diff options
Diffstat (limited to 'nova/virt/libvirt/driver.py')
| -rw-r--r-- | nova/virt/libvirt/driver.py | 198 |
1 files changed, 151 insertions, 47 deletions
diff --git a/nova/virt/libvirt/driver.py b/nova/virt/libvirt/driver.py index ce884dfe30..73134d8391 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 @@ -412,6 +413,8 @@ class LibvirtDriver(driver.ComputeDriver): not CONF.force_raw_images) requires_ploop_image = CONF.libvirt.virt_type == 'parallels' + self.image_backend = imagebackend.Backend(CONF.use_cow_images) + self.capabilities = { "has_imagecache": True, "supports_evacuate": True, @@ -439,6 +442,10 @@ class LibvirtDriver(driver.ComputeDriver): "supports_bfv_rescue": True, "supports_vtpm": CONF.libvirt.swtpm_enabled, "supports_socket_pci_numa_affinity": True, + "supports_ephemeral_encryption": + self.image_backend.backend().SUPPORTS_LUKS, + "supports_ephemeral_encryption_luks": + self.image_backend.backend().SUPPORTS_LUKS, } super(LibvirtDriver, self).__init__(virtapi) @@ -463,7 +470,6 @@ class LibvirtDriver(driver.ComputeDriver): self._disk_cachemode = None self.image_cache_manager = imagecache.ImageCacheManager() - self.image_backend = imagebackend.Backend(CONF.use_cow_images) self.disk_cachemodes = {} @@ -812,6 +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() @@ -984,33 +1002,26 @@ class LibvirtDriver(driver.ComputeDriver): msg = _("The cpu_models option is required when cpu_mode=custom") raise exception.Invalid(msg) - cpu = vconfig.LibvirtConfigGuestCPU() - for model in models: - cpu.model = self._get_cpu_model_mapping(model) - try: - self._compare_cpu(cpu, self._get_cpu_info(), None) - except exception.InvalidCPUInfo as e: - msg = (_("Configured CPU model: %(model)s is not " - "compatible with host CPU. Please correct your " - "config and try again. %(e)s") % { - 'model': model, 'e': e}) - raise exception.InvalidCPUInfo(msg) - - # Use guest CPU model to check the compatibility between guest CPU and - # configured extra_flags - cpu = vconfig.LibvirtConfigGuestCPU() - cpu.model = self._host.get_capabilities().host.cpu.model - for flag in set(x.lower() for x in CONF.libvirt.cpu_model_extra_flags): - cpu_feature = self._prepare_cpu_flag(flag) - cpu.add_feature(cpu_feature) - try: - self._compare_cpu(cpu, self._get_cpu_info(), None) - except exception.InvalidCPUInfo as e: - msg = (_("Configured extra flag: %(flag)s it not correct, or " - "the host CPU does not support this flag. Please " - "correct the config and try again. %(e)s") % { - 'flag': flag, 'e': e}) - raise exception.InvalidCPUInfo(msg) + if not CONF.workarounds.skip_cpu_compare_at_startup: + # Use guest CPU model to check the compatibility between + # guest CPU and configured extra_flags + for model in models: + cpu = vconfig.LibvirtConfigGuestCPU() + cpu.model = self._get_cpu_model_mapping(model) + for flag in set(x.lower() for + x in CONF.libvirt.cpu_model_extra_flags): + cpu_feature = self._prepare_cpu_flag(flag) + cpu.add_feature(cpu_feature) + try: + self._compare_cpu(cpu, self._get_cpu_info(), None) + except exception.InvalidCPUInfo as e: + msg = (_("Configured CPU model: %(model)s " + "and CPU Flags %(flags)s ar not " + "compatible with host CPU. Please correct your " + "config and try again. %(e)s") % { + 'model': model, 'e': e, + 'flags': CONF.libvirt.cpu_model_extra_flags}) + raise exception.InvalidCPUInfo(msg) def _check_vtpm_support(self) -> None: # TODO(efried): A key manager must be configured to create/retrieve @@ -1514,6 +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): @@ -3166,6 +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) @@ -3250,7 +3264,13 @@ class LibvirtDriver(driver.ComputeDriver): '[Error Code %(error_code)s] %(ex)s') % {'instance_name': instance.name, 'error_code': error_code, 'ex': err_msg}) - raise exception.InternalError(msg) + + if error_code == libvirt.VIR_ERR_AGENT_UNRESPONSIVE: + msg += (", libvirt cannot connect to the qemu-guest-agent" + " inside the instance.") + raise exception.InstanceQuiesceFailed(reason=msg) + else: + raise exception.InternalError(msg) def quiesce(self, context, instance, image_meta): """Freeze the guest filesystems to prepare for snapshot. @@ -7296,6 +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 @@ -7637,6 +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 @@ -7741,15 +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 @@ -8227,15 +8256,52 @@ class LibvirtDriver(driver.ComputeDriver): def _get_mediated_device_information(self, devname): """Returns a dict of a mediated device.""" - virtdev = self._host.device_lookup_by_name(devname) + # LP #1951656 - In Libvirt 7.7, the mdev name now includes the PCI + # address of the parent device (e.g. mdev_<uuid>_<pci_address>) due to + # the mdevctl allowing for multiple mediated devs having the same UUID + # defined (only one can be active at a time). Since the guest + # information doesn't have the parent ID, try to lookup which + # mediated device is available that matches the UUID. If multiple + # devices are found that match the UUID, then this is an error + # condition. + try: + virtdev = self._host.device_lookup_by_name(devname) + except libvirt.libvirtError as ex: + if ex.get_error_code() != libvirt.VIR_ERR_NO_NODE_DEVICE: + raise + mdevs = [dev for dev in self._host.list_mediated_devices() + if dev.startswith(devname)] + # If no matching devices are found, simply raise the original + # exception indicating that no devices are found. + if not mdevs: + raise + elif len(mdevs) > 1: + msg = ("The mediated device name %(devname)s refers to a UUID " + "that is present in multiple libvirt mediated devices. " + "Matching libvirt mediated devices are %(devices)s. " + "Mediated device UUIDs must be unique for Nova." % + {'devname': devname, + 'devices': ', '.join(mdevs)}) + raise exception.InvalidLibvirtMdevConfig(reason=msg) + + LOG.debug('Found requested device %s as %s. Using that.', + devname, mdevs[0]) + virtdev = self._host.device_lookup_by_name(mdevs[0]) xmlstr = virtdev.XMLDesc(0) cfgdev = vconfig.LibvirtConfigNodeDevice() cfgdev.parse_str(xmlstr) + # Starting with Libvirt 7.3, the uuid information is available in the + # node device information. If its there, use that. Otherwise, + # fall back to the previous behavior of parsing the uuid from the + # devname. + if cfgdev.mdev_information.uuid: + mdev_uuid = cfgdev.mdev_information.uuid + else: + mdev_uuid = libvirt_utils.mdev_name2uuid(cfgdev.name) device = { "dev_id": cfgdev.name, - # name is like mdev_00ead764_fdc0_46b6_8db9_2963f5c815b4 - "uuid": libvirt_utils.mdev_name2uuid(cfgdev.name), + "uuid": mdev_uuid, # the physical GPU PCI device "parent": cfgdev.parent, "type": cfgdev.mdev_information.type, @@ -8323,6 +8389,7 @@ class LibvirtDriver(driver.ComputeDriver): :param requested_types: Filter out the result for only mediated devices having those types. """ + LOG.debug('Searching for available mdevs...') allocated_mdevs = self._get_all_assigned_mediated_devices() mdevs = self._get_mediated_devices(requested_types) available_mdevs = set() @@ -8338,6 +8405,7 @@ class LibvirtDriver(driver.ComputeDriver): available_mdevs.add(mdev["uuid"]) available_mdevs -= set(allocated_mdevs) + LOG.info('Available mdevs at: %s.', available_mdevs) return available_mdevs def _create_new_mediated_device(self, parent, uuid=None): @@ -8349,6 +8417,7 @@ class LibvirtDriver(driver.ComputeDriver): :returns: the newly created mdev UUID or None if not possible """ + LOG.debug('Attempting to create new mdev...') supported_types = self.supported_vgpu_types # Try to see if we can still create a new mediated device devices = self._get_mdev_capable_devices(supported_types) @@ -8360,6 +8429,7 @@ class LibvirtDriver(driver.ComputeDriver): # The device is not the one that was called, not creating # the mdev continue + LOG.debug('Trying on: %s.', dev_name) dev_supported_type = self._get_vgpu_type_per_pgpu(dev_name) if dev_supported_type and device['types'][ dev_supported_type]['availableInstances'] > 0: @@ -8369,7 +8439,13 @@ class LibvirtDriver(driver.ComputeDriver): pci_addr = "{}:{}:{}.{}".format(*dev_name[4:].split('_')) chosen_mdev = nova.privsep.libvirt.create_mdev( pci_addr, dev_supported_type, uuid=uuid) + LOG.info('Created mdev: %s on pGPU: %s.', + chosen_mdev, pci_addr) return chosen_mdev + LOG.debug('Failed: No available instances on device.') + LOG.info('Failed to create mdev. ' + 'No free space found among the following devices: %s.', + [dev['dev_id'] for dev in devices]) @utils.synchronized(VGPU_RESOURCE_SEMAPHORE) def _allocate_mdevs(self, allocations): @@ -8452,6 +8528,8 @@ class LibvirtDriver(driver.ComputeDriver): # Take the first available mdev chosen_mdev = mdevs_available.pop() else: + LOG.debug('No available mdevs where found. ' + 'Creating an new one...') chosen_mdev = self._create_new_mediated_device(parent_device) if not chosen_mdev: # If we can't find devices having available VGPUs, just raise @@ -8459,6 +8537,7 @@ class LibvirtDriver(driver.ComputeDriver): reason='mdev-capable resource is not available') else: chosen_mdevs.append(chosen_mdev) + LOG.info('Allocated mdev: %s.', chosen_mdev) return chosen_mdevs def _detach_mediated_devices(self, guest): @@ -9461,6 +9540,7 @@ class LibvirtDriver(driver.ComputeDriver): data["hypervisor_type"] = self._host.get_driver_type() data["hypervisor_version"] = self._host.get_version() data["hypervisor_hostname"] = self._host.get_hostname() + data["uuid"] = self._host.get_node_uuid() # TODO(berrange): why do we bother converting the # libvirt capabilities XML into a special JSON format ? # The data format is different across all the drivers @@ -9914,7 +9994,7 @@ class LibvirtDriver(driver.ComputeDriver): try: cpu_xml = cpu.to_xml() LOG.debug("cpu compare xml: %s", cpu_xml, instance=instance) - ret = self._host.compare_cpu(cpu_xml) + ret = self._host.compare_hypervisor_cpu(cpu_xml) except libvirt.libvirtError as e: error_code = e.get_error_code() if error_code == libvirt.VIR_ERR_NO_SUPPORT: @@ -10997,16 +11077,37 @@ class LibvirtDriver(driver.ComputeDriver): if not CONF.workarounds.enable_qemu_monitor_announce_self: return - LOG.info('Sending announce-self command to QEMU monitor', - instance=instance) + current_attempt = 0 - try: - guest = self._host.get_guest(instance) - guest.announce_self() - except Exception: - LOG.warning('Failed to send announce-self command to QEMU monitor', - instance=instance) - LOG.exception() + max_attempts = ( + CONF.workarounds.qemu_monitor_announce_self_count) + # qemu_monitor_announce_retry_interval specified in seconds + announce_pause = ( + CONF.workarounds.qemu_monitor_announce_self_interval) + + while(current_attempt < max_attempts): + # Increment attempt + current_attempt += 1 + + # Only use announce_pause after the first attempt to avoid + # pausing before calling announce_self for the first attempt + if current_attempt != 1: + greenthread.sleep(announce_pause) + + LOG.info('Sending announce-self command to QEMU monitor. ' + 'Attempt %(current_attempt)s of %(max_attempts)s', + {'current_attempt': current_attempt, + 'max_attempts': max_attempts}, instance=instance) + try: + guest = self._host.get_guest(instance) + guest.announce_self() + except Exception: + LOG.warning('Failed to send announce-self command to ' + 'QEMU monitor. Attempt %(current_attempt)s of ' + '%(max_attempts)s', + {'current_attempt': current_attempt, + 'max_attempts': max_attempts}, instance=instance) + LOG.exception() def post_live_migration_at_destination(self, context, instance, @@ -11256,6 +11357,9 @@ class LibvirtDriver(driver.ComputeDriver): def get_available_nodes(self, refresh=False): return [self._host.get_hostname()] + 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.""" return self._host.get_cpu_stats() |