.assertIsNone(port['binding:host_id']) + self.assertNotIn('binding:status', port) diff --git a/nova/tests/functional/test_service.py b/nova/tests/functional/test_service.py index 65b41594bd..21e9a519ee 100644 --- a/nova/tests/functional/test_service.py +++ b/nova/tests/functional/test_service.py @@ -10,8 +10,12 @@ # License for the specific language governing permissions and limitations # under the License. +import functools from unittest import mock +import fixtures +from oslo_utils.fixture import uuidsentinel as uuids + from nova import context as nova_context from nova import exception from nova.objects import service @@ -19,6 +23,7 @@ from nova import test from nova.tests import fixtures as nova_fixtures from nova.tests.functional import fixtures as func_fixtures from nova.tests.functional import integrated_helpers +from nova.virt import node class ServiceTestCase(test.TestCase, @@ -137,3 +142,83 @@ class TestOldComputeCheck( return_value=old_version): self.assertRaises( exception.TooOldComputeService, self._start_compute, 'host1') + + +class TestComputeStartupChecks(test.TestCase): + STUB_COMPUTE_ID = False + + def setUp(self): + super().setUp() + self.useFixture(nova_fixtures.RealPolicyFixture()) + self.useFixture(nova_fixtures.NeutronFixture(self)) + self.useFixture(nova_fixtures.GlanceFixture(self)) + self.useFixture(func_fixtures.PlacementFixture()) + + self._local_uuid = str(uuids.node) + + self.useFixture(fixtures.MockPatch( + 'nova.virt.node.get_local_node_uuid', + functools.partial(self.local_uuid, True))) + self.useFixture(fixtures.MockPatch( + 'nova.virt.node.read_local_node_uuid', + self.local_uuid)) + self.useFixture(fixtures.MockPatch( + 'nova.virt.node.write_local_node_uuid', + mock.DEFAULT)) + self.flags(compute_driver='fake.FakeDriverWithoutFakeNodes') + + def local_uuid(self, get=False): + if get and not self._local_uuid: + # Simulate the get_local_node_uuid behavior of calling write once + self._local_uuid = str(uuids.node) + node.write_local_node_uuid(self._local_uuid) + return self._local_uuid + + def test_compute_node_identity_greenfield(self): + # Level-set test case to show that starting and re-starting without + # any error cases works as expected. + + # Start with no local compute_id + self._local_uuid = None + self.start_service('compute') + + # Start should have generated and written a compute id + node.write_local_node_uuid.assert_called_once_with(str(uuids.node)) + + # Starting again should succeed and not cause another write + self.start_service('compute') + node.write_local_node_uuid.assert_called_once_with(str(uuids.node)) + + def test_compute_node_identity_deleted(self): + self.start_service('compute') + + # Simulate the compute_id file being deleted + self._local_uuid = None + + # Should refuse to start because it's not our first time and the file + # being missing is a hard error. + exc = self.assertRaises(exception.InvalidConfiguration, + self.start_service, 'compute') + self.assertIn('lost that state', str(exc)) + + def test_compute_node_hostname_changed(self): + # Start our compute once to create the node record + self.start_service('compute') + + # Starting with a different hostname should trigger the abort + exc = self.assertRaises(exception.InvalidConfiguration, + self.start_service, 'compute', host='other') + self.assertIn('hypervisor_hostname', str(exc)) + + def test_compute_node_uuid_changed(self): + # Start our compute once to create the node record + self.start_service('compute') + + # Simulate a changed local compute_id file + self._local_uuid = str(uuids.othernode) + + # We should fail to create the compute node record again, but with a + # useful error message about why. + exc = self.assertRaises(exception.InvalidConfiguration, + self.start_service, 'compute') + self.assertIn('Duplicate compute node record', str(exc)) diff --git a/nova/tests/unit/compute/test_compute_mgr.py b/nova/tests/unit/compute/test_compute_mgr.py index c923b0e967..1c69cd8f1c 100644 --- a/nova/tests/unit/compute/test_compute_mgr.py +++ b/nova/tests/unit/compute/test_compute_mgr.py @@ -91,6 +91,7 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase, # os-brick>=5.1 now uses external file system locks instead of internal # locks so we need to set up locking REQUIRES_LOCKING = True + STUB_COMPUTE_ID = False def setUp(self): super(ComputeManagerUnitTestCase, self).setUp() @@ -1162,24 +1163,48 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase, mock_get_nodes.return_value.keys()) @mock.patch.object(objects.ComputeNodeList, 'get_all_by_uuids') - @mock.patch.object(fake_driver.FakeDriver, 'get_available_nodes') + @mock.patch.object(fake_driver.FakeDriver, 'get_nodenames_by_uuid') def test_get_nodes(self, mock_driver_get_nodes, mock_get_by_uuid): - mock_driver_get_nodes.return_value = ['fake_node1', 'fake_node2'] + mock_driver_get_nodes.return_value = {uuids.node_fake_node1: 'host', + uuids.node_fake_node2: 'host'} # NOTE(danms): The fake driver, by default, uses - # uuidsentinel.$node_name, so we can predict the uuids it will + # uuidsentinel.node_$node_name, so we can predict the uuids it will # return here. - cn1 = objects.ComputeNode(uuid=uuids.fake_node1) - cn2 = objects.ComputeNode(uuid=uuids.fake_node2) + cn1 = objects.ComputeNode(uuid=uuids.node_fake_node1, + hypervisor_hostname='host') + cn2 = objects.ComputeNode(uuid=uuids.node_fake_node2, + hypervisor_hostname='host') mock_get_by_uuid.return_value = [cn1, cn2] nodes = self.compute._get_nodes(self.context) - self.assertEqual({uuids.fake_node1: cn1, uuids.fake_node2: cn2}, nodes) + self.assertEqual({uuids.node_fake_node1: cn1, + uuids.node_fake_node2: cn2}, nodes) mock_driver_get_nodes.assert_called_once_with() mock_get_by_uuid.assert_called_once_with(self.context, - [uuids.fake_node1, - uuids.fake_node2]) + [uuids.node_fake_node1, + uuids.node_fake_node2]) + + @mock.patch.object(objects.ComputeNodeList, 'get_all_by_uuids') + @mock.patch.object(fake_driver.FakeDriver, 'get_nodenames_by_uuid') + def test_get_nodes_mismatch(self, mock_driver_get_nodes, mock_get_by_uuid): + # Virt driver reports a (hypervisor_) hostname of 'host1' + mock_driver_get_nodes.return_value = {uuids.node_fake_node1: 'host1', + uuids.node_fake_node2: 'host1'} + + # The database records for our compute nodes (by UUID) show a + # hypervisor_hostname of 'host2' + cn1 = objects.ComputeNode(uuid=uuids.node_fake_node1, + hypervisor_hostname='host2') + cn2 = objects.ComputeNode(uuid=uuids.node_fake_node2, + hypervisor_hostname='host2') + mock_get_by_uuid.return_value = [cn1, cn2] + + # Possible hostname (as reported by the virt driver) rename, + # which should abort our startup + self.assertRaises(exception.InvalidConfiguration, + self.compute._get_nodes, self.context) @mock.patch.object(manager.LOG, 'warning') @mock.patch.object( @@ -1202,11 +1227,11 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase, @mock.patch.object(manager.LOG, 'warning') @mock.patch.object(objects.ComputeNodeList, 'get_all_by_uuids') - @mock.patch.object(fake_driver.FakeDriver, 'get_available_node_uuids') + @mock.patch.object(fake_driver.FakeDriver, 'get_nodenames_by_uuid') def test_get_nodes_node_not_found( self, mock_driver_get_nodes, mock_get_all_by_uuids, mock_log_warning): - mock_driver_get_nodes.return_value = ['fake-node1'] + mock_driver_get_nodes.return_value = {uuids.node_1: 'fake-node1'} mock_get_all_by_uuids.return_value = [] nodes = self.compute._get_nodes(self.context) @@ -1215,11 +1240,11 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase, mock_driver_get_nodes.assert_called_once_with() mock_get_all_by_uuids.assert_called_once_with(self.context, - ['fake-node1']) + [uuids.node_1]) mock_log_warning.assert_called_once_with( "Compute nodes %s for host %s were not found in the database. " "If this is the first time this service is starting on this host, " - "then you can ignore this warning.", ['fake-node1'], 'fake-mini') + "then you can ignore this warning.", [uuids.node_1], 'fake-mini') def test_init_host_disk_devices_configuration_failure(self): self.flags(max_disk_devices_to_attach=0, group='compute') @@ -6337,13 +6362,15 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase, 'two-image': 'existing'}, r) @mock.patch.object(virt_node, 'write_local_node_uuid') - def test_ensure_node_uuid_not_needed_version(self, mock_node): + @mock.patch.object(virt_node, 'read_local_node_uuid') + def test_ensure_node_uuid_not_needed_version(self, mock_read, mock_write): # Make sure an up-to-date service bypasses the persistence service_ref = service_obj.Service() self.assertEqual(service_obj.SERVICE_VERSION, service_ref.version) - mock_node.assert_not_called() + mock_read.return_value = 'not none' + mock_write.assert_not_called() self.compute._ensure_existing_node_identity(service_ref) - mock_node.assert_not_called() + mock_write.assert_not_called() @mock.patch.object(virt_node, 'write_local_node_uuid') def test_ensure_node_uuid_not_needed_ironic(self, mock_node): @@ -6428,6 +6455,20 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase, mock_get_cn.assert_called_once_with(mock.ANY, self.compute.host) mock_write_node.assert_called_once_with(str(uuids.compute)) + @mock.patch.object(virt_node, 'read_local_node_uuid') + def test_ensure_node_uuid_missing_file_ironic(self, mock_read): + mock_service = mock.MagicMock( + version=service_obj.NODE_IDENTITY_VERSION) + mock_read.return_value = None + self.assertRaises(exception.InvalidConfiguration, + self.compute._ensure_existing_node_identity, + mock_service) + mock_read.assert_called_once_with() + + # Now make sure that ironic causes this exact configuration to pass + self.flags(compute_driver='ironic') + self.compute._ensure_existing_node_identity(mock_service) + def test_ensure_node_uuid_called_by_init_host(self): # test_init_host() above ensures that we do not call # _ensure_existing_node_identity() in the service_ref=None case. @@ -9607,9 +9648,15 @@ class ComputeManagerMigrationTestCase(test.NoDBTestCase, self.assertEqual(driver_console.get_connection_info.return_value, console) + @mock.patch('nova.utils.pass_context') @mock.patch('nova.compute.manager.ComputeManager.' '_do_live_migration') - def _test_max_concurrent_live(self, mock_lm): + def _test_max_concurrent_live(self, mock_lm, mock_pass_context): + # pass_context wraps the function, which doesn't work with a mock + # So we simply mock it too + def _mock_pass_context(runner, func, *args, **kwargs): + return runner(func, *args, **kwargs) + mock_pass_context.side_effect = _mock_pass_context @mock.patch('nova.objects.Migration.save') def _do_it(mock_mig_save): diff --git a/nova/tests/unit/compute/test_resource_tracker.py b/nova/tests/unit/compute/test_resource_tracker.py index dfea323a9a..cd36b8987f 100644 --- a/nova/tests/unit/compute/test_resource_tracker.py +++ b/nova/tests/unit/compute/test_resource_tracker.py @@ -1552,6 +1552,20 @@ class TestInitComputeNode(BaseTestCase): self.assertEqual('fake-host', node.host) mock_update.assert_called() + @mock.patch.object(resource_tracker.ResourceTracker, + '_get_compute_node', + return_value=None) + @mock.patch('nova.objects.compute_node.ComputeNode.create') + def test_create_failed_conflict(self, mock_create, mock_getcn): + self._setup_rt() + resources = {'hypervisor_hostname': 'node1', + 'uuid': uuids.node1} + mock_create.side_effect = exc.DuplicateRecord(target='foo') + self.assertRaises(exc.InvalidConfiguration, + self.rt._init_compute_node, + mock.MagicMock, + resources) + @ddt.ddt class TestUpdateComputeNode(BaseTestCase): diff --git a/nova/tests/unit/compute/test_shelve.py b/nova/tests/unit/compute/test_shelve.py index f95a722ced..0a1e3f54fc 100644 --- a/nova/tests/unit/compute/test_shelve.py +++ b/nova/tests/unit/compute/test_shelve.py @@ -209,6 +209,7 @@ class ShelveComputeManagerTestCase(test_compute.BaseTestCase): instance = self._shelve_offload(clean_shutdown=False) mock_power_off.assert_called_once_with(instance, 0, 0) + @mock.patch.object(neutron_api.API, 'unbind_ports') @mock.patch.object(compute_utils, 'EventReporter') @mock.patch.object(objects.BlockDeviceMappingList, 'get_by_instance_uuid') @mock.patch.object(nova.compute.manager.ComputeManager, @@ -225,7 +226,7 @@ class ShelveComputeManagerTestCase(test_compute.BaseTestCase): def _shelve_offload(self, mock_notify, mock_notify_instance_usage, mock_get_power_state, mock_update_resource_tracker, mock_delete_alloc, mock_terminate, mock_get_bdms, - mock_event, clean_shutdown=True): + mock_event, mock_unbind_ports, clean_shutdown=True): host = 'fake-mini' instance = self._create_fake_instance_obj(params={'host': host}) instance.task_state = task_states.SHELVING @@ -278,6 +279,9 @@ class ShelveComputeManagerTestCase(test_compute.BaseTestCase): instance.uuid, graceful_exit=False) + mock_unbind_ports.assert_called_once_with( + self.context, mock.ANY, detach=False) + return instance @mock.patch('nova.compute.utils.' diff --git a/nova/tests/unit/objects/test_compute_node.py b/nova/tests/unit/objects/test_compute_node.py index 63b070c543..84c4e87785 100644 --- a/nova/tests/unit/objects/test_compute_node.py +++ b/nova/tests/unit/objects/test_compute_node.py @@ -16,6 +16,7 @@ import copy from unittest import mock import netaddr +from oslo_db import exception as db_exc from oslo_serialization import jsonutils from oslo_utils.fixture import uuidsentinel from oslo_utils import timeutils @@ -341,6 +342,14 @@ class _TestComputeNodeObject(object): 'uuid': uuidsentinel.fake_compute_node} mock_create.assert_called_once_with(self.context, param_dict) + @mock.patch('nova.db.main.api.compute_node_create') + def test_create_duplicate(self, mock_create): + mock_create.side_effect = db_exc.DBDuplicateEntry + compute = compute_node.ComputeNode(context=self.context) + compute.service_id = 456 + compute.hypervisor_hostname = 'node1' + self.assertRaises(exception.DuplicateRecord, compute.create) + @mock.patch.object(db, 'compute_node_update') @mock.patch( 'nova.db.main.api.compute_node_get', return_value=fake_compute_node) diff --git a/nova/tests/unit/scheduler/test_manager.py b/nova/tests/unit/scheduler/test_manager.py index e7866069b3..e992fe6034 100644 --- a/nova/tests/unit/scheduler/test_manager.py +++ b/nova/tests/unit/scheduler/test_manager.py @@ -19,6 +19,7 @@ Tests For Scheduler from unittest import mock +from keystoneauth1 import exceptions as ks_exc import oslo_messaging as messaging from oslo_serialization import jsonutils from oslo_utils.fixture import uuidsentinel as uuids @@ -1688,6 +1689,41 @@ class SchedulerManagerTestCase(test.NoDBTestCase): mock_log_warning.assert_not_called() mock_log_debug.assert_called_once_with(msg) + @mock.patch('nova.scheduler.client.report.report_client_singleton') + @mock.patch.object(manager, 'LOG') + @mock.patch('nova.scheduler.host_manager.HostManager') + @mock.patch('nova.servicegroup.API') + @mock.patch('nova.rpc.get_notifier') + def test_init_lazy_placement_client(self, mock_rpc, mock_sg, mock_hm, + mock_log, mock_report): + # Simulate keytone or placement being offline at startup + mock_report.side_effect = ks_exc.RequestTimeout + mgr = manager.SchedulerManager() + mock_report.assert_called_once_with() + self.assertTrue(mock_log.warning.called) + + # Make sure we're raising the actual error to subsequent callers + self.assertRaises(ks_exc.RequestTimeout, lambda: mgr.placement_client) + + # Simulate recovery of the keystone or placement service + mock_report.reset_mock(side_effect=True) + mgr.placement_client + mock_report.assert_called_once_with() + + @mock.patch('nova.scheduler.client.report.report_client_singleton') + @mock.patch('nova.scheduler.host_manager.HostManager') + @mock.patch('nova.servicegroup.API') + @mock.patch('nova.rpc.get_notifier') + def test_init_lazy_placement_client_failures(self, mock_rpc, mock_sg, + mock_hm, mock_report): + # Certain keystoneclient exceptions are fatal + mock_report.side_effect = ks_exc.Unauthorized + self.assertRaises(ks_exc.Unauthorized, manager.SchedulerManager) + + # Anything else is fatal + mock_report.side_effect = test.TestingException + self.assertRaises(test.TestingException, manager.SchedulerManager) + class SchedulerManagerAllocationCandidateTestCase(test.NoDBTestCase): diff --git a/nova/tests/unit/test_filesystem.py b/nova/tests/unit/test_filesystem.py new file mode 100644 index 0000000000..85f16157ee --- /dev/null +++ b/nova/tests/unit/test_filesystem.py @@ -0,0 +1,52 @@ +# 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. + +import os +from unittest import mock + +from nova import exception +from nova import filesystem +from nova import test + + +class TestFSCommon(test.NoDBTestCase): + + def test_read_sys(self): + open_mock = mock.mock_open(read_data='bar') + with mock.patch('builtins.open', open_mock) as m_open: + self.assertEqual('bar', filesystem.read_sys('foo')) + expected_path = os.path.join(filesystem.SYS, 'foo') + m_open.assert_called_once_with(expected_path, mode='r') + + def test_read_sys_error(self): + with mock.patch('builtins.open', + side_effect=OSError('error')) as m_open: + self.assertRaises(exception.FileNotFound, + filesystem.read_sys, 'foo') + expected_path = os.path.join(filesystem.SYS, 'foo') + m_open.assert_called_once_with(expected_path, mode='r') + + def test_write_sys(self): + open_mock = mock.mock_open() + with mock.patch('builtins.open', open_mock) as m_open: + self.assertIsNone(filesystem.write_sys('foo', 'bar')) + expected_path = os.path.join(filesystem.SYS, 'foo') + m_open.assert_called_once_with(expected_path, mode='w') + open_mock().write.assert_called_once_with('bar') + + def test_write_sys_error(self): + with mock.patch('builtins.open', + side_effect=OSError('fake_error')) as m_open: + self.assertRaises(exception.FileNotFound, + filesystem.write_sys, 'foo', 'bar') + expected_path = os.path.join(filesystem.SYS, 'foo') + m_open.assert_called_once_with(expected_path, mode='w') diff --git a/nova/tests/unit/virt/libvirt/cpu/__init__.py b/nova/tests/unit/virt/libvirt/cpu/__init__.py new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/nova/tests/unit/virt/libvirt/cpu/__init__.py diff --git a/nova/tests/unit/virt/libvirt/cpu/test_api.py b/nova/tests/unit/virt/libvirt/cpu/test_api.py new file mode 100644 index 0000000000..b5bcb762f3 --- /dev/null +++ b/nova/tests/unit/virt/libvirt/cpu/test_api.py @@ -0,0 +1,194 @@ +# 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. + +from unittest import mock + +from nova import exception +from nova import objects +from nova import test +from nova.virt.libvirt.cpu import api +from nova.virt.libvirt.cpu import core + + +class TestAPI(test.NoDBTestCase): + + def setUp(self): + super(TestAPI, self).setUp() + self.core_1 = api.Core(1) + + # Create a fake instance with two pinned CPUs but only one is on the + # dedicated set + numa_topology = objects.InstanceNUMATopology(cells=[ + objects.InstanceNUMACell(cpu_pinning_raw={'0': '0', '2': '2'}), + ]) + self.fake_inst = objects.Instance(numa_topology=numa_topology) + + @mock.patch.object(core, 'get_online') + def test_online(self, mock_get_online): + mock_get_online.return_value = True + self.assertTrue(self.core_1.online) + mock_get_online.assert_called_once_with(self.core_1.ident) + + @mock.patch.object(core, 'set_online') + def test_set_online(self, mock_set_online): + self.core_1.online = True + mock_set_online.assert_called_once_with(self.core_1.ident) + + @mock.patch.object(core, 'set_offline') + def test_set_offline(self, mock_set_offline): + self.core_1.online = False + mock_set_offline.assert_called_once_with(self.core_1.ident) + + def test_hash(self): + self.assertEqual(hash(self.core_1.ident), hash(self.core_1)) + + @mock.patch.object(core, 'get_governor') + def test_governor(self, mock_get_governor): + mock_get_governor.return_value = 'fake_governor' + self.assertEqual('fake_governor', self.core_1.governor) + mock_get_governor.assert_called_once_with(self.core_1.ident) + + @mock.patch.object(core, 'set_governor') + def test_set_governor_low(self, mock_set_governor): + self.flags(cpu_power_governor_low='fake_low_gov', group='libvirt') + self.core_1.set_low_governor() + mock_set_governor.assert_called_once_with(self.core_1.ident, + 'fake_low_gov') + + @mock.patch.object(core, 'set_governor') + def test_set_governor_high(self, mock_set_governor): + self.flags(cpu_power_governor_high='fake_high_gov', group='libvirt') + self.core_1.set_high_governor() + mock_set_governor.assert_called_once_with(self.core_1.ident, + 'fake_high_gov') + + @mock.patch.object(core, 'set_online') + def test_power_up_online(self, mock_online): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + + api.power_up(self.fake_inst) + # only core #0 can be set as core #2 is not on the dedicated set + # As a reminder, core(i).online calls set_online(i) + mock_online.assert_called_once_with(0) + + @mock.patch.object(core, 'set_governor') + def test_power_up_governor(self, mock_set_governor): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_power_management_strategy='governor', group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + + api.power_up(self.fake_inst) + # only core #0 can be set as core #2 is not on the dedicated set + # As a reminder, core(i).set_high_governor calls set_governor(i) + mock_set_governor.assert_called_once_with(0, 'performance') + + @mock.patch.object(core, 'set_online') + def test_power_up_skipped(self, mock_online): + self.flags(cpu_power_management=False, group='libvirt') + api.power_up(self.fake_inst) + mock_online.assert_not_called() + + @mock.patch.object(core, 'set_online') + def test_power_up_skipped_if_standard_instance(self, mock_online): + self.flags(cpu_power_management=True, group='libvirt') + api.power_up(objects.Instance(numa_topology=None)) + mock_online.assert_not_called() + + @mock.patch.object(core, 'set_offline') + def test_power_down_offline(self, mock_offline): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + + api.power_down(self.fake_inst) + # only core #0 can be set as core #2 is not on the dedicated set + # As a reminder, core(i).online calls set_online(i) + mock_offline.assert_called_once_with(0) + + @mock.patch.object(core, 'set_governor') + def test_power_down_governor(self, mock_set_governor): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_power_management_strategy='governor', group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + + api.power_down(self.fake_inst) + # only core #0 can be set as core #2 is not on the dedicated set + # As a reminder, core(i).set_high_governor calls set_governor(i) + mock_set_governor.assert_called_once_with(0, 'powersave') + + @mock.patch.object(core, 'set_offline') + def test_power_down_skipped(self, mock_offline): + self.flags(cpu_power_management=False, group='libvirt') + api.power_down(self.fake_inst) + mock_offline.assert_not_called() + + @mock.patch.object(core, 'set_offline') + def test_power_down_skipped_if_standard_instance(self, mock_offline): + self.flags(cpu_power_management=True, group='libvirt') + api.power_down(objects.Instance(numa_topology=None)) + mock_offline.assert_not_called() + + @mock.patch.object(core, 'set_offline') + def test_power_down_all_dedicated_cpus_offline(self, mock_offline): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + + api.power_down_all_dedicated_cpus() + # All dedicated CPUs are turned offline + mock_offline.assert_has_calls([mock.call(0), mock.call(1)]) + + @mock.patch.object(core, 'set_governor') + def test_power_down_all_dedicated_cpus_governor(self, mock_set_governor): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_power_management_strategy='governor', group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + + api.power_down_all_dedicated_cpus() + # All dedicated CPUs are turned offline + mock_set_governor.assert_has_calls([mock.call(0, 'powersave'), + mock.call(1, 'powersave')]) + + @mock.patch.object(core, 'set_offline') + def test_power_down_all_dedicated_cpus_skipped(self, mock_offline): + self.flags(cpu_power_management=False, group='libvirt') + api.power_down_all_dedicated_cpus() + mock_offline.assert_not_called() + + def test_power_down_all_dedicated_cpus_wrong_config(self): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_dedicated_set=None, group='compute') + self.assertRaises(exception.InvalidConfiguration, + api.power_down_all_dedicated_cpus) + + @mock.patch.object(core, 'get_governor') + @mock.patch.object(core, 'get_online') + def test_validate_all_dedicated_cpus_for_governor(self, mock_get_online, + mock_get_governor): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + self.flags(cpu_power_management_strategy='governor', group='libvirt') + mock_get_governor.return_value = 'performance' + mock_get_online.side_effect = (True, False) + self.assertRaises(exception.InvalidConfiguration, + api.validate_all_dedicated_cpus) + + @mock.patch.object(core, 'get_governor') + @mock.patch.object(core, 'get_online') + def test_validate_all_dedicated_cpus_for_cpu_state(self, mock_get_online, + mock_get_governor): + self.flags(cpu_power_management=True, group='libvirt') + self.flags(cpu_dedicated_set='0-1', group='compute') + self.flags(cpu_power_management_strategy='cpu_state', group='libvirt') + mock_get_online.return_value = True + mock_get_governor.side_effect = ('powersave', 'performance') + self.assertRaises(exception.InvalidConfiguration, + api.validate_all_dedicated_cpus) diff --git a/nova/tests/unit/virt/libvirt/cpu/test_core.py b/nova/tests/unit/virt/libvirt/cpu/test_core.py new file mode 100644 index 0000000000..a3cba00d3b --- /dev/null +++ b/nova/tests/unit/virt/libvirt/cpu/test_core.py @@ -0,0 +1,122 @@ +# 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. + +from unittest import mock + +from nova import exception +from nova import test +from nova.tests import fixtures +from nova.virt.libvirt.cpu import core + + +class TestCore(test.NoDBTestCase): + + @mock.patch.object(core.filesystem, 'read_sys') + @mock.patch.object(core.hardware, 'parse_cpu_spec') + def test_get_available_cores(self, mock_parse_cpu_spec, mock_read_sys): + mock_read_sys.return_value = '1-2' + mock_parse_cpu_spec.return_value = set([1, 2]) + self.assertEqual(set([1, 2]), core.get_available_cores()) + mock_read_sys.assert_called_once_with(core.AVAILABLE_PATH) + mock_parse_cpu_spec.assert_called_once_with('1-2') + + @mock.patch.object(core.filesystem, 'read_sys') + @mock.patch.object(core.hardware, 'parse_cpu_spec') + def test_get_available_cores_none( + self, mock_parse_cpu_spec, mock_read_sys): + mock_read_sys.return_value = '' + self.assertEqual(set(), core.get_available_cores()) + mock_parse_cpu_spec.assert_not_called() + + @mock.patch.object(core, 'get_available_cores') + def test_exists(self, mock_get_available_cores): + mock_get_available_cores.return_value = set([1]) + self.assertTrue(core.exists(1)) + mock_get_available_cores.assert_called_once_with() + self.assertFalse(core.exists(2)) + + @mock.patch.object( + core, 'CPU_PATH_TEMPLATE', + new_callable=mock.PropertyMock(return_value='/sys/blah%(core)s')) + @mock.patch.object(core, 'exists') + def test_gen_cpu_path(self, mock_exists, mock_cpu_path): + mock_exists.return_value = True + self.assertEqual('/sys/blah1', core.gen_cpu_path(1)) + mock_exists.assert_called_once_with(1) + + @mock.patch.object(core, 'exists') + def test_gen_cpu_path_raises(self, mock_exists): + mock_exists.return_value = False + self.assertRaises(ValueError, core.gen_cpu_path, 1) + self.assertIn('Unable to access CPU: 1', self.stdlog.logger.output) + + +class TestCoreHelpers(test.NoDBTestCase): + + def setUp(self): + super(TestCoreHelpers, self).setUp() + self.useFixture(fixtures.PrivsepFixture()) + _p1 = mock.patch.object(core, 'exists', return_value=True) + self.mock_exists = _p1.start() + self.addCleanup(_p1.stop) + + _p2 = mock.patch.object(core, 'gen_cpu_path', + side_effect=lambda x: '/fakesys/blah%s' % x) + self.mock_gen_cpu_path = _p2.start() + self.addCleanup(_p2.stop) + + @mock.patch.object(core.filesystem, 'read_sys') + def test_get_online(self, mock_read_sys): + mock_read_sys.return_value = '1' + self.assertTrue(core.get_online(1)) + mock_read_sys.assert_called_once_with('/fakesys/blah1/online') + + @mock.patch.object(core.filesystem, 'read_sys') + def test_get_online_not_exists(self, mock_read_sys): + mock_read_sys.side_effect = exception.FileNotFound(file_path='foo') + self.assertTrue(core.get_online(1)) + mock_read_sys.assert_called_once_with('/fakesys/blah1/online') + + @mock.patch.object(core.filesystem, 'write_sys') + @mock.patch.object(core, 'get_online') + def test_set_online(self, mock_get_online, mock_write_sys): + mock_get_online.return_value = True + self.assertTrue(core.set_online(1)) + mock_write_sys.assert_called_once_with('/fakesys/blah1/online', + data='1') + mock_get_online.assert_called_once_with(1) + + @mock.patch.object(core.filesystem, 'write_sys') + @mock.patch.object(core, 'get_online') + def test_set_offline(self, mock_get_online, mock_write_sys): + mock_get_online.return_value = False + self.assertTrue(core.set_offline(1)) + mock_write_sys.assert_called_once_with('/fakesys/blah1/online', + data='0') + mock_get_online.assert_called_once_with(1) + + @mock.patch.object(core.filesystem, 'read_sys') + def test_get_governor(self, mock_read_sys): + mock_read_sys.return_value = 'fake_gov' + self.assertEqual('fake_gov', core.get_governor(1)) + mock_read_sys.assert_called_once_with( + '/fakesys/blah1/cpufreq/scaling_governor') + + @mock.patch.object(core, 'get_governor') + @mock.patch.object(core.filesystem, 'write_sys') + def test_set_governor(self, mock_write_sys, mock_get_governor): + mock_get_governor.return_value = 'fake_gov' + self.assertEqual('fake_gov', + core.set_governor(1, 'fake_gov')) + mock_write_sys.assert_called_once_with( + '/fakesys/blah1/cpufreq/scaling_governor', data='fake_gov') + mock_get_governor.assert_called_once_with(1) diff --git a/nova/tests/unit/virt/libvirt/test_config.py b/nova/tests/unit/virt/libvirt/test_config.py index 8f840e8859..3d0b5ae685 100644 --- a/nova/tests/unit/virt/libvirt/test_config.py +++ b/nova/tests/unit/virt/libvirt/test_config.py @@ -1537,7 +1537,7 @@ class LibvirtConfigGuestInputTest(LibvirtConfigBaseTest): class LibvirtConfigGuestGraphicsTest(LibvirtConfigBaseTest): - def test_config_graphics(self): + def test_config_graphics_vnc(self): obj = config.LibvirtConfigGuestGraphics() obj.type = "vnc" obj.autoport = True @@ -1549,6 +1549,30 @@ class LibvirtConfigGuestGraphicsTest(LibvirtConfigBaseTest): <graphics type="vnc" autoport="yes" keymap="en_US" listen="127.0.0.1"/> """) + def test_config_graphics_spice(self): + obj = config.LibvirtConfigGuestGraphics() + obj.type = "spice" + obj.autoport = False + obj.keymap = "en_US" + obj.listen = "127.0.0.1" + + obj.image_compression = "auto_glz" + obj.jpeg_compression = "auto" + obj.zlib_compression = "always" + obj.playback_compression = True + obj.streaming_mode = "filter" + + xml = obj.to_xml() + self.assertXmlEqual(xml, """ + <graphics type="spice" autoport="no" keymap="en_US" listen="127.0.0.1"> + <image compression="auto_glz"/> + <jpeg compression="auto"/> + <zlib compression="always"/> + <playback compression="on"/> + <streaming mode="filter"/> + </graphics> + """) + class LibvirtConfigGuestHostdev(LibvirtConfigBaseTest): diff --git a/nova/tests/unit/virt/libvirt/test_driver.py b/nova/tests/unit/virt/libvirt/test_driver.py index e9b7a2133e..2b58c7df8b 100644 --- a/nova/tests/unit/virt/libvirt/test_driver.py +++ b/nova/tests/unit/virt/libvirt/test_driver.py @@ -3402,7 +3402,8 @@ class LibvirtConnTestCase(test.NoDBTestCase, self.assertEqual( "Memory encryption requested by hw:mem_encryption extra spec in " "m1.fake flavor but image fake_image doesn't have " - "'hw_firmware_type' property set to 'uefi'", str(exc)) + "'hw_firmware_type' property set to 'uefi' or volume-backed " + "instance was requested", str(exc)) def test_sev_enabled_host_extra_spec_no_machine_type(self): exc = self.assertRaises(exception.InvalidMachineType, @@ -5839,6 +5840,11 @@ class LibvirtConnTestCase(test.NoDBTestCase, self.assertEqual(cfg.devices[3].type, 'vnc') self.assertEqual(cfg.devices[3].listen, '10.0.0.1') self.assertIsNone(cfg.devices[3].keymap) + self.assertIsNone(cfg.devices[3].image_compression) + self.assertIsNone(cfg.devices[3].jpeg_compression) + self.assertIsNone(cfg.devices[3].zlib_compression) + self.assertIsNone(cfg.devices[3].playback_compression) + self.assertIsNone(cfg.devices[3].streaming_mode) def test_get_guest_config_with_vnc_and_tablet(self): self.flags(enabled=True, group='vnc') @@ -5869,6 +5875,11 @@ class LibvirtConnTestCase(test.NoDBTestCase, vconfig.LibvirtConfigMemoryBalloon) self.assertEqual(cfg.devices[3].type, 'vnc') + self.assertIsNone(cfg.devices[3].image_compression) + self.assertIsNone(cfg.devices[3].jpeg_compression) + self.assertIsNone(cfg.devices[3].zlib_compression) + self.assertIsNone(cfg.devices[3].playback_compression) + self.assertIsNone(cfg.devices[3].streaming_mode) self.assertEqual(cfg.devices[5].type, 'tablet') def test_get_guest_config_with_spice_and_tablet(self): @@ -5905,6 +5916,11 @@ class LibvirtConnTestCase(test.NoDBTestCase, self.assertEqual(cfg.devices[3].type, 'spice') self.assertEqual(cfg.devices[3].listen, '10.0.0.1') self.assertIsNone(cfg.devices[3].keymap) + self.assertIsNone(cfg.devices[3].image_compression) + self.assertIsNone(cfg.devices[3].jpeg_compression) + self.assertIsNone(cfg.devices[3].zlib_compression) + self.assertIsNone(cfg.devices[3].playback_compression) + self.assertIsNone(cfg.devices[3].streaming_mode) self.assertEqual(cfg.devices[5].type, 'tablet') @mock.patch.object(host.Host, "_check_machine_type", new=mock.Mock()) @@ -5964,8 +5980,57 @@ class LibvirtConnTestCase(test.NoDBTestCase, self.assertEqual(cfg.devices[3].target_name, "com.redhat.spice.0") self.assertEqual(cfg.devices[3].type, 'spicevmc') self.assertEqual(cfg.devices[4].type, "spice") + self.assertIsNone(cfg.devices[4].image_compression) + self.assertIsNone(cfg.devices[4].jpeg_compression) + self.assertIsNone(cfg.devices[4].zlib_compression) + self.assertIsNone(cfg.devices[4].playback_compression) + self.assertIsNone(cfg.devices[4].streaming_mode) self.assertEqual(cfg.devices[5].type, video_type) + def test_get_guest_config_with_spice_compression(self): + self.flags(enabled=False, group='vnc') + self.flags(virt_type='kvm', group='libvirt') + self.flags(enabled=True, + agent_enabled=False, + image_compression='auto_lz', + jpeg_compression='never', + zlib_compression='always', + playback_compression=False, + streaming_mode='all', + server_listen='10.0.0.1', + group='spice') + self.flags(pointer_model='usbtablet') + + cfg = self._get_guest_config_with_graphics() + + self.assertEqual(len(cfg.devices), 9) + self.assertIsInstance(cfg.devices[0], + vconfig.LibvirtConfigGuestDisk) + self.assertIsInstance(cfg.devices[1], + vconfig.LibvirtConfigGuestDisk) + self.assertIsInstance(cfg.devices[2], + vconfig.LibvirtConfigGuestSerial) + self.assertIsInstance(cfg.devices[3], + vconfig.LibvirtConfigGuestGraphics) + self.assertIsInstance(cfg.devices[4], + vconfig.LibvirtConfigGuestVideo) + self.assertIsInstance(cfg.devices[5], + vconfig.LibvirtConfigGuestInput) + self.assertIsInstance(cfg.devices[6], + vconfig.LibvirtConfigGuestRng) + self.assertIsInstance(cfg.devices[7], + vconfig.LibvirtConfigGuestUSBHostController) + self.assertIsInstance(cfg.devices[8], + vconfig.LibvirtConfigMemoryBalloon) + + self.assertEqual(cfg.devices[3].type, 'spice') + self.assertEqual(cfg.devices[3].listen, '10.0.0.1') + self.assertEqual(cfg.devices[3].image_compression, 'auto_lz') + self.assertEqual(cfg.devices[3].jpeg_compression, 'never') + self.assertEqual(cfg.devices[3].zlib_compression, 'always') + self.assertFalse(cfg.devices[3].playback_compression) + self.assertEqual(cfg.devices[3].streaming_mode, 'all') + @mock.patch.object(host.Host, 'get_guest') @mock.patch.object(libvirt_driver.LibvirtDriver, '_get_serial_ports_from_guest') @@ -9190,6 +9255,34 @@ class LibvirtConnTestCase(test.NoDBTestCase, drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False) self.assertRaises(exception.Invalid, drvr._get_pcpu_available) + @mock.patch('nova.virt.libvirt.host.Host.get_available_cpus', + return_value=set([0, 1, 2, 3])) + def test_get_pcpu_available_for_power_mgmt(self, get_available_cpus): + """Test what happens when the '[compute] cpu_dedicated_set' config + option is set and power management is defined. + """ + self.flags(vcpu_pin_set=None) + self.flags(cpu_dedicated_set='2-3', cpu_shared_set=None, + group='compute') + self.flags(cpu_power_management=True, group='libvirt') + drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False) + pcpus = drvr._get_pcpu_available() + self.assertEqual(set([2, 3]), pcpus) + + @mock.patch('nova.virt.libvirt.host.Host.get_available_cpus', + return_value=set([4, 5])) + def test_get_pcpu_available__cpu_dedicated_set_invalid_for_pm(self, + get_available_cpus): + """Test what happens when the '[compute] cpu_dedicated_set' config + option is set but it's invalid with power management set. + """ + self.flags(vcpu_pin_set=None) + self.flags(cpu_dedicated_set='4-6', cpu_shared_set=None, + group='compute') + self.flags(cpu_power_management=True, group='libvirt') + drvr = libvirt_driver.LibvirtDriver(fake.FakeVirtAPI(), False) + self.assertRaises(exception.Invalid, drvr._get_pcpu_available) + @mock.patch('nova.virt.libvirt.host.Host.get_online_cpus', return_value=set([0, 1, 2, 3])) def test_get_vcpu_available(self, get_online_cpus): diff --git a/nova/tests/unit/virt/libvirt/test_host.py b/nova/tests/unit/virt/libvirt/test_host.py index 3afd6c139d..631b10d81a 100644 --- a/nova/tests/unit/virt/libvirt/test_host.py +++ b/nova/tests/unit/virt/libvirt/test_host.py @@ -1052,6 +1052,12 @@ Active: 8381604 kB 'iowait': 6121490000000}, stats) + @mock.patch.object(fakelibvirt.virConnect, "getCPUMap") + def test_get_available_cpus(self, mock_map): + mock_map.return_value = (4, [True, True, False, False], None) + result = self.host.get_available_cpus() + self.assertEqual(result, {0, 1, 2, 3}) + @mock.patch.object(fakelibvirt.virConnect, "defineXML") def test_write_instance_config(self, mock_defineXML): fake_dom_xml = """ diff --git a/nova/tests/unit/virt/test_hardware.py b/nova/tests/unit/virt/test_hardware.py index 016c478f8c..753ee41550 100644 --- a/nova/tests/unit/virt/test_hardware.py +++ b/nova/tests/unit/virt/test_hardware.py @@ -5364,7 +5364,7 @@ class MemEncryptionRequestedWithoutUEFITestCase( expected_error = ( "Memory encryption requested by %(requesters)s but image " "%(image_name)s doesn't have 'hw_firmware_type' property " - "set to 'uefi'" + "set to 'uefi' or volume-backed instance was requested" ) def _test_encrypted_memory_support_no_uefi(self, enc_extra_spec, @@ -5491,6 +5491,25 @@ class MemEncryptionRequiredTestCase(test.NoDBTestCase): (self.flavor_name, self.image_id) ) + def test_encrypted_memory_support_flavor_for_volume(self): + extra_specs = {'hw:mem_encryption': True} + + flavor = objects.Flavor(name=self.flavor_name, + extra_specs=extra_specs) + # Following image_meta is typical for root Cinder volume + image_meta = objects.ImageMeta.from_dict({ + 'min_disk': 0, + 'min_ram': 0, + 'properties': {}, + 'size': 0, + 'status': 'active'}) + # Confirm that exception.FlavorImageConflict is raised when + # flavor with hw:mem_encryption flag is used to create + # volume-backed instance + self.assertRaises(exception.FlavorImageConflict, + hw.get_mem_encryption_constraint, flavor, + image_meta) + class PCINUMAAffinityPolicyTest(test.NoDBTestCase): diff --git a/nova/utils.py b/nova/utils.py index 664056a09f..b5d45c58b5 100644 --- a/nova/utils.py +++ b/nova/utils.py @@ -632,15 +632,13 @@ def _serialize_profile_info(): return trace_info -def spawn(func, *args, **kwargs): - """Passthrough method for eventlet.spawn. - - This utility exists so that it can be stubbed for testing without - interfering with the service spawns. +def pass_context(runner, func, *args, **kwargs): + """Generalised passthrough method - It will also grab the context from the threadlocal store and add it to - the store on the new thread. This allows for continuity in logging the - context when using this method to spawn a new thread. + It will grab the context from the threadlocal store and add it to + the store on the runner. This allows for continuity in logging the + context when using this method to spawn a new thread through the + runner function """ _context = common_context.get_current() profiler_info = _serialize_profile_info() @@ -655,11 +653,11 @@ def spawn(func, *args, **kwargs): profiler.init(**profiler_info) return func(*args, **kwargs) - return eventlet.spawn(context_wrapper, *args, **kwargs) + return runner(context_wrapper, *args, **kwargs) -def spawn_n(func, *args, **kwargs): - """Passthrough method for eventlet.spawn_n. +def spawn(func, *args, **kwargs): + """Passthrough method for eventlet.spawn. This utility exists so that it can be stubbed for testing without interfering with the service spawns. @@ -668,25 +666,26 @@ def spawn_n(func, *args, **kwargs): the store on the new thread. This allows for continuity in logging the context when using this method to spawn a new thread. """ - _context = common_context.get_current() - profiler_info = _serialize_profile_info() - @functools.wraps(func) - def context_wrapper(*args, **kwargs): - # NOTE: If update_store is not called after spawn_n it won't be - # available for the logger to pull from threadlocal storage. - if _context is not None: - _context.update_store() - if profiler_info and profiler: - profiler.init(**profiler_info) - func(*args, **kwargs) + return pass_context(eventlet.spawn, func, *args, **kwargs) + + +def spawn_n(func, *args, **kwargs): + """Passthrough method for eventlet.spawn_n. + + This utility exists so that it can be stubbed for testing without + interfering with the service spawns. - eventlet.spawn_n(context_wrapper, *args, **kwargs) + It will also grab the context from the threadlocal store and add it to + the store on the new thread. This allows for continuity in logging the + context when using this method to spawn a new thread. + """ + pass_context(eventlet.spawn_n, func, *args, **kwargs) def tpool_execute(func, *args, **kwargs): """Run func in a native thread""" - tpool.execute(func, *args, **kwargs) + return pass_context(tpool.execute, func, *args, **kwargs) def is_none_string(val): diff --git a/nova/virt/driver.py b/nova/virt/driver.py index b6297bb785..5d42a392d8 100644 --- a/nova/virt/driver.py +++ b/nova/virt/driver.py @@ -1596,8 +1596,10 @@ class ComputeDriver(object): """ raise NotImplementedError() - def get_available_node_uuids(self, refresh=False): - return [nova.virt.node.get_local_node_uuid()] + def get_nodenames_by_uuid(self, refresh=False): + """Returns a dict of {uuid: nodename} for all managed nodes.""" + nodename = self.get_available_nodes()[0] + return {nova.virt.node.get_local_node_uuid(): nodename} def node_is_available(self, nodename): """Return whether this compute service manages a particular node.""" diff --git a/nova/virt/fake.py b/nova/virt/fake.py index f86db8fcb2..bf7dc8fc72 100644 --- a/nova/virt/fake.py +++ b/nova/virt/fake.py @@ -49,6 +49,7 @@ from nova.objects import migrate_data from nova.virt import driver from nova.virt import hardware from nova.virt.ironic import driver as ironic +import nova.virt.node from nova.virt import virtapi CONF = nova.conf.CONF @@ -510,7 +511,7 @@ class FakeDriver(driver.ComputeDriver): # control our node uuids. Make sure we return a unique and # consistent uuid for each node we are responsible for to # avoid the persistent local node identity from taking over. - host_status['uuid'] = str(getattr(uuids, nodename)) + host_status['uuid'] = str(getattr(uuids, 'node_%s' % nodename)) return host_status def update_provider_tree(self, provider_tree, nodename, allocations=None): @@ -653,8 +654,9 @@ class FakeDriver(driver.ComputeDriver): def get_available_nodes(self, refresh=False): return self._nodes - def get_available_node_uuids(self, refresh=False): - return [str(getattr(uuids, n)) for n in self.get_available_nodes()] + def get_nodenames_by_uuid(self, refresh=False): + return {str(getattr(uuids, 'node_%s' % n)): n + for n in self.get_available_nodes()} def instance_on_disk(self, instance): return False @@ -1129,3 +1131,22 @@ class EphEncryptionDriverPLAIN(MediumFakeDriver): FakeDriver.capabilities, supports_ephemeral_encryption=True, supports_ephemeral_encryption_plain=True) + + +class FakeDriverWithoutFakeNodes(FakeDriver): + """FakeDriver that behaves like a real single-node driver. + + This behaves like a real virt driver from the perspective of its + nodes, with a stable nodename and use of the global node identity + stuff to provide a stable node UUID. + """ + + def get_available_resource(self, nodename): + resources = super().get_available_resource(nodename) + resources['uuid'] = nova.virt.node.get_local_node_uuid() + return resources + + def get_nodenames_by_uuid(self, refresh=False): + return { + nova.virt.node.get_local_node_uuid(): self.get_available_nodes()[0] + } diff --git a/nova/virt/hardware.py b/nova/virt/hardware.py index 96a7198db2..c8f8bb2481 100644 --- a/nova/virt/hardware.py +++ b/nova/virt/hardware.py @@ -1213,10 +1213,13 @@ def _check_for_mem_encryption_requirement_conflicts( "image %(image_name)s which has hw_mem_encryption property " "explicitly set to %(image_val)s" ) + # image_meta.name is not set if image object represents root + # Cinder volume. + image_name = (image_meta.name if 'name' in image_meta else None) data = { 'flavor_name': flavor.name, 'flavor_val': flavor_mem_enc_str, - 'image_name': image_meta.name, + 'image_name': image_name, 'image_val': image_mem_enc, } raise exception.FlavorImageConflict(emsg % data) @@ -1228,10 +1231,15 @@ def _check_mem_encryption_uses_uefi_image(requesters, image_meta): emsg = _( "Memory encryption requested by %(requesters)s but image " - "%(image_name)s doesn't have 'hw_firmware_type' property set to 'uefi'" + "%(image_name)s doesn't have 'hw_firmware_type' property set to " + "'uefi' or volume-backed instance was requested" ) + # image_meta.name is not set if image object represents root Cinder + # volume, for this case FlavorImageConflict should be raised, but + # image_meta.name can't be extracted. + image_name = (image_meta.name if 'name' in image_meta else None) data = {'requesters': " and ".join(requesters), - 'image_name': image_meta.name} + 'image_name': image_name} raise exception.FlavorImageConflict(emsg % data) @@ -1260,12 +1268,14 @@ def _check_mem_encryption_machine_type(image_meta, machine_type=None): if mach_type is None: return + # image_meta.name is not set if image object represents root Cinder volume. + image_name = (image_meta.name if 'name' in image_meta else None) # Could be something like pc-q35-2.11 if a specific version of the # machine type is required, so do substring matching. if 'q35' not in mach_type: raise exception.InvalidMachineType( mtype=mach_type, - image_id=image_meta.id, image_name=image_meta.name, + image_id=image_meta.id, image_name=image_name, reason=_("q35 type is required for SEV to work")) diff --git a/nova/virt/ironic/driver.py b/nova/virt/ironic/driver.py index b437b4a959..77fefb81ea 100644 --- a/nova/virt/ironic/driver.py +++ b/nova/virt/ironic/driver.py @@ -839,8 +839,12 @@ class IronicDriver(virt_driver.ComputeDriver): return node_uuids - def get_available_node_uuids(self, refresh=False): - return self.get_available_nodes(refresh=refresh) + def get_nodenames_by_uuid(self, refresh=False): + nodes = self.get_available_nodes(refresh=refresh) + # We use the uuid for compute_node.uuid and + # compute_node.hypervisor_hostname, so the dict keys and values are + # the same. + return dict(zip(nodes, nodes)) def update_provider_tree(self, provider_tree, nodename, allocations=None): """Update a ProviderTree object with current resource provider and diff --git a/nova/virt/libvirt/config.py b/nova/virt/libvirt/config.py index 0db2dc6b67..231283b8dd 100644 --- a/nova/virt/libvirt/config.py +++ b/nova/virt/libvirt/config.py @@ -2047,6 +2047,12 @@ class LibvirtConfigGuestGraphics(LibvirtConfigGuestDevice): self.keymap = None self.listen = None + self.image_compression = None + self.jpeg_compression = None + self.zlib_compression = None + self.playback_compression = None + self.streaming_mode = None + def format_dom(self): dev = super(LibvirtConfigGuestGraphics, self).format_dom() @@ -2057,6 +2063,24 @@ class LibvirtConfigGuestGraphics(LibvirtConfigGuestDevice): if self.listen: dev.set("listen", self.listen) + if self.type == "spice": + if self.image_compression is not None: + dev.append(etree.Element( + 'image', compression=self.image_compression)) + if self.jpeg_compression is not None: + dev.append(etree.Element( + 'jpeg', compression=self.jpeg_compression)) + if self.zlib_compression is not None: + dev.append(etree.Element( + 'zlib', compression=self.zlib_compression)) + if self.playback_compression is not None: + dev.append(etree.Element( + 'playback', compression=self.get_on_off_str( + self.playback_compression))) + if self.streaming_mode is not None: + dev.append(etree.Element( + 'streaming', mode=self.streaming_mode)) + return dev diff --git a/nova/virt/libvirt/cpu/__init__.py b/nova/virt/libvirt/cpu/__init__.py new file mode 100644 index 0000000000..e69de29bb2 --- /dev/null +++ b/nova/virt/libvirt/cpu/__init__.py diff --git a/nova/virt/libvirt/cpu/api.py b/nova/virt/libvirt/cpu/api.py new file mode 100644 index 0000000000..1c17458d6b --- /dev/null +++ b/nova/virt/libvirt/cpu/api.py @@ -0,0 +1,157 @@ +# 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. + +from dataclasses import dataclass + +from oslo_log import log as logging + +import nova.conf +from nova import exception +from nova.i18n import _ +from nova import objects +from nova.virt import hardware +from nova.virt.libvirt.cpu import core + +LOG = logging.getLogger(__name__) + +CONF = nova.conf.CONF + + +@dataclass +class Core: + """Class to model a CPU core as reported by sysfs. + + It may be a physical CPU core or a hardware thread on a shared CPU core + depending on if the system supports SMT. + """ + + # NOTE(sbauza): ident is a mandatory field. + # The CPU core id/number + ident: int + + @property + def online(self) -> bool: + return core.get_online(self.ident) + + @online.setter + def online(self, state: bool) -> None: + if state: + core.set_online(self.ident) + else: + core.set_offline(self.ident) + + def __hash__(self): + return hash(self.ident) + + def __eq__(self, other): + return self.ident == other.ident + + def __str__(self): + return str(self.ident) + + @property + def governor(self) -> str: + return core.get_governor(self.ident) + + def set_high_governor(self) -> None: + core.set_governor(self.ident, CONF.libvirt.cpu_power_governor_high) + + def set_low_governor(self) -> None: + core.set_governor(self.ident, CONF.libvirt.cpu_power_governor_low) + + +def power_up(instance: objects.Instance) -> None: + if not CONF.libvirt.cpu_power_management: + return + if instance.numa_topology is None: + return + + cpu_dedicated_set = hardware.get_cpu_dedicated_set() or set() + pcpus = instance.numa_topology.cpu_pinning + powered_up = set() + for pcpu in pcpus: + if pcpu in cpu_dedicated_set: + pcpu = Core(pcpu) + if CONF.libvirt.cpu_power_management_strategy == 'cpu_state': + pcpu.online = True + else: + pcpu.set_high_governor() + powered_up.add(str(pcpu)) + LOG.debug("Cores powered up : %s", powered_up) + + +def power_down(instance: objects.Instance) -> None: + if not CONF.libvirt.cpu_power_management: + return + if instance.numa_topology is None: + return + + cpu_dedicated_set = hardware.get_cpu_dedicated_set() or set() + pcpus = instance.numa_topology.cpu_pinning + powered_down = set() + for pcpu in pcpus: + if pcpu in cpu_dedicated_set: + pcpu = Core(pcpu) + if CONF.libvirt.cpu_power_management_strategy == 'cpu_state': + pcpu.online = False + else: + pcpu.set_low_governor() + powered_down.add(str(pcpu)) + LOG.debug("Cores powered down : %s", powered_down) + + +def power_down_all_dedicated_cpus() -> None: + if not CONF.libvirt.cpu_power_management: + return + if (CONF.libvirt.cpu_power_management and + not CONF.compute.cpu_dedicated_set + ): + msg = _("'[compute]/cpu_dedicated_set' is mandatory to be set if " + "'[libvirt]/cpu_power_management' is set." + "Please provide the CPUs that can be pinned or don't use the " + "power management if you only use shared CPUs.") + raise exception.InvalidConfiguration(msg) + + cpu_dedicated_set = hardware.get_cpu_dedicated_set() or set() + for pcpu in cpu_dedicated_set: + pcpu = Core(pcpu) + if CONF.libvirt.cpu_power_management_strategy == 'cpu_state': + pcpu.online = False + else: + pcpu.set_low_governor() + LOG.debug("Cores powered down : %s", cpu_dedicated_set) + + +def validate_all_dedicated_cpus() -> None: + if not CONF.libvirt.cpu_power_management: + return + cpu_dedicated_set = hardware.get_cpu_dedicated_set() or set() + governors = set() + cpu_states = set() + for pcpu in cpu_dedicated_set: + pcpu = Core(pcpu) + # we need to collect the governors strategy and the CPU states + governors.add(pcpu.governor) + cpu_states.add(pcpu.online) + if CONF.libvirt.cpu_power_management_strategy == 'cpu_state': + # all the cores need to have the same governor strategy + if len(governors) > 1: + msg = _("All the cores need to have the same governor strategy" + "before modifying the CPU states. You can reboot the " + "compute node if you prefer.") + raise exception.InvalidConfiguration(msg) + elif CONF.libvirt.cpu_power_management_strategy == 'governor': + # all the cores need to be online + if False in cpu_states: + msg = _("All the cores need to be online before modifying the " + "governor strategy.") + raise exception.InvalidConfiguration(msg) diff --git a/nova/virt/libvirt/cpu/core.py b/nova/virt/libvirt/cpu/core.py new file mode 100644 index 0000000000..782f028fee --- /dev/null +++ b/nova/virt/libvirt/cpu/core.py @@ -0,0 +1,78 @@ +# 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. + +import os +import typing as ty + +from oslo_log import log as logging + +from nova import exception +from nova import filesystem +import nova.privsep +from nova.virt import hardware + +LOG = logging.getLogger(__name__) + +AVAILABLE_PATH = '/sys/devices/system/cpu/present' + +CPU_PATH_TEMPLATE = '/sys/devices/system/cpu/cpu%(core)s' + + +def get_available_cores() -> ty.Set[int]: + cores = filesystem.read_sys(AVAILABLE_PATH) + return hardware.parse_cpu_spec(cores) if cores else set() + + +def exists(core: int) -> bool: + return core in get_available_cores() + + +def gen_cpu_path(core: int) -> str: + if not exists(core): + LOG.warning('Unable to access CPU: %s', core) + raise ValueError('CPU: %(core)s does not exist', core) + return CPU_PATH_TEMPLATE % {'core': core} + + +def get_online(core: int) -> bool: + try: + online = filesystem.read_sys( + os.path.join(gen_cpu_path(core), 'online')).strip() + except exception.FileNotFound: + # The online file may not exist if we haven't written it yet. + # By default, this means that the CPU is online. + online = '1' + return online == '1' + + +@nova.privsep.sys_admin_pctxt.entrypoint +def set_online(core: int) -> bool: + filesystem.write_sys(os.path.join(gen_cpu_path(core), 'online'), data='1') + return get_online(core) + + +def set_offline(core: int) -> bool: + filesystem.write_sys(os.path.join(gen_cpu_path(core), 'online'), data='0') + return not get_online(core) + + +def get_governor(core: int) -> str: + return filesystem.read_sys( + os.path.join(gen_cpu_path(core), 'cpufreq/scaling_governor')).strip() + + +@nova.privsep.sys_admin_pctxt.entrypoint +def set_governor(core: int, governor: str) -> str: + filesystem.write_sys( + os.path.join(gen_cpu_path(core), 'cpufreq/scaling_governor'), + data=governor) + return get_governor(core) diff --git a/nova/virt/libvirt/driver.py b/nova/virt/libvirt/driver.py index 4b34759bb7..fe48960296 100644 --- a/nova/virt/libvirt/driver.py +++ b/nova/virt/libvirt/driver.py @@ -114,6 +114,7 @@ from nova.virt.image import model as imgmodel from nova.virt import images from nova.virt.libvirt import blockinfo from nova.virt.libvirt import config as vconfig +from nova.virt.libvirt.cpu import api as libvirt_cpu from nova.virt.libvirt import designer from nova.virt.libvirt import event as libvirtevent from nova.virt.libvirt import guest as libvirt_guest @@ -817,6 +818,18 @@ class LibvirtDriver(driver.ComputeDriver): "force_raw_images to True.") raise exception.InvalidConfiguration(msg) + # NOTE(sbauza): We verify first if the dedicated CPU performances were + # modified by Nova before. Note that it can provide an exception if + # either the governor strategies are different between the cores or if + # the cores are offline. + libvirt_cpu.validate_all_dedicated_cpus() + # NOTE(sbauza): We powerdown all dedicated CPUs but if some instances + # exist that are pinned for some CPUs, then we'll later powerup those + # CPUs when rebooting the instance in _init_instance() + # Note that it can provide an exception if the config options are + # wrongly modified. + libvirt_cpu.power_down_all_dedicated_cpus() + # TODO(sbauza): Remove this code once mediated devices are persisted # across reboots. self._recreate_assigned_mediated_devices() @@ -1512,6 +1525,8 @@ class LibvirtDriver(driver.ComputeDriver): # NOTE(GuanQiang): teardown container to avoid resource leak if CONF.libvirt.virt_type == 'lxc': self._teardown_container(instance) + # We're sure the instance is gone, we can shutdown the core if so + libvirt_cpu.power_down(instance) def destroy(self, context, instance, network_info, block_device_info=None, destroy_disks=True, destroy_secrets=True): @@ -3164,6 +3179,7 @@ class LibvirtDriver(driver.ComputeDriver): current_power_state = guest.get_power_state(self._host) + libvirt_cpu.power_up(instance) # TODO(stephenfin): Any reason we couldn't use 'self.resume' here? guest.launch(pause=current_power_state == power_state.PAUSED) @@ -7300,6 +7316,11 @@ class LibvirtDriver(driver.ComputeDriver): graphics = vconfig.LibvirtConfigGuestGraphics() graphics.type = "spice" graphics.listen = CONF.spice.server_listen + graphics.image_compression = CONF.spice.image_compression + graphics.jpeg_compression = CONF.spice.jpeg_compression + graphics.zlib_compression = CONF.spice.zlib_compression + graphics.playback_compression = CONF.spice.playback_compression + graphics.streaming_mode = CONF.spice.streaming_mode guest.add_device(graphics) add_video_driver = True @@ -7615,7 +7636,7 @@ class LibvirtDriver(driver.ComputeDriver): instance: 'objects.Instance', power_on: bool = True, pause: bool = False, - post_xml_callback: ty.Callable = None, + post_xml_callback: ty.Optional[ty.Callable] = None, ) -> libvirt_guest.Guest: """Create a Guest from XML. @@ -7641,6 +7662,7 @@ class LibvirtDriver(driver.ComputeDriver): post_xml_callback() if power_on or pause: + libvirt_cpu.power_up(instance) guest.launch(pause=pause) return guest @@ -7675,7 +7697,7 @@ class LibvirtDriver(driver.ComputeDriver): block_device_info: ty.Optional[ty.Dict[str, ty.Any]], power_on: bool = True, vifs_already_plugged: bool = False, - post_xml_callback: ty.Callable = None, + post_xml_callback: ty.Optional[ty.Callable] = None, external_events: ty.Optional[ty.List[ty.Tuple[str, str]]] = None, cleanup_instance_dir: bool = False, cleanup_instance_disks: bool = False, @@ -7745,15 +7767,18 @@ class LibvirtDriver(driver.ComputeDriver): if not CONF.compute.cpu_dedicated_set: return set() - online_cpus = self._host.get_online_cpus() + if CONF.libvirt.cpu_power_management: + available_cpus = self._host.get_available_cpus() + else: + available_cpus = self._host.get_online_cpus() dedicated_cpus = hardware.get_cpu_dedicated_set() - if not dedicated_cpus.issubset(online_cpus): + if not dedicated_cpus.issubset(available_cpus): msg = _("Invalid '[compute] cpu_dedicated_set' config: one or " - "more of the configured CPUs is not online. Online " - "cpuset(s): %(online)s, configured cpuset(s): %(req)s") + "more of the configured CPUs is not available. Available " + "cpuset(s): %(available)s, configured cpuset(s): %(req)s") raise exception.Invalid(msg % { - 'online': sorted(online_cpus), + 'available': sorted(available_cpus), 'req': sorted(dedicated_cpus)}) return dedicated_cpus @@ -11332,8 +11357,8 @@ class LibvirtDriver(driver.ComputeDriver): def get_available_nodes(self, refresh=False): return [self._host.get_hostname()] - def get_available_node_uuids(self, refresh=False): - return [self._host.get_node_uuid()] + def get_nodenames_by_uuid(self, refresh=False): + return {self._host.get_node_uuid(): self._host.get_hostname()} def get_host_cpu_stats(self): """Return the current CPU state of the host.""" diff --git a/nova/virt/libvirt/event.py b/nova/virt/libvirt/event.py index a7d2a3624f..56951dc11c 100644 --- a/nova/virt/libvirt/event.py +++ b/nova/virt/libvirt/event.py @@ -9,6 +9,8 @@ # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the # License for the specific language governing permissions and limitations # under the License. +import typing as ty + from nova.virt import event @@ -22,7 +24,10 @@ class LibvirtEvent(event.InstanceEvent): class DeviceEvent(LibvirtEvent): """Base class for device related libvirt events""" - def __init__(self, uuid: str, dev: str, timestamp: float = None): + def __init__(self, + uuid: str, + dev: str, + timestamp: ty.Optional[float] = None): super().__init__(uuid, timestamp) self.dev = dev diff --git a/nova/virt/libvirt/host.py b/nova/virt/libvirt/host.py index b986702401..9658a5791d 100644 --- a/nova/virt/libvirt/host.py +++ b/nova/virt/libvirt/host.py @@ -740,6 +740,14 @@ class Host(object): return doms + def get_available_cpus(self): + """Get the set of CPUs that exist on the host. + + :returns: set of CPUs, raises libvirtError on error + """ + cpus, cpu_map, online = self.get_connection().getCPUMap() + return {cpu for cpu in range(cpus)} + def get_online_cpus(self): """Get the set of CPUs that are online on the host diff --git a/nova/virt/libvirt/utils.py b/nova/virt/libvirt/utils.py index adb2ec45a1..e1298ee5c8 100644 --- a/nova/virt/libvirt/utils.py +++ b/nova/virt/libvirt/utils.py @@ -261,8 +261,8 @@ def copy_image( dest: str, host: ty.Optional[str] = None, receive: bool = False, - on_execute: ty.Callable = None, - on_completion: ty.Callable = None, + on_execute: ty.Optional[ty.Callable] = None, + on_completion: ty.Optional[ty.Callable] = None, compression: bool = True, ) -> None: """Copy a disk image to an existing directory @@ -639,7 +639,7 @@ def mdev_name2uuid(mdev_name: str) -> str: return str(uuid.UUID(mdev_uuid)) -def mdev_uuid2name(mdev_uuid: str, parent: str = None) -> str: +def mdev_uuid2name(mdev_uuid: str, parent: ty.Optional[str] = None) -> str: """Convert an mdev uuid (of the form 8-4-4-4-12) and optionally its parent device to a name (of the form mdev_<uuid_with_underscores>[_<pciid>]). |