path: root/nova/compute/api.py

# Copyright 2010 United States Government as represented by the
# Administrator of the National Aeronautics and Space Administration.
# Copyright 2011 Piston Cloud Computing, Inc.
# Copyright 2012-2013 Red Hat, Inc.
# All Rights Reserved.
#
#    Licensed under the Apache License, Version 2.0 (the "License"); you may
#    not use this file except in compliance with the License. You may obtain
#    a copy of the License at
#
#         http://www.apache.org/licenses/LICENSE-2.0
#
#    Unless required by applicable law or agreed to in writing, software
#    distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
#    WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
#    License for the specific language governing permissions and limitations
#    under the License.

"""Handles all requests relating to compute resources (e.g. guest VMs,
networking and storage of VMs, and compute hosts on which they run)."""

import collections
import functools
import re
import typing as ty

from castellan import key_manager
import os_traits
from oslo_log import log as logging
from oslo_messaging import exceptions as oslo_exceptions
from oslo_serialization import base64 as base64utils
from oslo_utils import excutils
from oslo_utils import strutils
from oslo_utils import timeutils
from oslo_utils import units
from oslo_utils import uuidutils

from nova.accelerator import cyborg
from nova import availability_zones
from nova import block_device
from nova.compute import flavors
from nova.compute import instance_actions
from nova.compute import instance_list
from nova.compute import migration_list
from nova.compute import power_state
from nova.compute import rpcapi as compute_rpcapi
from nova.compute import task_states
from nova.compute import utils as compute_utils
from nova.compute.utils import wrap_instance_event
from nova.compute import vm_states
from nova import conductor
import nova.conf
from nova import context as nova_context
from nova import crypto
from nova.db.api import api as api_db_api
from nova.db.main import api as main_db_api
from nova import exception
from nova import exception_wrapper
from nova.i18n import _
from nova.image import glance
from nova.limit import local as local_limit
from nova.limit import placement as placement_limits
from nova.network import constants
from nova.network import model as network_model
from nova.network import neutron
from nova.network import security_group_api
from nova import objects
from nova.objects import block_device as block_device_obj
from nova.objects import external_event as external_event_obj
from nova.objects import fields as fields_obj
from nova.objects import image_meta as image_meta_obj
from nova.objects import keypair as keypair_obj
from nova.objects import quotas as quotas_obj
from nova.objects import service as service_obj
from nova.pci import request as pci_request
from nova.policies import servers as servers_policies
from nova.policies import shelve as shelve_policies
import nova.policy
from nova import profiler
from nova import rpc
from nova.scheduler.client import query
from nova.scheduler.client import report
from nova.scheduler import utils as scheduler_utils
from nova import servicegroup
from nova import utils
from nova.virt import hardware
from nova.volume import cinder

LOG = logging.getLogger(__name__)

# NOTE(gibi): legacy notification used compute as a service but these
# calls still run on the client side of the compute service which is
# nova-api. By setting the binary to nova-api below, we can make sure
# that the new versioned notifications has the right publisher_id but the
# legacy notifications does not change.
wrap_exception = functools.partial(
    exception_wrapper.wrap_exception, service='compute', binary='nova-api')
CONF = nova.conf.CONF

AGGREGATE_ACTION_UPDATE = 'Update'
AGGREGATE_ACTION_UPDATE_META = 'UpdateMeta'
AGGREGATE_ACTION_DELETE = 'Delete'
AGGREGATE_ACTION_ADD = 'Add'

MIN_COMPUTE_SYNC_COMPUTE_STATUS_DISABLED = 38
MIN_COMPUTE_CROSS_CELL_RESIZE = 47
MIN_COMPUTE_SAME_HOST_COLD_MIGRATE = 48

# TODO(huaqiang): Remove in Wallaby
MIN_VER_NOVA_COMPUTE_MIXED_POLICY = 52

SUPPORT_ACCELERATOR_SERVICE_FOR_REBUILD = 53

SUPPORT_VNIC_TYPE_ACCELERATOR = 57

MIN_COMPUTE_BOOT_WITH_EXTENDED_RESOURCE_REQUEST = 58
MIN_COMPUTE_MOVE_WITH_EXTENDED_RESOURCE_REQUEST = 59
MIN_COMPUTE_INT_ATTACH_WITH_EXTENDED_RES_REQ = 60

SUPPORT_VNIC_TYPE_REMOTE_MANAGED = 61
MIN_COMPUTE_VDPA_ATTACH_DETACH = 62
MIN_COMPUTE_VDPA_HOTPLUG_LIVE_MIGRATION = 63


# FIXME(danms): Keep a global cache of the cells we find the
# first time we look. This needs to be refreshed on a timer or
# trigger.
CELLS = []


def check_instance_state(vm_state=None, task_state=(None,),
                         must_have_launched=True):
    """Decorator to check VM and/or task state before entry to API functions.

    If the instance is in the wrong state, or has not been successfully
    started at least once the wrapper will raise an exception.
    """

    if vm_state is not None and not isinstance(vm_state, set):
        vm_state = set(vm_state)
    if task_state is not None and not isinstance(task_state, set):
        task_state = set(task_state)

    def outer(f):
        @functools.wraps(f)
        def inner(self, context, instance, *args, **kw):
            if vm_state is not None and instance.vm_state not in vm_state:
                raise exception.InstanceInvalidState(
                    attr='vm_state',
                    instance_uuid=instance.uuid,
                    state=instance.vm_state,
                    method=f.__name__)
            if (task_state is not None and
                    instance.task_state not in task_state):
                raise exception.InstanceInvalidState(
                    attr='task_state',
                    instance_uuid=instance.uuid,
                    state=instance.task_state,
                    method=f.__name__)
            if must_have_launched and not instance.launched_at:
                raise exception.InstanceInvalidState(
                    attr='launched_at',
                    instance_uuid=instance.uuid,
                    state=instance.launched_at,
                    method=f.__name__)

            return f(self, context, instance, *args, **kw)
        return inner
    return outer


def _set_or_none(q):
    return q if q is None or isinstance(q, set) else set(q)


def reject_instance_state(vm_state=None, task_state=None):
    """Decorator.  Raise InstanceInvalidState if instance is in any of the
    given states.
    """

    vm_state = _set_or_none(vm_state)
    task_state = _set_or_none(task_state)

    def outer(f):
        @functools.wraps(f)
        def inner(self, context, instance, *args, **kw):
            _InstanceInvalidState = functools.partial(
                exception.InstanceInvalidState,
                instance_uuid=instance.uuid,
                method=f.__name__)

            if vm_state is not None and instance.vm_state in vm_state:
                raise _InstanceInvalidState(
                    attr='vm_state', state=instance.vm_state)

            if task_state is not None and instance.task_state in task_state:
                raise _InstanceInvalidState(
                    attr='task_state', state=instance.task_state)

            return f(self, context, instance, *args, **kw)
        return inner
    return outer


def check_instance_host(check_is_up=False):
    """Validate the instance.host before performing the operation.

    At a minimum this method will check that the instance.host is set.

    :param check_is_up: If True, check that the instance.host status is UP
        or MAINTENANCE (disabled but not down).
    :raises: InstanceNotReady if the instance.host is not set
    :raises: ServiceUnavailable if check_is_up=True and the instance.host
        compute service status is not UP or MAINTENANCE
    """
    def outer(function):
        @functools.wraps(function)
        def wrapped(self, context, instance, *args, **kwargs):
            if not instance.host:
                raise exception.InstanceNotReady(instance_id=instance.uuid)
            if check_is_up:
                # Make sure the source compute service is not down otherwise we
                # cannot proceed.
                service = [
                    service for service in instance.services
                        if service.binary == 'nova-compute'][0]
                if not self.servicegroup_api.service_is_up(service):
                    # ComputeServiceUnavailable would make more sense here but
                    # we do not want to leak hostnames to end users.
                    raise exception.ServiceUnavailable()
            return function(self, context, instance, *args, **kwargs)
        return wrapped
    return outer


def check_instance_lock(function):
    @functools.wraps(function)
    def inner(self, context, instance, *args, **kwargs):
        if instance.locked and not context.is_admin:
            raise exception.InstanceIsLocked(instance_uuid=instance.uuid)
        return function(self, context, instance, *args, **kwargs)
    return inner


def reject_sev_instances(operation):
    """Reject requests to decorated function if instance has SEV enabled.

    Raise OperationNotSupportedForSEV if instance has SEV enabled.
    """

    def outer(f):
        @functools.wraps(f)
        def inner(self, context, instance, *args, **kw):
            if hardware.get_mem_encryption_constraint(instance.flavor,
                                                      instance.image_meta):
                raise exception.OperationNotSupportedForSEV(
                    instance_uuid=instance.uuid,
                    operation=operation)
            return f(self, context, instance, *args, **kw)
        return inner
    return outer


def reject_vtpm_instances(operation):
    """Reject requests to decorated function if instance has vTPM enabled.

    Raise OperationNotSupportedForVTPM if instance has vTPM enabled.
    """

    def outer(f):
        @functools.wraps(f)
        def inner(self, context, instance, *args, **kw):
            if hardware.get_vtpm_constraint(
                instance.flavor, instance.image_meta,
            ):
                raise exception.OperationNotSupportedForVTPM(
                    instance_uuid=instance.uuid, operation=operation)
            return f(self, context, instance, *args, **kw)
        return inner
    return outer


def reject_vdpa_instances(operation, until=None):
    """Reject requests to decorated function if instance has vDPA interfaces.

    Raise OperationNotSupportedForVDPAInterfaces if operations involves one or
        more vDPA interfaces.
    """

    def outer(f):
        @functools.wraps(f)
        def inner(self, context, instance, *args, **kw):
            if any(
                vif['vnic_type'] == network_model.VNIC_TYPE_VDPA
                for vif in instance.get_network_info()
            ):
                reject = True
                if until is not None:
                    min_ver = objects.service.get_minimum_version_all_cells(
                        nova_context.get_admin_context(), ['nova-compute']
                    )
                    if min_ver >= until:
                        reject = False

                if reject:
                    raise exception.OperationNotSupportedForVDPAInterface(
                        instance_uuid=instance.uuid, operation=operation
                    )
            return f(self, context, instance, *args, **kw)
        return inner
    return outer


def load_cells():
    global CELLS
    if not CELLS:
        CELLS = objects.CellMappingList.get_all(
            nova_context.get_admin_context())
        LOG.debug('Found %(count)i cells: %(cells)s',
                  dict(count=len(CELLS),
                       cells=','.join([c.identity for c in CELLS])))

    if not CELLS:
        LOG.error('No cells are configured, unable to continue')


def _get_image_meta_obj(image_meta_dict):
    try:
        image_meta = objects.ImageMeta.from_dict(image_meta_dict)
    except ValueError as e:
        # there must be invalid values in the image meta properties so
        # consider this an invalid request
        msg = _('Invalid image metadata. Error: %s') % str(e)
        raise exception.InvalidRequest(msg)
    return image_meta


def block_accelerators(until_service=None):
    def inner(func):
        @functools.wraps(func)
        def wrapper(self, context, instance, *args, **kwargs):
            # NOTE(brinzhang): Catch a request operating a mixed instance,
            # make sure all nova-compute services have been upgraded and
            # support the accelerators.
            dp_name = instance.flavor.extra_specs.get('accel:device_profile')
            service_support = False
            if not dp_name:
                service_support = True
            elif until_service:
                min_version = objects.service.get_minimum_version_all_cells(
                    nova_context.get_admin_context(), ['nova-compute'])
                if min_version >= until_service:
                    service_support = True
            if not service_support:
                raise exception.ForbiddenWithAccelerators()
            return func(self, context, instance, *args, **kwargs)
        return wrapper
    return inner


def block_port_accelerators():
    def inner(func):
        @functools.wraps(func)
        def wrapper(self, context, instance, *args, **kwargs):
            # Catch a request operating a instance with accelerators
            # attach to ports.
            nw_info = instance.get_network_info()
            for vif in nw_info:
                vnic_type = vif['vnic_type']
                if vnic_type in (network_model.VNIC_TYPE_ACCELERATOR_DIRECT,
                    network_model.VNIC_TYPE_ACCELERATOR_DIRECT_PHYSICAL):
                    raise exception.ForbiddenPortsWithAccelerator()
            return func(self, context, instance, *args, **kwargs)
        return wrapper
    return inner


def block_extended_resource_request(function):
    @functools.wraps(function)
    def inner(self, context, instance, *args, **kwargs):
        if self.network_api.instance_has_extended_resource_request(
                instance.uuid
        ):
            version = service_obj.get_minimum_version_all_cells(
                context, ["nova-compute"])
            if version < MIN_COMPUTE_MOVE_WITH_EXTENDED_RESOURCE_REQUEST:
                raise exception.ExtendedResourceRequestOldCompute()
        return function(self, context, instance, *args, **kwargs)
    return inner


@profiler.trace_cls("compute_api")
class API:
    """API for interacting with the compute manager."""

    _sentinel = object()

    def __init__(self, image_api=None, network_api=None, volume_api=None):
        self.image_api = image_api or glance.API()
        self.network_api = network_api or neutron.API()
        self.volume_api = volume_api or cinder.API()
        self.compute_rpcapi = compute_rpcapi.ComputeAPI()
        self.compute_task_api = conductor.ComputeTaskAPI()
        self.servicegroup_api = servicegroup.API()
        self.host_api = HostAPI(self.compute_rpcapi, self.servicegroup_api)
        self.notifier = rpc.get_notifier('compute')
        if CONF.ephemeral_storage_encryption.enabled:
            self.key_manager = key_manager.API()
        # Help us to record host in EventReporter
        self.host = CONF.host

    def _record_action_start(self, context, instance, action):
        objects.InstanceAction.action_start(context, instance.uuid,
                                            action, want_result=False)

    def _check_injected_file_quota(self, context, injected_files):
        """Enforce quota limits on injected files.

        Raises a OverQuota if any limit is exceeded.
        """
        if not injected_files:
            return

        # Check number of files first
        try:
            objects.Quotas.limit_check(context,
                                       injected_files=len(injected_files))
            local_limit.enforce_api_limit(local_limit.INJECTED_FILES,
                                          len(injected_files))
        except exception.OnsetFileLimitExceeded:
            raise
        except exception.OverQuota:
            raise exception.OnsetFileLimitExceeded()

        # OK, now count path and content lengths; we're looking for
        # the max...
        max_path = 0
        max_content = 0
        for path, content in injected_files:
            max_path = max(max_path, len(path))
            max_content = max(max_content, len(content))

        try:
            objects.Quotas.limit_check(context,
                                       injected_file_path_bytes=max_path,
                                       injected_file_content_bytes=max_content)
            # TODO(johngarbutt) we can simplify the except clause when
            # the above legacy quota check is removed.
            local_limit.enforce_api_limit(
                local_limit.INJECTED_FILES_PATH, max_path)
            local_limit.enforce_api_limit(
                local_limit.INJECTED_FILES_CONTENT, max_content)
        except exception.OnsetFilePathLimitExceeded:
            raise
        except exception.OnsetFileContentLimitExceeded:
            raise
        except exception.OverQuota as exc:
            # Favor path limit over content limit for reporting
            # purposes
            if 'injected_file_path_bytes' in exc.kwargs['overs']:
                raise exception.OnsetFilePathLimitExceeded(
                      allowed=exc.kwargs['quotas']['injected_file_path_bytes'])
            else:
                raise exception.OnsetFileContentLimitExceeded(
                   allowed=exc.kwargs['quotas']['injected_file_content_bytes'])

    def _check_metadata_properties_quota(self, context, metadata=None):
        """Enforce quota limits on metadata properties."""
        if not metadata:
            return
        if not isinstance(metadata, dict):
            msg = (_("Metadata type should be dict."))
            raise exception.InvalidMetadata(reason=msg)
        num_metadata = len(metadata)
        try:
            objects.Quotas.limit_check(context, metadata_items=num_metadata)
            local_limit.enforce_api_limit(
                local_limit.SERVER_METADATA_ITEMS, num_metadata)
        except exception.MetadataLimitExceeded:
            raise
        except exception.OverQuota as exc:
            quota_metadata = exc.kwargs['quotas']['metadata_items']
            raise exception.MetadataLimitExceeded(allowed=quota_metadata)

        # Because metadata is stored in the DB, we hard-code the size limits
        # In future, we may support more variable length strings, so we act
        #  as if this is quota-controlled for forwards compatibility.
        # Those are only used in V2 API, from V2.1 API, those checks are
        # validated at API layer schema validation.
        for k, v in metadata.items():
            try:
                utils.check_string_length(v)
                utils.check_string_length(k, min_length=1)
            except exception.InvalidInput as e:
                raise exception.InvalidMetadata(reason=e.format_message())

            if len(k) > 255:
                msg = _("Metadata property key greater than 255 characters")
                raise exception.InvalidMetadataSize(reason=msg)
            if len(v) > 255:
                msg = _("Metadata property value greater than 255 characters")
                raise exception.InvalidMetadataSize(reason=msg)

    def _check_requested_secgroups(self, context, secgroups):
        """Check if the security group requested exists and belongs to
        the project.

        :param context: The nova request context.
        :type context: nova.context.RequestContext
        :param secgroups: list of requested security group names
        :type secgroups: list
        :returns: list of requested security group UUIDs; note that 'default'
            is a special case and will be unmodified if it's requested.
        """
        security_groups = []
        for secgroup in secgroups:
            # NOTE(sdague): default is handled special
            if secgroup == "default":
                security_groups.append(secgroup)
                continue
            secgroup_uuid = security_group_api.validate_name(context, secgroup)
            security_groups.append(secgroup_uuid)

        return security_groups

    def _check_requested_networks(self, context, requested_networks,
                                  max_count):
        """Check if the networks requested belongs to the project
        and the fixed IP address for each network provided is within
        same the network block
        """
        if requested_networks is not None:
            if requested_networks.no_allocate:
                # If the network request was specifically 'none' meaning don't
                # allocate any networks, we just return the number of requested
                # instances since quotas don't change at all.
                return max_count

            # NOTE(danms): Temporary transition
            requested_networks = requested_networks.as_tuples()

        return self.network_api.validate_networks(context, requested_networks,
                                                  max_count)

    def _handle_kernel_and_ramdisk(self, context, kernel_id, ramdisk_id,
                                   image):
        """Choose kernel and ramdisk appropriate for the instance.

        The kernel and ramdisk can be chosen in one of two ways:

            1. Passed in with create-instance request.

            2. Inherited from image metadata.

        If inherited from image metadata, and if that image metadata value is
        set to 'nokernel', both kernel and ramdisk will default to None.
        """
        # Inherit from image if not specified
        image_properties = image.get('properties', {})

        if kernel_id is None:
            kernel_id = image_properties.get('kernel_id')

        if ramdisk_id is None:
            ramdisk_id = image_properties.get('ramdisk_id')

        # Force to None if kernel_id indicates that a kernel is not to be used
        if kernel_id == 'nokernel':
            kernel_id = None
            ramdisk_id = None

        # Verify kernel and ramdisk exist (fail-fast)
        if kernel_id is not None:
            kernel_image = self.image_api.get(context, kernel_id)
            # kernel_id could have been a URI, not a UUID, so to keep behaviour
            # from before, which leaked that implementation detail out to the
            # caller, we return the image UUID of the kernel image and ramdisk
            # image (below) and not any image URIs that might have been
            # supplied.
            # TODO(jaypipes): Get rid of this silliness once we move to a real
            # Image object and hide all of that stuff within nova.image.glance
            kernel_id = kernel_image['id']

        if ramdisk_id is not None:
            ramdisk_image = self.image_api.get(context, ramdisk_id)
            ramdisk_id = ramdisk_image['id']

        return kernel_id, ramdisk_id

    @staticmethod
    def parse_availability_zone(context, availability_zone):
        # NOTE(vish): We have a legacy hack to allow admins to specify hosts
        #             via az using az:host:node. It might be nice to expose an
        #             api to specify specific hosts to force onto, but for
        #             now it just supports this legacy hack.
        # NOTE(deva): It is also possible to specify az::node, in which case
        #             the host manager will determine the correct host.
        forced_host = None
        forced_node = None
        if availability_zone and ':' in availability_zone:
            c = availability_zone.count(':')
            if c == 1:
                availability_zone, forced_host = availability_zone.split(':')
            elif c == 2:
                if '::' in availability_zone:
                    availability_zone, forced_node = \
                            availability_zone.split('::')
                else:
                    availability_zone, forced_host, forced_node = \
                            availability_zone.split(':')
            else:
                raise exception.InvalidInput(
                        reason="Unable to parse availability_zone")

        if not availability_zone:
            availability_zone = CONF.default_schedule_zone

        return availability_zone, forced_host, forced_node

    def _ensure_auto_disk_config_is_valid(self, auto_disk_config_img,
                                          auto_disk_config, image):
        auto_disk_config_disabled = \
                utils.is_auto_disk_config_disabled(auto_disk_config_img)
        if auto_disk_config_disabled and auto_disk_config:
            raise exception.AutoDiskConfigDisabledByImage(image=image)

    def _inherit_properties_from_image(self, image, auto_disk_config):
        image_properties = image.get('properties', {})
        auto_disk_config_img = \
                utils.get_auto_disk_config_from_image_props(image_properties)
        self._ensure_auto_disk_config_is_valid(auto_disk_config_img,
                                               auto_disk_config,
                                               image.get("id"))
        if auto_disk_config is None:
            auto_disk_config = strutils.bool_from_string(auto_disk_config_img)

        return {
            'os_type': image_properties.get('os_type'),
            'architecture': image_properties.get('architecture'),
            'vm_mode': image_properties.get('vm_mode'),
            'auto_disk_config': auto_disk_config
        }

    def _check_config_drive(self, config_drive):
        if config_drive:
            try:
                bool_val = strutils.bool_from_string(config_drive,
                                                     strict=True)
            except ValueError:
                raise exception.ConfigDriveInvalidValue(option=config_drive)
        else:
            bool_val = False
        # FIXME(comstud):  Bug ID 1193438 filed for this. This looks silly,
        # but this is because the config drive column is a String.  False
        # is represented by using an empty string.  And for whatever
        # reason, we rely on the DB to cast True to a String.
        return True if bool_val else ''

    def _validate_flavor_image(
        self, context, image_id, image, flavor, root_bdm, validate_numa=True,
    ):
        """Validate the flavor and image.

        This is called from the API service to ensure that the flavor
        extra-specs and image properties are self-consistent and compatible
        with each other.

        :param context: A context.RequestContext
        :param image_id: UUID of the image
        :param image: a dict representation of the image including properties,
            enforces the image status is active.
        :param flavor: Flavor object
        :param root_bdm: BlockDeviceMapping for root disk.  Will be None for
            the resize case.
        :param validate_numa: Flag to indicate whether or not to validate
            the NUMA-related metadata.
        :raises: Many different possible exceptions. See
            api.openstack.compute.servers.INVALID_FLAVOR_IMAGE_EXCEPTIONS
            for the full list.
        """
        if image and image['status'] != 'active':
            raise exception.ImageNotActive(image_id=image_id)
        self._validate_flavor_image_nostatus(
            context, image, flavor, root_bdm, validate_numa)

    @staticmethod
    def _detect_nonbootable_image_from_properties(image_id, image):
        """Check image for a property indicating it's nonbootable.

        This is called from the API service to ensure that there are
        no known image properties indicating that this image is of a
        type that we do not support booting from.

        Currently the only such property is 'cinder_encryption_key_id'.

        :param image_id: UUID of the image
        :param image: a dict representation of the image including properties
        :raises: ImageUnacceptable if the image properties indicate
                 that booting this image is not supported
        """
        if not image:
            return

        image_properties = image.get('properties', {})
        # NOTE(lyarwood) Skip this check when image_id is None indicating that
        # the instance is booting from a volume that was itself initially
        # created from an image. As such we don't care if
        # cinder_encryption_key_id was against the original image as we are now
        # booting from an encrypted volume.
        if image_properties.get('cinder_encryption_key_id') and image_id:
            reason = _('Direct booting of an image uploaded from an '
                       'encrypted volume is unsupported.')
            raise exception.ImageUnacceptable(image_id=image_id,
                                              reason=reason)

    @staticmethod
    def _validate_flavor_image_nostatus(
        context, image, flavor, root_bdm, validate_numa=True,
        validate_pci=False,
    ):
        """Validate the flavor and image.

        This is called from the API service to ensure that the flavor
        extra-specs and image properties are self-consistent and compatible
        with each other.

        :param context: A context.RequestContext
        :param image: a dict representation of the image including properties
        :param flavor: Flavor object
        :param root_bdm: BlockDeviceMapping for root disk.  Will be None for
               the resize case.
        :param validate_numa: Flag to indicate whether or not to validate
               the NUMA-related metadata.
        :param validate_pci: Flag to indicate whether or not to validate
               the PCI-related metadata.
        :raises: Many different possible exceptions.  See
                 api.openstack.compute.servers.INVALID_FLAVOR_IMAGE_EXCEPTIONS
                 for the full list.
        """
        if not image:
            return

        image_properties = image.get('properties', {})
        config_drive_option = image_properties.get(
            'img_config_drive', 'optional')
        if config_drive_option not in ['optional', 'mandatory']:
            raise exception.InvalidImageConfigDrive(
                config_drive=config_drive_option)

        if flavor['memory_mb'] < int(image.get('min_ram') or 0):
            raise exception.FlavorMemoryTooSmall()

        # Image min_disk is in gb, size is in bytes. For sanity, have them both
        # in bytes.
        image_min_disk = int(image.get('min_disk') or 0) * units.Gi
        image_size = int(image.get('size') or 0)

        # Target disk is a volume. Don't check flavor disk size because it
        # doesn't make sense, and check min_disk against the volume size.
        if root_bdm is not None and root_bdm.is_volume:
            # There are 2 possibilities here:
            #
            # 1. The target volume already exists but bdm.volume_size is not
            #    yet set because this method is called before
            #    _bdm_validate_set_size_and_instance during server create.
            # 2. The target volume doesn't exist, in which case the bdm will
            #    contain the intended volume size
            #
            # Note that rebuild also calls this method with potentially a new
            # image but you can't rebuild a volume-backed server with a new
            # image (yet).
            #
            # Cinder does its own check against min_disk, so if the target
            # volume already exists this has already been done and we don't
            # need to check it again here. In this case, volume_size may not be
            # set on the bdm.
            #
            # If we're going to create the volume, the bdm will contain
            # volume_size. Therefore we should check it if it exists. This will
            # still be checked again by cinder when the volume is created, but
            # that will not happen until the request reaches a host. By
            # checking it here, the user gets an immediate and useful failure
            # indication.
            #
            # The third possibility is that we have failed to consider
            # something, and there are actually more than 2 possibilities. In
            # this case cinder will still do the check at volume creation time.
            # The behaviour will still be correct, but the user will not get an
            # immediate failure from the api, and will instead have to
            # determine why the instance is in an error state with a task of
            # block_device_mapping.
            #
            # We could reasonably refactor this check into _validate_bdm at
            # some future date, as the various size logic is already split out
            # in there.
            dest_size = root_bdm.volume_size
            if dest_size is not None:
                dest_size *= units.Gi

                if image_min_disk > dest_size:
                    raise exception.VolumeSmallerThanMinDisk(
                        volume_size=dest_size, image_min_disk=image_min_disk)

        # Target disk is a local disk whose size is taken from the flavor
        else:
            dest_size = flavor['root_gb'] * units.Gi

            # NOTE(johannes): root_gb is allowed to be 0 for legacy reasons
            # since libvirt interpreted the value differently than other
            # drivers. A value of 0 means don't check size.
            if dest_size != 0:
                if image_size > dest_size:
                    raise exception.FlavorDiskSmallerThanImage(
                        flavor_size=dest_size, image_size=image_size)

                if image_min_disk > dest_size:
                    raise exception.FlavorDiskSmallerThanMinDisk(
                        flavor_size=dest_size, image_min_disk=image_min_disk)
            else:
                # The user is attempting to create a server with a 0-disk
                # image-backed flavor, which can lead to issues with a large
                # image consuming an unexpectedly large amount of local disk
                # on the compute host. Check to see if the deployment will
                # allow that.
                if not context.can(
                        servers_policies.ZERO_DISK_FLAVOR, fatal=False):
                    raise exception.BootFromVolumeRequiredForZeroDiskFlavor()

        API._validate_flavor_image_numa_pci(
            image, flavor, validate_numa=validate_numa,
            validate_pci=validate_pci)

    # TODO(huaqiang): Remove in Wallaby when there is no nova-compute node
    # having a version prior to Victoria.
    @staticmethod
    def _check_compute_service_for_mixed_instance(numa_topology):
        """Check if the nova-compute service is ready to support mixed instance
        when the CPU allocation policy is 'mixed'.
        """
        # No need to check the instance with no NUMA topology associated with.
        if numa_topology is None:
            return

        # No need to check if instance CPU policy is not 'mixed'
        if numa_topology.cpu_policy != fields_obj.CPUAllocationPolicy.MIXED:
            return

        # Catch a request creating a mixed instance, make sure all nova-compute
        # service have been upgraded and support the mixed policy.
        minimal_version = objects.service.get_minimum_version_all_cells(
            nova_context.get_admin_context(), ['nova-compute'])
        if minimal_version < MIN_VER_NOVA_COMPUTE_MIXED_POLICY:
            raise exception.MixedInstanceNotSupportByComputeService()

    @staticmethod
    def _validate_flavor_image_numa_pci(
        image, flavor, validate_numa=True, validate_pci=False,
    ):
        """Validate the flavor and image NUMA/PCI values.

        This is called from the API service to ensure that the flavor
        extra-specs and image properties are self-consistent and compatible
        with each other.

        :param image: a dict representation of the image including properties
        :param flavor: Flavor object
        :param validate_numa: Flag to indicate whether or not to validate
               the NUMA-related metadata.
        :param validate_pci: Flag to indicate whether or not to validate
               the PCI-related metadata.
        :raises: Many different possible exceptions.  See
                 api.openstack.compute.servers.INVALID_FLAVOR_IMAGE_EXCEPTIONS
                 for the full list.
        """
        image_meta = _get_image_meta_obj(image)

        # Only validate values of flavor/image so the return results of
        # following 'get' functions are not used.
        hardware.get_mem_encryption_constraint(flavor, image_meta)
        hardware.get_pmu_constraint(flavor, image_meta)
        hardware.get_number_of_serial_ports(flavor, image_meta)
        hardware.get_realtime_cpu_constraint(flavor, image_meta)
        hardware.get_cpu_topology_constraints(flavor, image_meta)
        hardware.get_vif_multiqueue_constraint(flavor, image_meta)
        if validate_numa:
            hardware.numa_get_constraints(flavor, image_meta)
        if validate_pci:
            pci_request.get_pci_requests_from_flavor(flavor)

    def _get_image_defined_bdms(self, flavor, image_meta, root_device_name):
        image_properties = image_meta.get('properties', {})

        # Get the block device mappings defined by the image.
        image_defined_bdms = image_properties.get('block_device_mapping', [])
        legacy_image_defined = not image_properties.get('bdm_v2', False)

        image_mapping = image_properties.get('mappings', [])

        if legacy_image_defined:
            image_defined_bdms = block_device.from_legacy_mapping(
                image_defined_bdms, None, root_device_name)
        else:
            image_defined_bdms = list(map(block_device.BlockDeviceDict,
                                          image_defined_bdms))

        if image_mapping:
            image_mapping = self._prepare_image_mapping(flavor, image_mapping)
            image_defined_bdms = self._merge_bdms_lists(
                image_mapping, image_defined_bdms)

        return image_defined_bdms

    def _get_flavor_defined_bdms(self, flavor, block_device_mapping):
        flavor_defined_bdms = []

        have_ephemeral_bdms = any(filter(
            block_device.new_format_is_ephemeral, block_device_mapping))
        have_swap_bdms = any(filter(
            block_device.new_format_is_swap, block_device_mapping))

        if flavor.get('ephemeral_gb') and not have_ephemeral_bdms:
            flavor_defined_bdms.append(
                block_device.create_blank_bdm(flavor['ephemeral_gb']))
        if flavor.get('swap') and not have_swap_bdms:
            flavor_defined_bdms.append(
                block_device.create_blank_bdm(flavor['swap'], 'swap'))

        return flavor_defined_bdms

    def _merge_bdms_lists(self, overridable_mappings, overrider_mappings):
        """Override any block devices from the first list by device name

        :param overridable_mappings: list which items are overridden
        :param overrider_mappings: list which items override

        :returns: A merged list of bdms
        """
        device_names = set(bdm['device_name'] for bdm in overrider_mappings
                           if bdm['device_name'])
        return (overrider_mappings +
                [bdm for bdm in overridable_mappings
                 if bdm['device_name'] not in device_names])

    def _check_and_transform_bdm(
        self, context, base_options, flavor, image_meta, min_count, max_count,
        block_device_mapping, legacy_bdm,
    ):
        # NOTE (ndipanov): Assume root dev name is 'vda' if not supplied.
        #                  It's needed for legacy conversion to work.
        root_device_name = (base_options.get('root_device_name') or 'vda')
        image_ref = base_options.get('image_ref', '')
        # If the instance is booted by image and has a volume attached,
        # the volume cannot have the same device name as root_device_name
        if image_ref:
            for bdm in block_device_mapping:
                if (bdm.get('destination_type') == 'volume' and
                    block_device.strip_dev(bdm.get(
                    'device_name')) == root_device_name):
                    msg = _('The volume cannot be assigned the same device'
                            ' name as the root device %s') % root_device_name
                    raise exception.InvalidRequest(msg)

        image_defined_bdms = self._get_image_defined_bdms(
            flavor, image_meta, root_device_name)
        root_in_image_bdms = (
            block_device.get_root_bdm(image_defined_bdms) is not None)

        if legacy_bdm:
            block_device_mapping = block_device.from_legacy_mapping(
                block_device_mapping, image_ref, root_device_name,
                no_root=root_in_image_bdms)
        elif root_in_image_bdms:
            # NOTE (ndipanov): client will insert an image mapping into the v2
            # block_device_mapping, but if there is a bootable device in image
            # mappings - we need to get rid of the inserted image
            # NOTE (gibi): another case is when a server is booted with an
            # image to bdm mapping where the image only contains a bdm to a
            # snapshot. In this case the other image to bdm mapping
            # contains an unnecessary device with boot_index == 0.
            # Also in this case the image_ref is None as we are booting from
            # an image to volume bdm.
            def not_image_and_root_bdm(bdm):
                return not (bdm.get('boot_index') == 0 and
                            bdm.get('source_type') == 'image')

            block_device_mapping = list(
                filter(not_image_and_root_bdm, block_device_mapping))

        block_device_mapping = self._merge_bdms_lists(
            image_defined_bdms, block_device_mapping)

        if min_count > 1 or max_count > 1:
            if any(map(lambda bdm: bdm['source_type'] == 'volume',
                       block_device_mapping)):
                msg = _('Cannot attach one or more volumes to multiple'
                        ' instances')
                raise exception.InvalidRequest(msg)

        block_device_mapping += self._get_flavor_defined_bdms(
            flavor, block_device_mapping)

        return block_device_obj.block_device_make_list_from_dicts(
                context, block_device_mapping)

    def _get_image(self, context, image_href):
        if not image_href:
            return None, {}

        image = self.image_api.get(context, image_href)
        return image['id'], image

    def _checks_for_create_and_rebuild(
        self, context, image_id, image, flavor, metadata, files_to_inject,
        root_bdm, validate_numa=True,
    ):
        self._check_metadata_properties_quota(context, metadata)
        self._check_injected_file_quota(context, files_to_inject)
        self._detect_nonbootable_image_from_properties(image_id, image)
        self._validate_flavor_image(context, image_id, image,
                                    flavor, root_bdm,
                                    validate_numa=validate_numa)

    def _check_support_vnic_accelerator(self, context, requested_networks):
        if requested_networks:
            for request_net in requested_networks:
                if request_net.device_profile:
                    min_version = (objects.service.
                        get_minimum_version_all_cells(
                            context,
                            ['nova-compute']))
                    if min_version < SUPPORT_VNIC_TYPE_ACCELERATOR:
                        msg = ("Port with cyborg profile is not avaliable"
                            " until upgrade finished.")
                        raise exception.ForbiddenPortsWithAccelerator(msg)

    def _check_vnic_remote_managed_min_version(self, context):
        min_version = (objects.service.get_minimum_version_all_cells(
            context, ['nova-compute']))
        if min_version < SUPPORT_VNIC_TYPE_REMOTE_MANAGED:
            msg = ("Remote-managed ports are not supported"
                   " until an upgrade is fully finished.")
            raise exception.ForbiddenWithRemoteManagedPorts(msg)

    def _check_support_vnic_remote_managed(self, context, requested_networks):
        if requested_networks:
            for request_net in requested_networks:
                if (request_net.port_id and
                        self.network_api.is_remote_managed_port(
                            context, request_net.port_id)):
                    self._check_vnic_remote_managed_min_version(context)

    def _validate_and_build_base_options(
        self, context, flavor, boot_meta, image_href, image_id, kernel_id,
        ramdisk_id, display_name, display_description, hostname, key_name,
        key_data, security_groups, availability_zone, user_data, metadata,
        access_ip_v4, access_ip_v6, requested_networks, config_drive,
        auto_disk_config, reservation_id, max_count,
        supports_port_resource_request,
    ):
        """Verify all the input parameters regardless of the provisioning
        strategy being performed.
        """
        if flavor['disabled']:
            raise exception.FlavorNotFound(flavor_id=flavor['id'])

        if user_data:
            try:
                base64utils.decode_as_bytes(user_data)
            except TypeError:
                raise exception.InstanceUserDataMalformed()

        # When using Neutron, _check_requested_secgroups will translate and
        # return any requested security group names to uuids.
        security_groups = self._check_requested_secgroups(
            context, security_groups)

        # Note:  max_count is the number of instances requested by the user,
        # max_network_count is the maximum number of instances taking into
        # account any network quotas
        max_network_count = self._check_requested_networks(
            context, requested_networks, max_count)

        kernel_id, ramdisk_id = self._handle_kernel_and_ramdisk(
            context, kernel_id, ramdisk_id, boot_meta)

        config_drive = self._check_config_drive(config_drive)

        if key_data is None and key_name is not None:
            key_pair = objects.KeyPair.get_by_name(context,
                                                   context.user_id,
                                                   key_name)
            key_data = key_pair.public_key
        else:
            key_pair = None

        root_device_name = block_device.prepend_dev(
                block_device.properties_root_device_name(
                    boot_meta.get('properties', {})))

        image_meta = _get_image_meta_obj(boot_meta)
        numa_topology = hardware.numa_get_constraints(flavor, image_meta)

        system_metadata = {}

        pci_numa_affinity_policy = hardware.get_pci_numa_policy_constraint(
            flavor, image_meta)

        # PCI requests come from two sources: instance flavor and
        # requested_networks. The first call in below returns an
        # InstancePCIRequests object which is a list of InstancePCIRequest
        # objects. The second call in below creates an InstancePCIRequest
        # object for each SR-IOV port, and append it to the list in the
        # InstancePCIRequests object
        pci_request_info = pci_request.get_pci_requests_from_flavor(
            flavor, affinity_policy=pci_numa_affinity_policy)
        result = self.network_api.create_resource_requests(
            context, requested_networks, pci_request_info,
            affinity_policy=pci_numa_affinity_policy)
        network_metadata, port_resource_requests, req_lvl_params = result

        self._check_support_vnic_accelerator(context, requested_networks)
        self._check_support_vnic_remote_managed(context, requested_networks)

        # Creating servers with ports that have resource requests, like QoS
        # minimum bandwidth rules, is only supported in a requested minimum
        # microversion.
        if port_resource_requests and not supports_port_resource_request:
            raise exception.CreateWithPortResourceRequestOldVersion()

        # TODO(gibi): remove this when Nova does not need to support Wallaby
        # computes any more.
        if (port_resource_requests and
            self.network_api.has_extended_resource_request_extension(context)
        ):
            # we only support the extended resource request if the computes are
            # upgraded to Xena.
            min_version = objects.service.get_minimum_version_all_cells(
                context, ["nova-compute"])
            if min_version < MIN_COMPUTE_BOOT_WITH_EXTENDED_RESOURCE_REQUEST:
                raise exception.ExtendedResourceRequestOldCompute()

        base_options = {
            'reservation_id': reservation_id,
            'image_ref': image_href,
            'kernel_id': kernel_id or '',
            'ramdisk_id': ramdisk_id or '',
            'power_state': power_state.NOSTATE,
            'vm_state': vm_states.BUILDING,
            'config_drive': config_drive,
            'user_id': context.user_id,
            'project_id': context.project_id,
            'instance_type_id': flavor['id'],
            'memory_mb': flavor['memory_mb'],
            'vcpus': flavor['vcpus'],
            'root_gb': flavor['root_gb'],
            'ephemeral_gb': flavor['ephemeral_gb'],
            'display_name': display_name,
            'display_description': display_description,
            'hostname': hostname,
            'user_data': user_data,
            'key_name': key_name,
            'key_data': key_data,
            'locked': False,
            'metadata': metadata or {},
            'access_ip_v4': access_ip_v4,
            'access_ip_v6': access_ip_v6,
            'availability_zone': availability_zone,
            'root_device_name': root_device_name,
            'progress': 0,
            'pci_requests': pci_request_info,
            'numa_topology': numa_topology,
            'system_metadata': system_metadata,
            'port_resource_requests': port_resource_requests,
            'request_level_params': req_lvl_params,
        }

        options_from_image = self._inherit_properties_from_image(
                boot_meta, auto_disk_config)

        base_options.update(options_from_image)

        # return the validated options and maximum number of instances allowed
        # by the network quotas
        return (base_options, max_network_count, key_pair, security_groups,
                network_metadata)

    @staticmethod
    @api_db_api.context_manager.writer
    def _create_reqspec_buildreq_instmapping(context, rs, br, im):
        """Create the request spec, build request, and instance mapping in a
        single database transaction.

        The RequestContext must be passed in to this method so that the
        database transaction context manager decorator will nest properly and
        include each create() into the same transaction context.
        """
        rs.create()
        br.create()
        im.create()

    def _validate_host_or_node(self, context, host, hypervisor_hostname):
        """Check whether compute nodes exist by validating the host
        and/or the hypervisor_hostname. There are three cases:
        1. If only host is supplied, we can lookup the HostMapping in
        the API DB.
        2. If only node is supplied, we can query a resource provider
        with that name in placement.
        3. If both host and node are supplied, we can get the cell from
        HostMapping and from that lookup the ComputeNode with the
        given cell.

        :param context: The API request context.
        :param host: Target host.
        :param hypervisor_hostname: Target node.
        :raises: ComputeHostNotFound if we find no compute nodes with host
                 and/or hypervisor_hostname.
        """

        if host:
            # When host is specified.
            try:
                host_mapping = objects.HostMapping.get_by_host(context, host)
            except exception.HostMappingNotFound:
                LOG.warning('No host-to-cell mapping found for host '
                            '%(host)s.', {'host': host})
                raise exception.ComputeHostNotFound(host=host)
            # When both host and node are specified.
            if hypervisor_hostname:
                cell = host_mapping.cell_mapping
                with nova_context.target_cell(context, cell) as cctxt:
                    # Here we only do an existence check, so we don't
                    # need to store the return value into a variable.
                    objects.ComputeNode.get_by_host_and_nodename(
                        cctxt, host, hypervisor_hostname)
        elif hypervisor_hostname:
            # When only node is specified.
            try:
                self.placementclient.get_provider_by_name(
                    context, hypervisor_hostname)
            except exception.ResourceProviderNotFound:
                raise exception.ComputeHostNotFound(host=hypervisor_hostname)

    def _get_volumes_for_bdms(self, context, bdms):
        """Get the pre-existing volumes from cinder for the list of BDMs.

        :param context: nova auth RequestContext
        :param bdms: BlockDeviceMappingList which has zero or more BDMs with
            a pre-existing volume_id specified.
        :return: dict, keyed by volume id, of volume dicts
        :raises: VolumeNotFound - if a given volume does not exist
        :raises: CinderConnectionFailed - if there are problems communicating
            with the cinder API
        :raises: Forbidden - if the user token does not have authority to see
            a volume
        """
        volumes = {}
        for bdm in bdms:
            if bdm.volume_id:
                volumes[bdm.volume_id] = self.volume_api.get(
                    context, bdm.volume_id)
        return volumes

    @staticmethod
    def _validate_vol_az_for_create(instance_az, volumes):
        """Performs cross_az_attach validation for the instance and volumes.

        If [cinder]/cross_az_attach=True (default) this method is a no-op.

        If [cinder]/cross_az_attach=False, this method will validate that:

        1. All volumes are in the same availability zone.
        2. The volume AZ matches the instance AZ. If the instance is being
           created without a specific AZ (either via the user request or the
           [DEFAULT]/default_schedule_zone option), and the volume AZ matches
           [DEFAULT]/default_availability_zone for compute services, then the
           method returns the volume AZ so it can be set in the RequestSpec as
           if the user requested the zone explicitly.

        :param instance_az: Availability zone for the instance. In this case
            the host is not yet selected so the instance AZ value should come
            from one of the following cases:

            * The user requested availability zone.
            * [DEFAULT]/default_schedule_zone (defaults to None) if the request
              does not specify an AZ (see parse_availability_zone).
        :param volumes: iterable of dicts of cinder volumes to be attached to
            the server being created
        :returns: None or volume AZ to set in the RequestSpec for the instance
        :raises: MismatchVolumeAZException if the instance and volume AZ do
            not match
        """
        if CONF.cinder.cross_az_attach:
            return

        if not volumes:
            return

        # First make sure that all of the volumes are in the same zone.
        vol_zones = [vol['availability_zone'] for vol in volumes]
        if len(set(vol_zones)) > 1:
            msg = (_("Volumes are in different availability zones: %s")
                   % ','.join(vol_zones))
            raise exception.MismatchVolumeAZException(reason=msg)

        volume_az = vol_zones[0]
        # In this case the instance.host should not be set so the instance AZ
        # value should come from instance.availability_zone which will be one
        # of the following cases:
        # * The user requested availability zone.
        # * [DEFAULT]/default_schedule_zone (defaults to None) if the request
        #   does not specify an AZ (see parse_availability_zone).

        # If the instance is not being created with a specific AZ (the AZ is
        # input via the API create request *or* [DEFAULT]/default_schedule_zone
        # is not None), then check to see if we should use the default AZ
        # (which by default matches the default AZ in Cinder, i.e. 'nova').
        if instance_az is None:
            # Check if the volume AZ is the same as our default AZ for compute
            # hosts (nova) and if so, assume we are OK because the user did not
            # request an AZ and will get the same default. If the volume AZ is
            # not the same as our default, return the volume AZ so the caller
            # can put it into the request spec so the instance is scheduled
            # to the same zone as the volume. Note that we are paranoid about
            # the default here since both nova and cinder's default backend AZ
            # is "nova" and we do not want to pin the server to that AZ since
            # it's special, i.e. just like we tell users in the docs to not
            # specify availability_zone='nova' when creating a server since we
            # might not be able to migrate it later.
            if volume_az != CONF.default_availability_zone:
                return volume_az  # indication to set in request spec
            # The volume AZ is the same as the default nova AZ so we will be OK
            return

        if instance_az != volume_az:
            msg = _("Server and volumes are not in the same availability "
                    "zone. Server is in: %(instance_az)s. Volumes are in: "
                    "%(volume_az)s") % {
                'instance_az': instance_az, 'volume_az': volume_az}
            raise exception.MismatchVolumeAZException(reason=msg)

    def _provision_instances(
        self, context, flavor, min_count,
        max_count, base_options, boot_meta, security_groups,
        block_device_mapping, shutdown_terminate,
        instance_group, check_server_group_quota, filter_properties,
        key_pair, tags, trusted_certs, supports_multiattach,
        network_metadata=None, requested_host=None,
        requested_hypervisor_hostname=None,
    ):
        # NOTE(boxiang): Check whether compute nodes exist by validating
        # the host and/or the hypervisor_hostname. Pass the destination
        # to the scheduler with host and/or hypervisor_hostname(node).
        destination = None
        if requested_host or requested_hypervisor_hostname:
            self._validate_host_or_node(context, requested_host,
                                        requested_hypervisor_hostname)
            destination = objects.Destination()
            if requested_host:
                destination.host = requested_host
            destination.node = requested_hypervisor_hostname
        # Check quotas
        num_instances = compute_utils.check_num_instances_quota(
            context, flavor, min_count, max_count)

        # Find out whether or not we are a BFV instance
        if block_device_mapping:
            root = block_device_mapping.root_bdm()
            is_bfv = bool(root and root.is_volume)
        else:
            # If we have no BDMs, we're clearly not BFV
            is_bfv = False

        # NOTE(johngarbutt) when unified limits not used, this just
        #   returns num_instances back again
        # NOTE: If we want to enforce quota on port or cyborg resources in the
        # future, this enforce call will need to move after we have populated
        # the RequestSpec with all of the requested resources and use the real
        # RequestSpec to get the overall resource usage of the instance.
        num_instances = placement_limits.enforce_num_instances_and_flavor(
                context, context.project_id, flavor,
                is_bfv, min_count, num_instances)

        security_groups = security_group_api.populate_security_groups(
            security_groups)
        port_resource_requests = base_options.pop('port_resource_requests')
        req_lvl_params = base_options.pop('request_level_params')
        instances_to_build = []
        # We could be iterating over several instances with several BDMs per
        # instance and those BDMs could be using a lot of the same images so
        # we want to cache the image API GET results for performance.
        image_cache = {}  # dict of image dicts keyed by image id
        # Before processing the list of instances get all of the requested
        # pre-existing volumes so we can do some validation here rather than
        # down in the bowels of _validate_bdm.
        volumes = self._get_volumes_for_bdms(context, block_device_mapping)
        volume_az = self._validate_vol_az_for_create(
            base_options['availability_zone'], volumes.values())
        if volume_az:
            # This means the instance is not being created in a specific zone
            # but needs to match the zone that the volumes are in so update
            # base_options to match the volume zone.
            base_options['availability_zone'] = volume_az
        LOG.debug("Going to run %s instances...", num_instances)
        extra_specs = flavor.extra_specs
        dp_name = extra_specs.get('accel:device_profile')
        dp_request_groups = []
        if dp_name:
            dp_request_groups = cyborg.get_device_profile_request_groups(
                context, dp_name)
        try:
            for idx in range(num_instances):
                # Create a uuid for the instance so we can store the
                # RequestSpec before the instance is created.
                instance_uuid = uuidutils.generate_uuid()
                # Store the RequestSpec that will be used for scheduling.
                req_spec = objects.RequestSpec.from_components(
                    context,
                    instance_uuid, boot_meta, flavor,
                    base_options['numa_topology'],
                    base_options['pci_requests'], filter_properties,
                    instance_group, base_options['availability_zone'],
                    security_groups=security_groups,
                    port_resource_requests=port_resource_requests,
                    request_level_params=req_lvl_params)
                req_spec.is_bfv = is_bfv

                # NOTE(danms): We need to record num_instances on the request
                # spec as this is how the conductor knows how many were in this
                # batch.
                req_spec.num_instances = num_instances

                # NOTE(stephenfin): The network_metadata field is not persisted
                # inside RequestSpec object.
                if network_metadata:
                    req_spec.network_metadata = network_metadata

                if destination:
                    req_spec.requested_destination = destination

                if dp_request_groups:
                    req_spec.requested_resources.extend(dp_request_groups)

                # Create an instance object, but do not store in db yet.
                instance = objects.Instance(context=context)
                instance.uuid = instance_uuid
                instance.update(base_options)
                instance.keypairs = objects.KeyPairList(objects=[])
                if key_pair:
                    instance.keypairs.objects.append(key_pair)

                instance.trusted_certs = self._retrieve_trusted_certs_object(
                    context, trusted_certs)

                self._populate_instance_for_create(
                    context, instance, boot_meta, idx,
                    security_groups, flavor,
                    num_instances, shutdown_terminate)

                block_device_mapping = (
                    self._bdm_validate_set_size_and_instance(context,
                        instance, flavor, block_device_mapping,
                        image_cache, volumes, supports_multiattach))
                instance_tags = self._transform_tags(tags, instance.uuid)

                build_request = objects.BuildRequest(context,
                        instance=instance, instance_uuid=instance.uuid,
                        project_id=instance.project_id,
                        block_device_mappings=block_device_mapping,
                        tags=instance_tags)

                # Create an instance_mapping.  The null cell_mapping indicates
                # that the instance doesn't yet exist in a cell, and lookups
                # for it need to instead look for the RequestSpec.
                # cell_mapping will be populated after scheduling, with a
                # scheduling failure using the cell_mapping for the special
                # cell0.
                inst_mapping = objects.InstanceMapping(context=context)
                inst_mapping.instance_uuid = instance_uuid
                inst_mapping.project_id = context.project_id
                inst_mapping.user_id = context.user_id
                inst_mapping.cell_mapping = None

                # Create the request spec, build request, and instance mapping
                # records in a single transaction so that if a DBError is
                # raised from any of them, all INSERTs will be rolled back and
                # no orphaned records will be left behind.
                self._create_reqspec_buildreq_instmapping(context, req_spec,
                                                          build_request,
                                                          inst_mapping)

                instances_to_build.append(
                    (req_spec, build_request, inst_mapping))

                if instance_group:
                    if check_server_group_quota:
                        try:
                            objects.Quotas.check_deltas(
                                context, {'server_group_members': 1},
                                instance_group, context.user_id)
                            local_limit.enforce_db_limit(
                                context, local_limit.SERVER_GROUP_MEMBERS,
                                entity_scope=instance_group.uuid, delta=1)
                        except exception.GroupMemberLimitExceeded:
                            raise
                        except exception.OverQuota:
                            msg = _("Quota exceeded, too many servers in "
                                    "group")
                            raise exception.OverQuota(msg)

                    members = objects.InstanceGroup.add_members(
                        context, instance_group.uuid, [instance.uuid])

                    # NOTE(melwitt): We recheck the quota after creating the
                    # object to prevent users from allocating more resources
                    # than their allowed quota in the event of a race. This is
                    # configurable because it can be expensive if strict quota
                    # limits are not required in a deployment.
                    if CONF.quota.recheck_quota and check_server_group_quota:
                        try:
                            objects.Quotas.check_deltas(
                                context, {'server_group_members': 0},
                                instance_group, context.user_id)
                            # TODO(johngarbutt): decide if we need this check
                            # The quota rechecking of limits is really just to
                            # protect against denial of service attacks that
                            # aim to fill up the database. Its usefulness could
                            # be debated.
                            local_limit.enforce_db_limit(
                                context, local_limit.SERVER_GROUP_MEMBERS,
                                entity_scope=instance_group.uuid, delta=0)
                        except exception.GroupMemberLimitExceeded:
                            with excutils.save_and_reraise_exception():
                                objects.InstanceGroup._remove_members_in_db(
                                    context, instance_group.id,
                                    [instance.uuid])
                        except exception.OverQuota:
                            objects.InstanceGroup._remove_members_in_db(
                                context, instance_group.id, [instance.uuid])
                            msg = _("Quota exceeded, too many servers in "
                                    "group")
                            raise exception.OverQuota(msg)
                    # list of members added to servers group in this iteration
                    # is needed to check quota of server group during add next
                    # instance
                    instance_group.members.extend(members)

        # In the case of any exceptions, attempt DB cleanup
        except Exception:
            with excutils.save_and_reraise_exception():
                self._cleanup_build_artifacts(None, instances_to_build)

        return instances_to_build

    @staticmethod
    def _retrieve_trusted_certs_object(context, trusted_certs, rebuild=False):
        """Convert user-requested trusted cert IDs to TrustedCerts object

        Also validates that the deployment is new enough to support trusted
        image certification validation.

        :param context: The user request auth context
        :param trusted_certs: list of user-specified trusted cert string IDs,
            may be None
        :param rebuild: True if rebuilding the server, False if creating a
            new server
        :returns: nova.objects.TrustedCerts object or None if no user-specified
            trusted cert IDs were given and nova is not configured with
            default trusted cert IDs
        """
        # Retrieve trusted_certs parameter, or use CONF value if certificate
        # validation is enabled
        if trusted_certs:
            certs_to_return = objects.TrustedCerts(ids=trusted_certs)
        elif (CONF.glance.verify_glance_signatures and
              CONF.glance.enable_certificate_validation and
              CONF.glance.default_trusted_certificate_ids):
            certs_to_return = objects.TrustedCerts(
                ids=CONF.glance.default_trusted_certificate_ids)
        else:
            return None

        return certs_to_return

    @staticmethod
    def _get_requested_instance_group(context, filter_properties):
        if (not filter_properties or
                not filter_properties.get('scheduler_hints')):
            return

        group_hint = filter_properties.get('scheduler_hints').get('group')
        if not group_hint:
            return

        return objects.InstanceGroup.get_by_uuid(context, group_hint)

    def _update_ephemeral_encryption_bdms(
        self,
        flavor: 'objects.Flavor',
        image_meta_dict: ty.Dict[str, ty.Any],
        block_device_mapping: 'objects.BlockDeviceMappingList',
    ) -> None:
        """Update local BlockDeviceMappings when ephemeral encryption requested

        Enable ephemeral encryption in all local BlockDeviceMappings
        when requested in the flavor or image. Also optionally set the format
        and options if also provided.

        :param flavor: The instance flavor for the request
        :param image_meta_dict: The image metadata for the request
        :block_device_mapping: The current block_device_mapping for the request
        """
        image_meta = _get_image_meta_obj(image_meta_dict)
        if not hardware.get_ephemeral_encryption_constraint(
                flavor, image_meta):
            return

        # NOTE(lyarwood): Attempt to find the format in the flavor and image,
        # if one isn't found then the compute will need to provide and save a
        # default format during a the initial build.
        eph_format = hardware.get_ephemeral_encryption_format(
            flavor, image_meta)

        # NOTE(lyarwood): The term ephemeral is overloaded in the codebase,
        # what it actually means in the context of ephemeral encryption is
        # anything local to the compute host so use the is_local property.
        # TODO(lyarwood): Add .get_local_devices() to BlockDeviceMappingList
        for bdm in [b for b in block_device_mapping if b.is_local]:
            bdm.encrypted = True
            if eph_format:
                bdm.encryption_format = eph_format

    def _create_instance(self, context, flavor,
               image_href, kernel_id, ramdisk_id,
               min_count, max_count,
               display_name, display_description, hostname,
               key_name, key_data, security_groups,
               availability_zone, user_data, metadata, injected_files,
               admin_password, access_ip_v4, access_ip_v6,
               requested_networks, config_drive,
               block_device_mapping, auto_disk_config, filter_properties,
               reservation_id=None, legacy_bdm=True, shutdown_terminate=False,
               check_server_group_quota=False, tags=None,
               supports_multiattach=False, trusted_certs=None,
               supports_port_resource_request=False,
               requested_host=None, requested_hypervisor_hostname=None):
        """Verify all the input parameters regardless of the provisioning
        strategy being performed and schedule the instance(s) for
        creation.
        """

        # Normalize and setup some parameters
        if reservation_id is None:
            reservation_id = utils.generate_uid('r')
        security_groups = security_groups or ['default']
        min_count = min_count or 1
        max_count = max_count or min_count
        block_device_mapping = block_device_mapping or []
        tags = tags or []

        if image_href:
            image_id, boot_meta = self._get_image(context, image_href)
        else:
            # This is similar to the logic in _retrieve_trusted_certs_object.
            if (trusted_certs or
                (CONF.glance.verify_glance_signatures and
                 CONF.glance.enable_certificate_validation and
                 CONF.glance.default_trusted_certificate_ids)):
                msg = _("Image certificate validation is not supported "
                        "when booting from volume")
                raise exception.CertificateValidationFailed(message=msg)
            image_id = None
            boot_meta = block_device.get_bdm_image_metadata(
                context, self.image_api, self.volume_api, block_device_mapping,
                legacy_bdm)

        self._check_auto_disk_config(image=boot_meta,
                                     auto_disk_config=auto_disk_config)

        (
            base_options, max_net_count, key_pair, security_groups,
            network_metadata,
        ) = self._validate_and_build_base_options(
            context, flavor, boot_meta, image_href, image_id,
            kernel_id, ramdisk_id, display_name, display_description,
            hostname, key_name, key_data, security_groups, availability_zone,
            user_data, metadata, access_ip_v4, access_ip_v6,
            requested_networks, config_drive, auto_disk_config,
            reservation_id, max_count, supports_port_resource_request,
        )

        # TODO(huaqiang): Remove in Wallaby
        # check nova-compute nodes have been updated to Victoria to support the
        # mixed CPU policy for creating a new instance.
        numa_topology = base_options.get('numa_topology')
        self._check_compute_service_for_mixed_instance(numa_topology)

        # max_net_count is the maximum number of instances requested by the
        # user adjusted for any network quota constraints, including
        # consideration of connections to each requested network
        if max_net_count < min_count:
            raise exception.PortLimitExceeded()
        elif max_net_count < max_count:
            LOG.info("max count reduced from %(max_count)d to "
                     "%(max_net_count)d due to network port quota",
                     {'max_count': max_count,
                      'max_net_count': max_net_count})
            max_count = max_net_count

        # _check_and_transform_bdm transforms block_device_mapping from API
        # bdms (dicts) to a BlockDeviceMappingList.
        block_device_mapping = self._check_and_transform_bdm(context,
            base_options, flavor, boot_meta, min_count, max_count,
            block_device_mapping, legacy_bdm)

        # Update any local BlockDeviceMapping objects if ephemeral encryption
        # has been requested though flavor extra specs or image properties
        self._update_ephemeral_encryption_bdms(
            flavor, boot_meta, block_device_mapping)

        # We can't do this check earlier because we need bdms from all sources
        # to have been merged in order to get the root bdm.
        # Set validate_numa=False since numa validation is already done by
        # _validate_and_build_base_options().
        self._checks_for_create_and_rebuild(context, image_id, boot_meta,
                flavor, metadata, injected_files,
                block_device_mapping.root_bdm(), validate_numa=False)

        instance_group = self._get_requested_instance_group(
            context, filter_properties)

        tags = self._create_tag_list_obj(context, tags)

        instances_to_build = self._provision_instances(
            context, flavor, min_count, max_count, base_options,
            boot_meta, security_groups, block_device_mapping,
            shutdown_terminate, instance_group, check_server_group_quota,
            filter_properties, key_pair, tags, trusted_certs,
            supports_multiattach, network_metadata,
            requested_host, requested_hypervisor_hostname)

        instances = []
        request_specs = []
        build_requests = []
        for rs, build_request, im in instances_to_build:
            build_requests.append(build_request)
            instance = build_request.get_new_instance(context)
            instances.append(instance)
            # NOTE(sbauza): Add the requested networks so the related scheduler
            # pre-filter can verify them
            if requested_networks is not None:
                rs.requested_networks = requested_networks
            request_specs.append(rs)

        self.compute_task_api.schedule_and_build_instances(
            context,
            build_requests=build_requests,
            request_spec=request_specs,
            image=boot_meta,
            admin_password=admin_password,
            injected_files=injected_files,
            requested_networks=requested_networks,
            block_device_mapping=block_device_mapping,
            tags=tags)

        return instances, reservation_id

    @staticmethod
    def _cleanup_build_artifacts(instances, instances_to_build):
        # instances_to_build is a list of tuples:
        # (RequestSpec, BuildRequest, InstanceMapping)

        # Be paranoid about artifacts being deleted underneath us.
        for instance in instances or []:
            try:
                instance.destroy()
            except exception.InstanceNotFound:
                pass
        for rs, build_request, im in instances_to_build or []:
            try:
                rs.destroy()
            except exception.RequestSpecNotFound:
                pass
            try:
                build_request.destroy()
            except exception.BuildRequestNotFound:
                pass
            try:
                im.destroy()
            except exception.InstanceMappingNotFound:
                pass

    @staticmethod
    def _volume_size(flavor, bdm):
        size = bdm.get('volume_size')
        # NOTE (ndipanov): inherit flavor size only for swap and ephemeral
        if (size is None and bdm.get('source_type') == 'blank' and
                bdm.get('destination_type') == 'local'):
            if bdm.get('guest_format') == 'swap':
                size = flavor.get('swap', 0)
            else:
                size = flavor.get('ephemeral_gb', 0)
        return size

    def _prepare_image_mapping(self, flavor, mappings):
        """Extract and format blank devices from image mappings."""

        prepared_mappings = []

        for bdm in block_device.mappings_prepend_dev(mappings):
            LOG.debug("Image bdm %s", bdm)

            virtual_name = bdm['virtual']
            if virtual_name == 'ami' or virtual_name == 'root':
                continue

            if not block_device.is_swap_or_ephemeral(virtual_name):
                continue

            guest_format = bdm.get('guest_format')
            if virtual_name == 'swap':
                guest_format = 'swap'
            if not guest_format:
                guest_format = CONF.default_ephemeral_format

            values = block_device.BlockDeviceDict({
                'device_name': bdm['device'],
                'source_type': 'blank',
                'destination_type': 'local',
                'device_type': 'disk',
                'guest_format': guest_format,
                'delete_on_termination': True,
                'boot_index': -1})

            values['volume_size'] = self._volume_size(
                flavor, values)
            if values['volume_size'] == 0:
                continue

            prepared_mappings.append(values)

        return prepared_mappings

    def _bdm_validate_set_size_and_instance(self, context, instance,
                                            flavor,
                                            block_device_mapping,
                                            image_cache, volumes,
                                            supports_multiattach=False):
        """Ensure the bdms are valid, then set size and associate with instance

        Because this method can be called multiple times when more than one
        instance is booted in a single request it makes a copy of the bdm list.

        :param context: nova auth RequestContext
        :param instance: Instance object
        :param flavor: Flavor object - used for swap and ephemeral BDMs
        :param block_device_mapping: BlockDeviceMappingList object
        :param image_cache: dict of image dicts keyed by id which is used as a
            cache in case there are multiple BDMs in the same request using
            the same image to avoid redundant GET calls to the image service
        :param volumes: dict, keyed by volume id, of volume dicts from cinder
        :param supports_multiattach: True if the request supports multiattach
            volumes, False otherwise
        """
        LOG.debug("block_device_mapping %s", list(block_device_mapping),
                  instance_uuid=instance.uuid)
        self._validate_bdm(
            context, instance, flavor, block_device_mapping,
            image_cache, volumes, supports_multiattach)
        instance_block_device_mapping = block_device_mapping.obj_clone()
        for bdm in instance_block_device_mapping:
            bdm.volume_size = self._volume_size(flavor, bdm)
            bdm.instance_uuid = instance.uuid
        return instance_block_device_mapping

    @staticmethod
    def _check_requested_volume_type(bdm, volume_type_id_or_name,
                                     volume_types):
        """If we are specifying a volume type, we need to get the
        volume type details from Cinder and make sure the ``volume_type``
        is available.
        """

        # NOTE(brinzhang): Verify that the specified volume type exists.
        # And save the volume type name internally for consistency in the
        # BlockDeviceMapping object.
        for vol_type in volume_types:
            if (volume_type_id_or_name == vol_type['id'] or
                        volume_type_id_or_name == vol_type['name']):
                bdm.volume_type = vol_type['name']
                break
        else:
            raise exception.VolumeTypeNotFound(
                id_or_name=volume_type_id_or_name)

    def _validate_bdm(
        self, context, instance, flavor, block_device_mappings, image_cache,
        volumes, supports_multiattach=False,
    ):
        """Validate requested block device mappings.

        :param context: nova auth RequestContext
        :param instance: Instance object
        :param flavor: Flavor object - used for swap and ephemeral BDMs
        :param block_device_mappings: BlockDeviceMappingList object
        :param image_cache: dict of image dicts keyed by id which is used as a
            cache in case there are multiple BDMs in the same request using
            the same image to avoid redundant GET calls to the image service
        :param volumes: dict, keyed by volume id, of volume dicts from cinder
        :param supports_multiattach: True if the request supports multiattach
            volumes, False otherwise
        """
        # Make sure that the boot indexes make sense.
        # Setting a negative value or None indicates that the device should not
        # be used for booting.
        boot_indexes = sorted([bdm.boot_index
                               for bdm in block_device_mappings
                               if bdm.boot_index is not None and
                               bdm.boot_index >= 0])

        # Each device which is capable of being used as boot device should
        # be given a unique boot index, starting from 0 in ascending order, and
        # there needs to be at least one boot device.
        if not boot_indexes or any(i != v for i, v in enumerate(boot_indexes)):
            # Convert the BlockDeviceMappingList to a list for repr details.
            LOG.debug('Invalid block device mapping boot sequence for '
                      'instance: %s', list(block_device_mappings),
                      instance=instance)
            raise exception.InvalidBDMBootSequence()

        volume_types = None
        for bdm in block_device_mappings:
            volume_type = bdm.volume_type
            if volume_type:
                if not volume_types:
                    # In order to reduce the number of hit cinder APIs,
                    # initialize our cache of volume types.
                    volume_types = self.volume_api.get_all_volume_types(
                        context)
                # NOTE(brinzhang): Ensure the validity of volume_type.
                self._check_requested_volume_type(bdm, volume_type,
                                                  volume_types)

            # NOTE(vish): For now, just make sure the volumes are accessible.
            # Additionally, check that the volume can be attached to this
            # instance.
            snapshot_id = bdm.snapshot_id
            volume_id = bdm.volume_id
            image_id = bdm.image_id
            if image_id is not None:
                if (image_id != instance.get('image_ref') and
                        image_id not in image_cache):
                    try:
                        # Cache the results of the image GET so we do not make
                        # the same request for the same image if processing
                        # multiple BDMs or multiple servers with the same image
                        image_cache[image_id] = self._get_image(
                            context, image_id)
                    except Exception:
                        raise exception.InvalidBDMImage(id=image_id)
                if (bdm.source_type == 'image' and
                        bdm.destination_type == 'volume' and
                        not bdm.volume_size):
                    raise exception.InvalidBDM(message=_("Images with "
                        "destination_type 'volume' need to have a non-zero "
                        "size specified"))
            elif volume_id is not None:
                try:
                    volume = volumes[volume_id]
                    # We do not validate the instance and volume AZ here
                    # because that is done earlier by _provision_instances.
                    self._check_attach_and_reserve_volume(
                        context, volume, instance, bdm, supports_multiattach,
                        validate_az=False)
                    bdm.volume_size = volume.get('size')
                except (exception.CinderConnectionFailed,
                        exception.InvalidVolume,
                        exception.MultiattachNotSupportedOldMicroversion):
                    raise
                except exception.InvalidInput as exc:
                    raise exception.InvalidVolume(reason=exc.format_message())
                except Exception as e:
                    LOG.info('Failed validating volume %s. Error: %s',
                             volume_id, e)
                    raise exception.InvalidBDMVolume(id=volume_id)
            elif snapshot_id is not None:
                try:
                    snap = self.volume_api.get_snapshot(context, snapshot_id)
                    bdm.volume_size = bdm.volume_size or snap.get('size')
                except exception.CinderConnectionFailed:
                    raise
                except Exception:
                    raise exception.InvalidBDMSnapshot(id=snapshot_id)
            elif (bdm.source_type == 'blank' and
                    bdm.destination_type == 'volume' and
                    not bdm.volume_size):
                raise exception.InvalidBDM(message=_("Blank volumes "
                    "(source: 'blank', dest: 'volume') need to have non-zero "
                    "size"))

            # NOTE(lyarwood): Ensure the disk_bus is at least known to Nova.
            # The virt driver may reject this later but for now just ensure
            # it's listed as an acceptable value of the DiskBus field class.
            disk_bus = bdm.disk_bus if 'disk_bus' in bdm else None
            if disk_bus and disk_bus not in fields_obj.DiskBus.ALL:
                raise exception.InvalidBDMDiskBus(disk_bus=disk_bus)

        ephemeral_size = sum(bdm.volume_size or flavor['ephemeral_gb']
                for bdm in block_device_mappings
                if block_device.new_format_is_ephemeral(bdm))
        if ephemeral_size > flavor['ephemeral_gb']:
            raise exception.InvalidBDMEphemeralSize()

        # There should be only one swap
        swap_list = block_device.get_bdm_swap_list(block_device_mappings)
        if len(swap_list) > 1:
            msg = _("More than one swap drive requested.")
            raise exception.InvalidBDMFormat(details=msg)

        if swap_list:
            swap_size = swap_list[0].volume_size or 0
            if swap_size > flavor['swap']:
                raise exception.InvalidBDMSwapSize()

        max_local = CONF.max_local_block_devices
        if max_local >= 0:
            num_local = len([bdm for bdm in block_device_mappings
                             if bdm.destination_type == 'local'])
            if num_local > max_local:
                raise exception.InvalidBDMLocalsLimit()

    def _populate_instance_names(self, instance, num_instances, index):
        """Populate instance display_name and hostname.

        :param instance: The instance to set the display_name, hostname for
        :type instance: nova.objects.Instance
        :param num_instances: Total number of instances being created in this
            request
        :param index: The 0-based index of this particular instance
        """
        # NOTE(mriedem): This is only here for test simplicity since a server
        # name is required in the REST API.
        if 'display_name' not in instance or instance.display_name is None:
            instance.display_name = 'Server %s' % instance.uuid

        # only set the hostname if the user hasn't already requested one
        if 'hostname' not in instance or not instance.hostname:
            # if we're booting multiple instances, we need to add an indexing
            # suffix to both instance.hostname and instance.display_name.
            # This is not necessary for a single instance.
            hostname = utils.sanitize_hostname(instance.display_name)
            if not hostname:
                hostname = f'Server-{instance.uuid}'
            elif num_instances > 1:
                hostname = f'{hostname}-{index + 1}'

            instance.hostname = hostname

        if num_instances > 1:
            instance.display_name = f'{instance.display_name}-{index + 1}'

    def _populate_instance_for_create(
        self, context, instance, image, index, security_groups, flavor,
        num_instances, shutdown_terminate,
    ):
        """Build the beginning of a new instance."""

        instance.launch_index = index
        instance.vm_state = vm_states.BUILDING
        instance.task_state = task_states.SCHEDULING
        info_cache = objects.InstanceInfoCache()
        info_cache.instance_uuid = instance.uuid
        info_cache.network_info = network_model.NetworkInfo()
        instance.info_cache = info_cache
        instance.flavor = flavor
        instance.old_flavor = None
        instance.new_flavor = None
        if CONF.ephemeral_storage_encryption.enabled:
            # NOTE(kfarr): dm-crypt expects the cipher in a
            # hyphenated format: cipher-chainmode-ivmode
            # (ex: aes-xts-plain64). The algorithm needs
            # to be parsed out to pass to the key manager (ex: aes).
            cipher = CONF.ephemeral_storage_encryption.cipher
            algorithm = cipher.split('-')[0] if cipher else None
            instance.ephemeral_key_uuid = self.key_manager.create_key(
                context,
                algorithm=algorithm,
                length=CONF.ephemeral_storage_encryption.key_size)
        else:
            instance.ephemeral_key_uuid = None

        # Store image properties so we can use them later
        # (for notifications, etc).  Only store what we can.
        if not instance.obj_attr_is_set('system_metadata'):
            instance.system_metadata = {}
        # Make sure we have the dict form that we need for instance_update.
        instance.system_metadata = utils.instance_sys_meta(instance)

        system_meta = utils.get_system_metadata_from_image(
            image, flavor)

        # In case we couldn't find any suitable base_image
        system_meta.setdefault('image_base_image_ref', instance.image_ref)

        system_meta['owner_user_name'] = context.user_name
        system_meta['owner_project_name'] = context.project_name

        instance.system_metadata.update(system_meta)

        # Since the removal of nova-network, we don't actually store anything
        # in the database. Instead, we proxy the security groups on the
        # instance from the ports attached to the instance.
        instance.security_groups = objects.SecurityGroupList()

        self._populate_instance_names(instance, num_instances, index)
        instance.shutdown_terminate = shutdown_terminate

        return instance

    def _create_tag_list_obj(self, context, tags):
        """Create TagList objects from simple string tags.

        :param context: security context.
        :param tags: simple string tags from API request.
        :returns: TagList object.
        """
        tag_list = [objects.Tag(context=context, tag=t) for t in tags]
        tag_list_obj = objects.TagList(objects=tag_list)
        return tag_list_obj

    def _transform_tags(self, tags, resource_id):
        """Change the resource_id of the tags according to the input param.

        Because this method can be called multiple times when more than one
        instance is booted in a single request it makes a copy of the tags
        list.

        :param tags: TagList object.
        :param resource_id: string.
        :returns: TagList object.
        """
        instance_tags = tags.obj_clone()
        for tag in instance_tags:
            tag.resource_id = resource_id
        return instance_tags

    def _check_multiple_instances_with_neutron_ports(self, requested_networks):
        """Check whether multiple instances are created from port id(s)."""
        for requested_net in requested_networks:
            if requested_net.port_id:
                msg = _("Unable to launch multiple instances with"
                        " a single configured port ID. Please launch your"
                        " instance one by one with different ports.")
                raise exception.MultiplePortsNotApplicable(reason=msg)

    def _check_multiple_instances_with_specified_ip(self, requested_networks):
        """Check whether multiple instances are created with specified ip."""

        for requested_net in requested_networks:
            if requested_net.network_id and requested_net.address:
                msg = _("max_count cannot be greater than 1 if an fixed_ip "
                        "is specified.")
                raise exception.InvalidFixedIpAndMaxCountRequest(reason=msg)

    def create(
        self, context, flavor,
        image_href, kernel_id=None, ramdisk_id=None,
        min_count=None, max_count=None,
        display_name=None, display_description=None, hostname=None,
        key_name=None, key_data=None, security_groups=None,
        availability_zone=None, forced_host=None, forced_node=None,
        user_data=None, metadata=None, injected_files=None,
        admin_password=None, block_device_mapping=None,
        access_ip_v4=None, access_ip_v6=None, requested_networks=None,
        config_drive=None, auto_disk_config=None, scheduler_hints=None,
        legacy_bdm=True, shutdown_terminate=False,
        check_server_group_quota=False, tags=None,
        supports_multiattach=False, trusted_certs=None,
        supports_port_resource_request=False,
        requested_host=None, requested_hypervisor_hostname=None,
    ):
        """Provision instances, sending instance information to the
        scheduler.  The scheduler will determine where the instance(s)
        go and will handle creating the DB entries.

        Returns a tuple of (instances, reservation_id)
        """
        if requested_networks and max_count is not None and max_count > 1:
            self._check_multiple_instances_with_specified_ip(
                requested_networks)
            self._check_multiple_instances_with_neutron_ports(
                requested_networks)

        if hostname and max_count is not None and max_count > 1:
            raise exception.AmbiguousHostnameForMultipleInstances()

        if availability_zone and forced_host is None:
            azs = availability_zones.get_availability_zones(
                context.elevated(), self.host_api, get_only_available=True)
            if availability_zone not in azs:
                msg = _('The requested availability zone is not available')
                raise exception.InvalidRequest(msg)

        filter_properties = scheduler_utils.build_filter_properties(
            scheduler_hints, forced_host, forced_node, flavor)

        return self._create_instance(
            context, flavor,
            image_href, kernel_id, ramdisk_id,
            min_count, max_count,
            display_name, display_description, hostname,
            key_name, key_data, security_groups,
            availability_zone, user_data, metadata,
            injected_files, admin_password,
            access_ip_v4, access_ip_v6,
            requested_networks, config_drive,
            block_device_mapping, auto_disk_config,
            filter_properties=filter_properties,
            legacy_bdm=legacy_bdm,
            shutdown_terminate=shutdown_terminate,
            check_server_group_quota=check_server_group_quota,
            tags=tags, supports_multiattach=supports_multiattach,
            trusted_certs=trusted_certs,
            supports_port_resource_request=supports_port_resource_request,
            requested_host=requested_host,
            requested_hypervisor_hostname=requested_hypervisor_hostname)

    def _check_auto_disk_config(self, instance=None, image=None,
                                auto_disk_config=None):
        if auto_disk_config is None:
            return
        if not image and not instance:
            return

        if image:
            image_props = image.get("properties", {})
            auto_disk_config_img = \
                utils.get_auto_disk_config_from_image_props(image_props)
            image_ref = image.get("id")
        else:
            sys_meta = utils.instance_sys_meta(instance)
            image_ref = sys_meta.get('image_base_image_ref')
            auto_disk_config_img = \
                utils.get_auto_disk_config_from_instance(sys_meta=sys_meta)

        self._ensure_auto_disk_config_is_valid(auto_disk_config_img,
                                               auto_disk_config,
                                               image_ref)

    def _lookup_instance(self, context, uuid):
        '''Helper method for pulling an instance object from a database.

        During the transition to cellsv2 there is some complexity around
        retrieving an instance from the database which this method hides. If
        there is an instance mapping then query the cell for the instance, if
        no mapping exists then query the configured nova database.

        Once we are past the point that all deployments can be assumed to be
        migrated to cellsv2 this method can go away.
        '''
        inst_map = None
        try:
            inst_map = objects.InstanceMapping.get_by_instance_uuid(
                context, uuid)
        except exception.InstanceMappingNotFound:
            # TODO(alaski): This exception block can be removed once we're
            # guaranteed everyone is using cellsv2.
            pass

        if inst_map is None or inst_map.cell_mapping is None:
            # If inst_map is None then the deployment has not migrated to
            # cellsv2 yet.
            # If inst_map.cell_mapping is None then the instance is not in a
            # cell yet. Until instance creation moves to the conductor the
            # instance can be found in the configured database, so attempt
            # to look it up.
            cell = None
            try:
                instance = objects.Instance.get_by_uuid(context, uuid)
            except exception.InstanceNotFound:
                # If we get here then the conductor is in charge of writing the
                # instance to the database and hasn't done that yet. It's up to
                # the caller of this method to determine what to do with that
                # information.
                return None, None
        else:
            cell = inst_map.cell_mapping
            with nova_context.target_cell(context, cell) as cctxt:
                try:
                    instance = objects.Instance.get_by_uuid(cctxt, uuid)
                except exception.InstanceNotFound:
                    # Since the cell_mapping exists we know the instance is in
                    # the cell, however InstanceNotFound means it's already
                    # deleted.
                    return None, None
        return cell, instance

    def _delete_while_booting(self, context, instance):
        """Handle deletion if the instance has not reached a cell yet

        Deletion before an instance reaches a cell needs to be handled
        differently. What we're attempting to do is delete the BuildRequest
        before the api level conductor does.  If we succeed here then the boot
        request stops before reaching a cell.  If not then the instance will
        need to be looked up in a cell db and the normal delete path taken.
        """
        deleted = self._attempt_delete_of_buildrequest(context, instance)
        if deleted:
            # If we've reached this block the successful deletion of the
            # buildrequest indicates that the build process should be halted by
            # the conductor.

            # NOTE(alaski): Though the conductor halts the build process it
            # does not currently delete the instance record. This is
            # because in the near future the instance record will not be
            # created if the buildrequest has been deleted here. For now we
            # ensure the instance has been set to deleted at this point.
            # Yes this directly contradicts the comment earlier in this
            # method, but this is a temporary measure.
            # Look up the instance because the current instance object was
            # stashed on the buildrequest and therefore not complete enough
            # to run .destroy().
            try:
                instance_uuid = instance.uuid
                cell, instance = self._lookup_instance(context, instance_uuid)
                if instance is not None:
                    # If instance is None it has already been deleted.
                    if cell:
                        with nova_context.target_cell(context, cell) as cctxt:
                            # FIXME: When the instance context is targeted,
                            # we can remove this
                            with compute_utils.notify_about_instance_delete(
                                    self.notifier, cctxt, instance):
                                instance.destroy()
                    else:
                        instance.destroy()
            except exception.InstanceNotFound:
                pass

            return True
        return False

    def _local_delete_cleanup(self, context, instance_uuid):
        # NOTE(aarents) Ensure instance allocation is cleared and instance
        # mapping queued as deleted before _delete() return
        try:
            self.placementclient.delete_allocation_for_instance(
                context, instance_uuid, force=True)
        except exception.AllocationDeleteFailed:
            LOG.info("Allocation delete failed during local delete cleanup.",
                     instance_uuid=instance_uuid)

        try:
            self._update_queued_for_deletion(context, instance_uuid, True)
        except exception.InstanceMappingNotFound:
            LOG.info("Instance Mapping does not exist while attempting "
                     "local delete cleanup.",
                     instance_uuid=instance_uuid)

    def _attempt_delete_of_buildrequest(self, context, instance):
        # If there is a BuildRequest then the instance may not have been
        # written to a cell db yet. Delete the BuildRequest here, which
        # will indicate that the Instance build should not proceed.
        try:
            build_req = objects.BuildRequest.get_by_instance_uuid(
                context, instance.uuid)
            build_req.destroy()
        except exception.BuildRequestNotFound:
            # This means that conductor has deleted the BuildRequest so the
            # instance is now in a cell and the delete needs to proceed
            # normally.
            return False

        # We need to detach from any volumes so they aren't orphaned.
        self._local_cleanup_bdm_volumes(
            build_req.block_device_mappings, instance, context)

        return True

    def _delete(self, context, instance, delete_type, cb, **instance_attrs):
        if instance.disable_terminate:
            LOG.info('instance termination disabled', instance=instance)
            return

        cell = None
        # If there is an instance.host (or the instance is shelved-offloaded or
        # in error state), the instance has been scheduled and sent to a
        # cell/compute which means it was pulled from the cell db.
        # Normal delete should be attempted.
        may_have_ports_or_volumes = compute_utils.may_have_ports_or_volumes(
            instance)

        # Save a copy of the instance UUID early, in case
        # _lookup_instance returns instance = None, to pass to
        # _local_delete_cleanup if needed.
        instance_uuid = instance.uuid

        if not instance.host and not may_have_ports_or_volumes:
            try:
                if self._delete_while_booting(context, instance):
                    self._local_delete_cleanup(context, instance.uuid)
                    return
                # If instance.host was not set it's possible that the Instance
                # object here was pulled from a BuildRequest object and is not
                # fully populated. Notably it will be missing an 'id' field
                # which will prevent instance.destroy from functioning
                # properly. A lookup is attempted which will either return a
                # full Instance or None if not found. If not found then it's
                # acceptable to skip the rest of the delete processing.

                cell, instance = self._lookup_instance(context, instance.uuid)
                if cell and instance:
                    try:
                        # Now destroy the instance from the cell it lives in.
                        with compute_utils.notify_about_instance_delete(
                                self.notifier, context, instance):
                            instance.destroy()
                    except exception.InstanceNotFound:
                        pass
                    # The instance was deleted or is already gone.
                    self._local_delete_cleanup(context, instance.uuid)
                    return
                if not instance:
                    # Instance is already deleted.
                    self._local_delete_cleanup(context, instance_uuid)
                    return
            except exception.ObjectActionError:
                # NOTE(melwitt): This means the instance.host changed
                # under us indicating the instance became scheduled
                # during the destroy(). Refresh the instance from the DB and
                # continue on with the delete logic for a scheduled instance.
                # NOTE(danms): If instance.host is set, we should be able to
                # do the following lookup. If not, there's not much we can
                # do to recover.
                cell, instance = self._lookup_instance(context, instance.uuid)
                if not instance:
                    # Instance is already deleted
                    self._local_delete_cleanup(context, instance_uuid)
                    return

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

        # At these states an instance has a snapshot associate.
        if instance.vm_state in (vm_states.SHELVED,
                                 vm_states.SHELVED_OFFLOADED):
            snapshot_id = instance.system_metadata.get('shelved_image_id')
            LOG.info("Working on deleting snapshot %s "
                     "from shelved instance...",
                     snapshot_id, instance=instance)
            try:
                self.image_api.delete(context, snapshot_id)
            except (exception.ImageNotFound,
                    exception.ImageNotAuthorized) as exc:
                LOG.warning("Failed to delete snapshot "
                            "from shelved instance (%s).",
                            exc.format_message(), instance=instance)
            except Exception:
                LOG.exception("Something wrong happened when trying to "
                              "delete snapshot from shelved instance.",
                              instance=instance)

        original_task_state = instance.task_state
        try:
            # NOTE(maoy): no expected_task_state needs to be set
            instance.update(instance_attrs)
            instance.progress = 0
            instance.save()

            if not instance.host and not may_have_ports_or_volumes:
                try:
                    with compute_utils.notify_about_instance_delete(
                            self.notifier, context, instance,
                            delete_type
                            if delete_type != 'soft_delete'
                            else 'delete'):
                        instance.destroy()
                    LOG.info('Instance deleted and does not have host '
                             'field, its vm_state is %(state)s.',
                             {'state': instance.vm_state},
                              instance=instance)
                    self._local_delete_cleanup(context, instance.uuid)
                    return
                except exception.ObjectActionError as ex:
                    # The instance's host likely changed under us as
                    # this instance could be building and has since been
                    # scheduled. Continue with attempts to delete it.
                    LOG.debug('Refreshing instance because: %s', ex,
                              instance=instance)
                    instance.refresh()

            if instance.vm_state == vm_states.RESIZED:
                self._confirm_resize_on_deleting(context, instance)
                # NOTE(neha_alhat): After confirm resize vm_state will become
                # 'active' and task_state will be set to 'None'. But for soft
                # deleting a vm, the _do_soft_delete callback requires
                # task_state in 'SOFT_DELETING' status. So, we need to set
                # task_state as 'SOFT_DELETING' again for soft_delete case.
                # After confirm resize and before saving the task_state to
                # "SOFT_DELETING", during the short window, user can submit
                # soft delete vm request again and system will accept and
                # process it without any errors.
                if delete_type == 'soft_delete':
                    instance.task_state = instance_attrs['task_state']
                    instance.save()

            is_local_delete = True
            try:
                # instance.host must be set in order to look up the service.
                if instance.host is not None:
                    service = objects.Service.get_by_compute_host(
                        context.elevated(), instance.host)
                    is_local_delete = not self.servicegroup_api.service_is_up(
                        service)
                if not is_local_delete:
                    if original_task_state in (task_states.DELETING,
                                                  task_states.SOFT_DELETING):
                        LOG.info('Instance is already in deleting state, '
                                 'ignoring this request',
                                 instance=instance)
                        return
                    self._record_action_start(context, instance,
                                              instance_actions.DELETE)

                    cb(context, instance, bdms)
            except exception.ComputeHostNotFound:
                LOG.debug('Compute host %s not found during service up check, '
                          'going to local delete instance', instance.host,
                          instance=instance)

            if is_local_delete:
                # If instance is in shelved_offloaded state or compute node
                # isn't up, delete instance from db and clean bdms info and
                # network info
                if cell is None:
                    # NOTE(danms): If we didn't get our cell from one of the
                    # paths above, look it up now.
                    try:
                        im = objects.InstanceMapping.get_by_instance_uuid(
                            context, instance.uuid)
                        cell = im.cell_mapping
                    except exception.InstanceMappingNotFound:
                        LOG.warning('During local delete, failed to find '
                                    'instance mapping', instance=instance)
                        return

                LOG.debug('Doing local delete in cell %s', cell.identity,
                          instance=instance)
                with nova_context.target_cell(context, cell) as cctxt:
                    self._local_delete(cctxt, instance, bdms, delete_type, cb)

        except exception.InstanceNotFound:
            # NOTE(comstud): Race condition. Instance already gone.
            pass

    def _confirm_resize_on_deleting(self, context, instance):
        # If in the middle of a resize, use confirm_resize to
        # ensure the original instance is cleaned up too along
        # with its allocations (and migration-based allocations)
        # in placement.
        migration = None
        for status in ('finished', 'confirming'):
            try:
                migration = objects.Migration.get_by_instance_and_status(
                        context.elevated(), instance.uuid, status)
                LOG.info('Found an unconfirmed migration during delete, '
                         'id: %(id)s, status: %(status)s',
                         {'id': migration.id,
                          'status': migration.status},
                         instance=instance)
                break
            except exception.MigrationNotFoundByStatus:
                pass

        if not migration:
            LOG.info('Instance may have been confirmed during delete',
                     instance=instance)
            return

        self._record_action_start(context, instance,
                                  instance_actions.CONFIRM_RESIZE)

        # If migration.cross_cell_move, we need to also cleanup the instance
        # data from the source cell database.
        if migration.cross_cell_move:
            self.compute_task_api.confirm_snapshot_based_resize(
                context, instance, migration, do_cast=False)
        else:
            self.compute_rpcapi.confirm_resize(context,
                    instance, migration, migration.source_compute, cast=False)

    def _local_cleanup_bdm_volumes(self, bdms, instance, context):
        """The method deletes the bdm records and, if a bdm is a volume, call
        the terminate connection and the detach volume via the Volume API.
        """
        elevated = context.elevated()
        for bdm in bdms:
            if bdm.is_volume:
                try:
                    if bdm.attachment_id:
                        self.volume_api.attachment_delete(context,
                                                          bdm.attachment_id)
                    else:
                        connector = compute_utils.get_stashed_volume_connector(
                            bdm, instance)
                        if connector:
                            self.volume_api.terminate_connection(context,
                                                                 bdm.volume_id,
                                                                 connector)
                        else:
                            LOG.debug('Unable to find connector for volume %s,'
                                      ' not attempting terminate_connection.',
                                      bdm.volume_id, instance=instance)
                        # Attempt to detach the volume. If there was no
                        # connection made in the first place this is just
                        # cleaning up the volume state in the Cinder DB.
                        self.volume_api.detach(elevated, bdm.volume_id,
                                               instance.uuid)

                    if bdm.delete_on_termination:
                        self.volume_api.delete(context, bdm.volume_id)
                except Exception as exc:
                    LOG.warning("Ignoring volume cleanup failure due to %s",
                                exc, instance=instance)
            # If we're cleaning up volumes from an instance that wasn't yet
            # created in a cell, i.e. the user deleted the server while
            # the BuildRequest still existed, then the BDM doesn't actually
            # exist in the DB to destroy it.
            if 'id' in bdm:
                bdm.destroy()

    @property
    def placementclient(self):
        return report.report_client_singleton()

    def _local_delete(self, context, instance, bdms, delete_type, cb):
        if instance.vm_state == vm_states.SHELVED_OFFLOADED:
            LOG.info("instance is in SHELVED_OFFLOADED state, cleanup"
                     " the instance's info from database.",
                     instance=instance)
        else:
            LOG.warning("instance's host %s is down, deleting from "
                        "database", instance.host, instance=instance)
        with compute_utils.notify_about_instance_delete(
                self.notifier, context, instance,
                delete_type if delete_type != 'soft_delete' else 'delete'):

            elevated = context.elevated()
            self.network_api.deallocate_for_instance(elevated, instance)

            # cleanup volumes
            self._local_cleanup_bdm_volumes(bdms, instance, context)

            # cleanup accelerator requests (ARQs)
            compute_utils.delete_arqs_if_needed(context, instance)

            # Cleanup allocations in Placement since we can't do it from the
            # compute service.
            self.placementclient.delete_allocation_for_instance(
                context, instance.uuid, force=True)
            cb(context, instance, bdms, local=True)
            instance.destroy()

    @staticmethod
    def _update_queued_for_deletion(context, instance_uuid, qfd):
        # NOTE(tssurya): We query the instance_mapping record of this instance
        # and update the queued_for_delete flag to True (or False according to
        # the state of the instance). This just means that the instance is
        # queued for deletion (or is no longer queued for deletion). It does
        # not guarantee its successful deletion (or restoration). Hence the
        # value could be stale which is fine, considering its use is only
        # during down cell (desperate) situation.
        im = objects.InstanceMapping.get_by_instance_uuid(context,
                                                          instance_uuid)
        im.queued_for_delete = qfd
        im.save()

    def _do_delete(self, context, instance, bdms, local=False):
        if local:
            instance.vm_state = vm_states.DELETED
            instance.task_state = None
            instance.terminated_at = timeutils.utcnow()
            instance.save()
        else:
            self.compute_rpcapi.terminate_instance(context, instance, bdms)
        self._update_queued_for_deletion(context, instance.uuid, True)

    def _do_soft_delete(self, context, instance, bdms, local=False):
        if local:
            instance.vm_state = vm_states.SOFT_DELETED
            instance.task_state = None
            instance.terminated_at = timeutils.utcnow()
            instance.save()
        else:
            self.compute_rpcapi.soft_delete_instance(context, instance)
        self._update_queued_for_deletion(context, instance.uuid, True)

    # NOTE(maoy): we allow delete to be called no matter what vm_state says.
    @check_instance_lock
    @check_instance_state(vm_state=None, task_state=None,
                          must_have_launched=True)
    def soft_delete(self, context, instance):
        """Terminate an instance."""
        LOG.debug('Going to try to soft delete instance',
                  instance=instance)

        self._delete(context, instance, 'soft_delete', self._do_soft_delete,
                     task_state=task_states.SOFT_DELETING,
                     deleted_at=timeutils.utcnow())

    def _delete_instance(self, context, instance):
        self._delete(context, instance, 'delete', self._do_delete,
                     task_state=task_states.DELETING)

    @check_instance_lock
    @check_instance_state(vm_state=None, task_state=None,
                          must_have_launched=False)
    def delete(self, context, instance):
        """Terminate an instance."""
        LOG.debug("Going to try to terminate instance", instance=instance)
        self._delete_instance(context, instance)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.SOFT_DELETED])
    def restore(self, context, instance):
        """Restore a previously deleted (but not reclaimed) instance."""
        # Check quotas
        flavor = instance.get_flavor()
        project_id, user_id = quotas_obj.ids_from_instance(context, instance)
        compute_utils.check_num_instances_quota(context, flavor, 1, 1,
                project_id=project_id, user_id=user_id)
        is_bfv = compute_utils.is_volume_backed_instance(context, instance)
        placement_limits.enforce_num_instances_and_flavor(context, project_id,
                                                         flavor, is_bfv, 1, 1)

        self._record_action_start(context, instance, instance_actions.RESTORE)

        if instance.host:
            instance.task_state = task_states.RESTORING
            instance.deleted_at = None
            instance.save(expected_task_state=[None])
            # TODO(melwitt): We're not rechecking for strict quota here to
            # guard against going over quota during a race at this time because
            # the resource consumption for this operation is written to the
            # database by compute.
            self.compute_rpcapi.restore_instance(context, instance)
        else:
            instance.vm_state = vm_states.ACTIVE
            instance.task_state = None
            instance.deleted_at = None
            instance.save(expected_task_state=[None])
        self._update_queued_for_deletion(context, instance.uuid, False)

    @check_instance_lock
    @check_instance_state(task_state=None,
                          must_have_launched=False)
    def force_delete(self, context, instance):
        """Force delete an instance in any vm_state/task_state."""
        self._delete(context, instance, 'force_delete', self._do_delete,
                     task_state=task_states.DELETING)

    def force_stop(self, context, instance, do_cast=True, clean_shutdown=True):
        LOG.debug("Going to try to stop instance", instance=instance)

        instance.task_state = task_states.POWERING_OFF
        instance.progress = 0
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance, instance_actions.STOP)

        self.compute_rpcapi.stop_instance(context, instance, do_cast=do_cast,
                                          clean_shutdown=clean_shutdown)

    @check_instance_lock
    @check_instance_host()
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.ERROR])
    def stop(self, context, instance, do_cast=True, clean_shutdown=True):
        """Stop an instance."""
        self.force_stop(context, instance, do_cast, clean_shutdown)

    @check_instance_lock
    @check_instance_host()
    @check_instance_state(vm_state=[vm_states.STOPPED])
    def start(self, context, instance):
        """Start an instance."""
        LOG.debug("Going to try to start instance", instance=instance)

        instance.task_state = task_states.POWERING_ON
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance, instance_actions.START)
        self.compute_rpcapi.start_instance(context, instance)

    @check_instance_lock
    @check_instance_host()
    @check_instance_state(vm_state=vm_states.ALLOW_TRIGGER_CRASH_DUMP)
    def trigger_crash_dump(self, context, instance):
        """Trigger crash dump in an instance."""
        LOG.debug("Try to trigger crash dump", instance=instance)

        self._record_action_start(context, instance,
                                  instance_actions.TRIGGER_CRASH_DUMP)

        self.compute_rpcapi.trigger_crash_dump(context, instance)

    def _generate_minimal_construct_for_down_cells(self, context,
                                                   down_cell_uuids,
                                                   project, limit):
        """Generate a list of minimal instance constructs for a given list of
        cells that did not respond to a list operation. This will list
        every instance mapping in the affected cells and return a minimal
        objects.Instance for each (non-queued-for-delete) mapping.

        :param context: RequestContext
        :param down_cell_uuids: A list of cell UUIDs that did not respond
        :param project: A project ID to filter mappings, or None
        :param limit: A numeric limit on the number of results, or None
        :returns: An InstanceList() of partial Instance() objects
        """
        unavailable_servers = objects.InstanceList()
        for cell_uuid in down_cell_uuids:
            LOG.warning("Cell %s is not responding and hence only "
                        "partial results are available from this "
                        "cell if any.", cell_uuid)
            instance_mappings = (objects.InstanceMappingList.
                get_not_deleted_by_cell_and_project(context, cell_uuid,
                                                    project, limit=limit))
            for im in instance_mappings:
                unavailable_servers.objects.append(
                    objects.Instance(
                        context=context,
                        uuid=im.instance_uuid,
                        project_id=im.project_id,
                        created_at=im.created_at
                    )
                )
            if limit is not None:
                limit -= len(instance_mappings)
                if limit <= 0:
                    break
        return unavailable_servers

    def _get_instance_map_or_none(self, context, instance_uuid):
        try:
            inst_map = objects.InstanceMapping.get_by_instance_uuid(
                    context, instance_uuid)
        except exception.InstanceMappingNotFound:
            # InstanceMapping should always be found generally. This exception
            # may be raised if a deployment has partially migrated the nova-api
            # services.
            inst_map = None
        return inst_map

    @staticmethod
    def _save_user_id_in_instance_mapping(mapping, instance):
        # TODO(melwitt): We take the opportunity to migrate user_id on the
        # instance mapping if it's not yet been migrated. This can be removed
        # in a future release, when all migrations are complete.
        # If the instance came from a RequestSpec because of a down cell, its
        # user_id could be None and the InstanceMapping.user_id field is
        # non-nullable. Avoid trying to set/save the user_id in that case.
        if 'user_id' not in mapping and instance.user_id is not None:
            mapping.user_id = instance.user_id
            mapping.save()

    def _get_instance_from_cell(self, context, im, expected_attrs,
                                cell_down_support):
        # NOTE(danms): Even though we're going to scatter/gather to the
        # right cell, other code depends on this being force targeted when
        # the get call returns.
        nova_context.set_target_cell(context, im.cell_mapping)

        uuid = im.instance_uuid
        result = nova_context.scatter_gather_single_cell(context,
            im.cell_mapping, objects.Instance.get_by_uuid, uuid,
            expected_attrs=expected_attrs)
        cell_uuid = im.cell_mapping.uuid
        if not nova_context.is_cell_failure_sentinel(result[cell_uuid]):
            inst = result[cell_uuid]
            self._save_user_id_in_instance_mapping(im, inst)
            return inst
        elif isinstance(result[cell_uuid], exception.InstanceNotFound):
            raise exception.InstanceNotFound(instance_id=uuid)
        elif cell_down_support:
            if im.queued_for_delete:
                # should be treated like deleted instance.
                raise exception.InstanceNotFound(instance_id=uuid)

            # instance in down cell, return a minimal construct
            LOG.warning("Cell %s is not responding and hence only "
                        "partial results are available from this "
                        "cell.", cell_uuid)
            try:
                rs = objects.RequestSpec.get_by_instance_uuid(context,
                                                              uuid)
                # For BFV case, we could have rs.image but rs.image.id might
                # still not be set. So we check the existence of both image
                # and its id.
                image_ref = (rs.image.id if rs.image and
                             'id' in rs.image else None)
                inst = objects.Instance(context=context, power_state=0,
                                        uuid=uuid,
                                        project_id=im.project_id,
                                        created_at=im.created_at,
                                        user_id=rs.user_id,
                                        flavor=rs.flavor,
                                        image_ref=image_ref,
                                        availability_zone=rs.availability_zone)
                self._save_user_id_in_instance_mapping(im, inst)
                return inst
            except exception.RequestSpecNotFound:
                # could be that a deleted instance whose request
                # spec has been archived is being queried.
                raise exception.InstanceNotFound(instance_id=uuid)
        else:
            if isinstance(result[cell_uuid], exception.NovaException):
                LOG.exception(result[cell_uuid])
            raise exception.NovaException(
                _("Cell %s is not responding or returned an exception, "
                  "hence instance info is not available.") % cell_uuid)

    def _get_instance(self, context, instance_uuid, expected_attrs,
                      cell_down_support=False):
        inst_map = self._get_instance_map_or_none(context, instance_uuid)
        if inst_map and (inst_map.cell_mapping is not None):
            instance = self._get_instance_from_cell(context, inst_map,
                expected_attrs, cell_down_support)
        elif inst_map and (inst_map.cell_mapping is None):
            # This means the instance has not been scheduled and put in
            # a cell yet. For now it also may mean that the deployer
            # has not created their cell(s) yet.
            try:
                build_req = objects.BuildRequest.get_by_instance_uuid(
                        context, instance_uuid)
                instance = build_req.instance
            except exception.BuildRequestNotFound:
                # Instance was mapped and the BuildRequest was deleted
                # while fetching. Try again.
                inst_map = self._get_instance_map_or_none(context,
                                                          instance_uuid)
                if inst_map and (inst_map.cell_mapping is not None):
                    instance = self._get_instance_from_cell(context, inst_map,
                        expected_attrs, cell_down_support)
                else:
                    raise exception.InstanceNotFound(instance_id=instance_uuid)
        else:
            # If we got here, we don't have an instance mapping, but we aren't
            # sure why. The instance mapping might be missing because the
            # upgrade is incomplete (map_instances wasn't run). Or because the
            # instance was deleted and the DB was archived at which point the
            # mapping is deleted. The former case is bad, but because of the
            # latter case we can't really log any kind of warning/error here
            # since it might be normal.
            raise exception.InstanceNotFound(instance_id=instance_uuid)

        return instance

    def get(self, context, instance_id, expected_attrs=None,
            cell_down_support=False):
        """Get a single instance with the given instance_id.

        :param cell_down_support: True if the API (and caller) support
                                  returning a minimal instance
                                  construct if the relevant cell is
                                  down. If False, an error is raised
                                  since the instance cannot be retrieved
                                  due to the cell being down.
        """
        if not expected_attrs:
            expected_attrs = []
        expected_attrs.extend(['metadata', 'system_metadata',
                               'security_groups', 'info_cache'])
        # NOTE(ameade): we still need to support integer ids for ec2
        try:
            if uuidutils.is_uuid_like(instance_id):
                LOG.debug("Fetching instance by UUID",
                           instance_uuid=instance_id)

                instance = self._get_instance(context, instance_id,
                    expected_attrs, cell_down_support=cell_down_support)
            else:
                LOG.debug("Failed to fetch instance by id %s", instance_id)
                raise exception.InstanceNotFound(instance_id=instance_id)
        except exception.InvalidID:
            LOG.debug("Invalid instance id %s", instance_id)
            raise exception.InstanceNotFound(instance_id=instance_id)

        return instance

    def get_all(self, context, search_opts=None, limit=None, marker=None,
                expected_attrs=None, sort_keys=None, sort_dirs=None,
                cell_down_support=False, all_tenants=False):
        """Get all instances filtered by one of the given parameters.

        If there is no filter and the context is an admin, it will retrieve
        all instances in the system.

        Deleted instances will be returned by default, unless there is a
        search option that says otherwise.

        The results will be sorted based on the list of sort keys in the
        'sort_keys' parameter (first value is primary sort key, second value is
        secondary sort ket, etc.). For each sort key, the associated sort
        direction is based on the list of sort directions in the 'sort_dirs'
        parameter.

        :param cell_down_support: True if the API (and caller) support
                                  returning a minimal instance
                                  construct if the relevant cell is
                                  down. If False, instances from
                                  unreachable cells will be omitted.
        :param all_tenants: True if the "all_tenants" filter was passed.

        """
        if search_opts is None:
            search_opts = {}

        LOG.debug("Searching by: %s", str(search_opts))

        # Fixups for the DB call
        filters = {}

        def _remap_flavor_filter(flavor_id):
            flavor = objects.Flavor.get_by_flavor_id(context, flavor_id)
            filters['instance_type_id'] = flavor.id

        def _remap_fixed_ip_filter(fixed_ip):
            # Turn fixed_ip into a regexp match. Since '.' matches
            # any character, we need to use regexp escaping for it.
            filters['ip'] = '^%s$' % fixed_ip.replace('.', '\\.')

        # search_option to filter_name mapping.
        filter_mapping = {
                'image': 'image_ref',
                'name': 'display_name',
                'tenant_id': 'project_id',
                'flavor': _remap_flavor_filter,
                'fixed_ip': _remap_fixed_ip_filter}

        # copy from search_opts, doing various remappings as necessary
        for opt, value in search_opts.items():
            # Do remappings.
            # Values not in the filter_mapping table are copied as-is.
            # If remapping is None, option is not copied
            # If the remapping is a string, it is the filter_name to use
            try:
                remap_object = filter_mapping[opt]
            except KeyError:
                filters[opt] = value
            else:
                # Remaps are strings to translate to, or functions to call
                # to do the translating as defined by the table above.
                if isinstance(remap_object, str):
                    filters[remap_object] = value
                else:
                    try:
                        remap_object(value)

                    # We already know we can't match the filter, so
                    # return an empty list
                    except ValueError:
                        return objects.InstanceList()

        # IP address filtering cannot be applied at the DB layer, remove any DB
        # limit so that it can be applied after the IP filter.
        filter_ip = 'ip6' in filters or 'ip' in filters
        skip_build_request = False
        orig_limit = limit
        if filter_ip:
            # We cannot skip build requests if there is a marker since the
            # the marker could be a build request.
            skip_build_request = marker is None
            if self.network_api.has_substr_port_filtering_extension(context):
                # We're going to filter by IP using Neutron so set filter_ip
                # to False so we don't attempt post-DB query filtering in
                # memory below.
                filter_ip = False
                instance_uuids = self._ip_filter_using_neutron(context,
                                                               filters)
                if instance_uuids:
                    # Note that 'uuid' is not in the 2.1 GET /servers query
                    # parameter schema, however, we allow additionalProperties
                    # so someone could filter instances by uuid, which doesn't
                    # make a lot of sense but we have to account for it.
                    if 'uuid' in filters and filters['uuid']:
                        filter_uuids = filters['uuid']
                        if isinstance(filter_uuids, list):
                            instance_uuids.extend(filter_uuids)
                        else:
                            # Assume a string. If it's a dict or tuple or
                            # something, well...that's too bad. This is why
                            # we have query parameter schema definitions.
                            if filter_uuids not in instance_uuids:
                                instance_uuids.append(filter_uuids)
                    filters['uuid'] = instance_uuids
                else:
                    # No matches on the ip filter(s), return an empty list.
                    return objects.InstanceList()
            elif limit:
                LOG.debug('Removing limit for DB query due to IP filter')
                limit = None

        # Skip get BuildRequest if filtering by IP address, as building
        # instances will not have IP addresses.
        if skip_build_request:
            build_requests = objects.BuildRequestList()
        else:
            # The ordering of instances will be
            # [sorted instances with no host] + [sorted instances with host].
            # This means BuildRequest and cell0 instances first, then cell
            # instances
            try:
                build_requests = objects.BuildRequestList.get_by_filters(
                    context, filters, limit=limit, marker=marker,
                    sort_keys=sort_keys, sort_dirs=sort_dirs)
                # If we found the marker in we need to set it to None
                # so we don't expect to find it in the cells below.
                marker = None
            except exception.MarkerNotFound:
                # If we didn't find the marker in the build requests then keep
                # looking for it in the cells.
                build_requests = objects.BuildRequestList()

        build_req_instances = objects.InstanceList(
            objects=[build_req.instance for build_req in build_requests])
        # Only subtract from limit if it is not None
        limit = (limit - len(build_req_instances)) if limit else limit

        # We could arguably avoid joining on security_groups if we're using
        # neutron (which is the default) but if you're using neutron then the
        # security_group_instance_association table should be empty anyway
        # and the DB should optimize out that join, making it insignificant.
        fields = ['metadata', 'info_cache', 'security_groups']
        if expected_attrs:
            fields.extend(expected_attrs)

        insts, down_cell_uuids = instance_list.get_instance_objects_sorted(
            context, filters, limit, marker, fields, sort_keys, sort_dirs,
            cell_down_support=cell_down_support)

        def _get_unique_filter_method():
            seen_uuids = set()

            def _filter(instance):
                # During a cross-cell move operation we could have the instance
                # in more than one cell database so we not only have to filter
                # duplicates but we want to make sure we only return the
                # "current" one which should also be the one that the instance
                # mapping points to, but we don't want to do that expensive
                # lookup here. The DB API will filter out hidden instances by
                # default but there is a small window where two copies of an
                # instance could be hidden=False in separate cell DBs.
                # NOTE(mriedem): We could make this better in the case that we
                # have duplicate instances that are both hidden=False by
                # showing the one with the newer updated_at value, but that
                # could be tricky if the user is filtering on
                # changes-since/before or updated_at, or sorting on updated_at,
                # but technically that was already potentially broken with this
                # _filter method if we return an older BuildRequest.instance,
                # and given the window should be very small where we have
                # duplicates, it's probably not worth the complexity.
                if instance.uuid in seen_uuids:
                    return False
                seen_uuids.add(instance.uuid)
                return True

            return _filter

        filter_method = _get_unique_filter_method()
        # Only subtract from limit if it is not None
        limit = (limit - len(insts)) if limit else limit
        # TODO(alaski): Clean up the objects concatenation when List objects
        # support it natively.
        instances = objects.InstanceList(
            objects=list(filter(filter_method,
                           build_req_instances.objects +
                           insts.objects)))

        if filter_ip:
            instances = self._ip_filter(instances, filters, orig_limit)

        if cell_down_support:
            # API and client want minimal construct instances for any cells
            # that didn't return, so generate and prefix those to the actual
            # results.
            project = search_opts.get('project_id', context.project_id)
            if all_tenants:
                # NOTE(tssurya): The only scenario where project has to be None
                # is when using "all_tenants" in which case we do not want
                # the query to be restricted based on the project_id.
                project = None
            limit = (orig_limit - len(instances)) if limit else limit
            return (self._generate_minimal_construct_for_down_cells(context,
                down_cell_uuids, project, limit) + instances)

        return instances

    @staticmethod
    def _ip_filter(inst_models, filters, limit):
        ipv4_f = re.compile(str(filters.get('ip')))
        ipv6_f = re.compile(str(filters.get('ip6')))

        def _match_instance(instance):
            nw_info = instance.get_network_info()
            for vif in nw_info:
                for fixed_ip in vif.fixed_ips():
                    address = fixed_ip.get('address')
                    if not address:
                        continue
                    version = fixed_ip.get('version')
                    if ((version == 4 and ipv4_f.match(address)) or
                        (version == 6 and ipv6_f.match(address))):
                        return True
            return False

        result_objs = []
        for instance in inst_models:
            if _match_instance(instance):
                result_objs.append(instance)
                if limit and len(result_objs) == limit:
                    break
        return objects.InstanceList(objects=result_objs)

    def _ip_filter_using_neutron(self, context, filters):
        ip4_address = filters.get('ip')
        ip6_address = filters.get('ip6')
        addresses = [ip4_address, ip6_address]
        uuids = []
        for address in addresses:
            if address:
                try:
                    ports = self.network_api.list_ports(
                        context, fixed_ips='ip_address_substr=' + address,
                        fields=['device_id'])['ports']
                    for port in ports:
                        uuids.append(port['device_id'])
                except Exception as e:
                    LOG.error('An error occurred while listing ports '
                              'with an ip_address filter value of "%s". '
                              'Error: %s',
                              address, str(e))
        return uuids

    def update_instance(self, context, instance, updates):
        """Updates a single Instance object with some updates dict.

        Returns the updated instance.
        """

        # NOTE(sbauza): Given we only persist the Instance object after we
        # create the BuildRequest, we are sure that if the Instance object
        # has an ID field set, then it was persisted in the right Cell DB.
        if instance.obj_attr_is_set('id'):
            instance.update(updates)
            instance.save()
        else:
            # Instance is not yet mapped to a cell, so we need to update
            # BuildRequest instead
            # TODO(sbauza): Fix the possible race conditions where BuildRequest
            # could be deleted because of either a concurrent instance delete
            # or because the scheduler just returned a destination right
            # after we called the instance in the API.
            try:
                build_req = objects.BuildRequest.get_by_instance_uuid(
                    context, instance.uuid)
                instance = build_req.instance
                instance.update(updates)
                # FIXME(sbauza): Here we are updating the current
                # thread-related BuildRequest object. Given that another worker
                # could have looking up at that BuildRequest in the API, it
                # means that it could pass it down to the conductor without
                # making sure that it's not updated, we could have some race
                # condition where it would missing the updated fields, but
                # that's something we could discuss once the instance record
                # is persisted by the conductor.
                build_req.save()
            except exception.BuildRequestNotFound:
                # Instance was mapped and the BuildRequest was deleted
                # while fetching (and possibly the instance could have been
                # deleted as well). We need to lookup again the Instance object
                # in order to correctly update it.
                # TODO(sbauza): Figure out a good way to know the expected
                # attributes by checking which fields are set or not.
                expected_attrs = ['flavor', 'pci_devices', 'numa_topology',
                                  'tags', 'metadata', 'system_metadata',
                                  'security_groups', 'info_cache']
                inst_map = self._get_instance_map_or_none(context,
                                                          instance.uuid)
                if inst_map and (inst_map.cell_mapping is not None):
                    with nova_context.target_cell(
                            context,
                            inst_map.cell_mapping) as cctxt:
                        instance = objects.Instance.get_by_uuid(
                            cctxt, instance.uuid,
                            expected_attrs=expected_attrs)
                        instance.update(updates)
                        instance.save()
                else:
                    # Conductor doesn't delete the BuildRequest until after the
                    # InstanceMapping record is created, so if we didn't get
                    # that and the BuildRequest doesn't exist, then the
                    # instance is already gone and we need to just error out.
                    raise exception.InstanceNotFound(instance_id=instance.uuid)
        return instance

    # NOTE(melwitt): We don't check instance lock for backup because lock is
    #                intended to prevent accidental change/delete of instances
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.PAUSED, vm_states.SUSPENDED])
    def backup(self, context, instance, name, backup_type, rotation,
               extra_properties=None):
        """Backup the given instance

        :param instance: nova.objects.instance.Instance object
        :param name: name of the backup
        :param backup_type: 'daily' or 'weekly'
        :param rotation: int representing how many backups to keep around;
            None if rotation shouldn't be used (as in the case of snapshots)
        :param extra_properties: dict of extra image properties to include
                                 when creating the image.
        :returns: A dict containing image metadata
        """
        props_copy = dict(extra_properties, backup_type=backup_type)

        if compute_utils.is_volume_backed_instance(context, instance):
            LOG.info("It's not supported to backup volume backed "
                     "instance.", instance=instance)
            raise exception.InvalidRequest(
                _('Backup is not supported for volume-backed instances.'))
        else:
            image_meta = compute_utils.create_image(
                context, instance, name, 'backup', self.image_api,
                extra_properties=props_copy)

        instance.task_state = task_states.IMAGE_BACKUP
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance,
                                  instance_actions.BACKUP)

        self.compute_rpcapi.backup_instance(context, instance,
                                            image_meta['id'],
                                            backup_type,
                                            rotation)
        return image_meta

    # NOTE(melwitt): We don't check instance lock for snapshot because lock is
    #                intended to prevent accidental change/delete of instances
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.PAUSED, vm_states.SUSPENDED])
    def snapshot(self, context, instance, name, extra_properties=None):
        """Snapshot the given instance.

        :param instance: nova.objects.instance.Instance object
        :param name: name of the snapshot
        :param extra_properties: dict of extra image properties to include
                                 when creating the image.
        :returns: A dict containing image metadata
        """
        image_meta = compute_utils.create_image(
            context, instance, name, 'snapshot', self.image_api,
            extra_properties=extra_properties)

        instance.task_state = task_states.IMAGE_SNAPSHOT_PENDING
        try:
            instance.save(expected_task_state=[None])
        except (exception.InstanceNotFound,
                exception.UnexpectedDeletingTaskStateError) as ex:
            # Changing the instance task state to use in raising the
            # InstanceInvalidException below
            LOG.debug('Instance disappeared during snapshot.',
                      instance=instance)
            try:
                image_id = image_meta['id']
                self.image_api.delete(context, image_id)
                LOG.info('Image %s deleted because instance '
                         'deleted before snapshot started.',
                         image_id, instance=instance)
            except exception.ImageNotFound:
                pass
            except Exception as exc:
                LOG.warning("Error while trying to clean up image %(img_id)s: "
                            "%(error_msg)s",
                            {"img_id": image_meta['id'],
                             "error_msg": str(exc)})
            attr = 'task_state'
            state = task_states.DELETING
            if type(ex) == exception.InstanceNotFound:
                attr = 'vm_state'
                state = vm_states.DELETED
            raise exception.InstanceInvalidState(attr=attr,
                                           instance_uuid=instance.uuid,
                                           state=state,
                                           method='snapshot')

        self._record_action_start(context, instance,
                                  instance_actions.CREATE_IMAGE)

        self.compute_rpcapi.snapshot_instance(context, instance,
                                              image_meta['id'])

        return image_meta

    # NOTE(melwitt): We don't check instance lock for snapshot because lock is
    #                intended to prevent accidental change/delete of instances
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.PAUSED, vm_states.SUSPENDED])
    def snapshot_volume_backed(self, context, instance, name,
                               extra_properties=None):
        """Snapshot the given volume-backed instance.

        :param instance: nova.objects.instance.Instance object
        :param name: name of the backup or snapshot
        :param extra_properties: dict of extra image properties to include

        :returns: the new image metadata
        """
        image_meta = compute_utils.initialize_instance_snapshot_metadata(
            context, instance, name, extra_properties)
        # the new image is simply a bucket of properties (particularly the
        # block device mapping, kernel and ramdisk IDs) with no image data,
        # hence the zero size
        image_meta['size'] = 0
        for attr in ('container_format', 'disk_format'):
            image_meta.pop(attr, None)
        properties = image_meta['properties']
        # clean properties before filling
        for key in ('block_device_mapping', 'bdm_v2', 'root_device_name'):
            properties.pop(key, None)
        if instance.root_device_name:
            properties['root_device_name'] = instance.root_device_name

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

        mapping = []  # list of BDM dicts that can go into the image properties
        # Do some up-front filtering of the list of BDMs from
        # which we are going to create snapshots.
        volume_bdms = []
        for bdm in bdms:
            if bdm.no_device:
                continue
            if bdm.is_volume:
                # These will be handled below.
                volume_bdms.append(bdm)
            else:
                mapping.append(bdm.get_image_mapping())

        # Check limits in Cinder before creating snapshots to avoid going over
        # quota in the middle of a list of volumes. This is a best-effort check
        # but concurrently running snapshot requests from the same project
        # could still fail to create volume snapshots if they go over limit.
        if volume_bdms:
            limits = self.volume_api.get_absolute_limits(context)
            total_snapshots_used = limits['totalSnapshotsUsed']
            max_snapshots = limits['maxTotalSnapshots']
            # -1 means there is unlimited quota for snapshots
            if (max_snapshots > -1 and
                    len(volume_bdms) + total_snapshots_used > max_snapshots):
                LOG.debug('Unable to create volume snapshots for instance. '
                          'Currently has %s snapshots, requesting %s new '
                          'snapshots, with a limit of %s.',
                          total_snapshots_used, len(volume_bdms),
                          max_snapshots, instance=instance)
                raise exception.OverQuota(overs='snapshots')

        quiesced = False
        if instance.vm_state == vm_states.ACTIVE:
            try:
                LOG.info("Attempting to quiesce instance before volume "
                         "snapshot.", instance=instance)
                self.compute_rpcapi.quiesce_instance(context, instance)
                quiesced = True
            except (exception.InstanceQuiesceNotSupported,
                    exception.QemuGuestAgentNotEnabled,
                    exception.NovaException, NotImplementedError) as err:
                if strutils.bool_from_string(instance.system_metadata.get(
                        'image_os_require_quiesce')):
                    raise

                if isinstance(err, exception.NovaException):
                    LOG.info('Skipping quiescing instance: %(reason)s.',
                             {'reason': err.format_message()},
                             instance=instance)
                else:
                    LOG.info('Skipping quiescing instance because the '
                             'operation is not supported by the underlying '
                             'compute driver.', instance=instance)
            # NOTE(tasker): discovered that an uncaught exception could occur
            #               after the instance has been frozen. catch and thaw.
            except Exception as ex:
                with excutils.save_and_reraise_exception():
                    LOG.error("An error occurred during quiesce of instance. "
                              "Unquiescing to ensure instance is thawed. "
                              "Error: %s", str(ex),
                              instance=instance)
                    self.compute_rpcapi.unquiesce_instance(context, instance,
                                                           mapping=None)

        @wrap_instance_event(prefix='api')
        def snapshot_instance(self, context, instance, bdms):
            try:
                for bdm in volume_bdms:
                    # create snapshot based on volume_id
                    volume = self.volume_api.get(context, bdm.volume_id)
                    # NOTE(yamahata): Should we wait for snapshot creation?
                    #                 Linux LVM snapshot creation completes in
                    #                 short time, it doesn't matter for now.
                    name = _('snapshot for %s') % image_meta['name']
                    LOG.debug('Creating snapshot from volume %s.',
                              volume['id'], instance=instance)
                    snapshot = self.volume_api.create_snapshot_force(
                        context, volume['id'],
                        name, volume['display_description'])
                    mapping_dict = block_device.snapshot_from_bdm(
                        snapshot['id'], bdm)
                    mapping_dict = mapping_dict.get_image_mapping()
                    mapping.append(mapping_dict)
                return mapping
            # NOTE(tasker): No error handling is done in the above for loop.
            # This means that if the snapshot fails and throws an exception
            # the traceback will skip right over the unquiesce needed below.
            # Here, catch any exception, unquiesce the instance, and raise the
            # error so that the calling function can do what it needs to in
            # order to properly treat a failed snap.
            except Exception:
                with excutils.save_and_reraise_exception():
                    if quiesced:
                        LOG.info("Unquiescing instance after volume snapshot "
                                 "failure.", instance=instance)
                        self.compute_rpcapi.unquiesce_instance(
                            context, instance, mapping)

        self._record_action_start(context, instance,
                                  instance_actions.CREATE_IMAGE)
        mapping = snapshot_instance(self, context, instance, bdms)

        if quiesced:
            self.compute_rpcapi.unquiesce_instance(context, instance, mapping)

        if mapping:
            properties['block_device_mapping'] = mapping
            properties['bdm_v2'] = True

        return self.image_api.create(context, image_meta)

    @check_instance_lock
    def reboot(self, context, instance, reboot_type):
        """Reboot the given instance."""
        if reboot_type == 'SOFT':
            self._soft_reboot(context, instance)
        else:
            self._hard_reboot(context, instance)

    @check_instance_state(vm_state=set(vm_states.ALLOW_SOFT_REBOOT),
                          task_state=[None])
    def _soft_reboot(self, context, instance):
        expected_task_state = [None]
        instance.task_state = task_states.REBOOTING
        instance.save(expected_task_state=expected_task_state)

        self._record_action_start(context, instance, instance_actions.REBOOT)

        self.compute_rpcapi.reboot_instance(context, instance=instance,
                                            block_device_info=None,
                                            reboot_type='SOFT')

    @check_instance_state(vm_state=set(vm_states.ALLOW_HARD_REBOOT),
                          task_state=task_states.ALLOW_REBOOT)
    def _hard_reboot(self, context, instance):
        instance.task_state = task_states.REBOOTING_HARD
        instance.save(expected_task_state=task_states.ALLOW_REBOOT)

        self._record_action_start(context, instance, instance_actions.REBOOT)

        self.compute_rpcapi.reboot_instance(context, instance=instance,
                                            block_device_info=None,
                                            reboot_type='HARD')

    def _check_image_arch(self, image=None):
        if image:
            img_arch = image.get("properties", {}).get('hw_architecture')
            if img_arch:
                fields_obj.Architecture.canonicalize(img_arch)

    @reject_vtpm_instances(instance_actions.REBUILD)
    @block_accelerators(until_service=SUPPORT_ACCELERATOR_SERVICE_FOR_REBUILD)
    # TODO(stephenfin): We should expand kwargs out to named args
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.ERROR])
    def rebuild(self, context, instance, image_href, admin_password,
                files_to_inject=None, **kwargs):
        """Rebuild the given instance with the provided attributes."""
        files_to_inject = files_to_inject or []
        metadata = kwargs.get('metadata', {})
        preserve_ephemeral = kwargs.get('preserve_ephemeral', False)
        auto_disk_config = kwargs.get('auto_disk_config')

        if 'key_name' in kwargs:
            key_name = kwargs.pop('key_name')
            if key_name:
                # NOTE(liuyulong): we are intentionally using the user_id from
                # the request context rather than the instance.user_id because
                # users own keys but instances are owned by projects, and
                # another user in the same project can rebuild an instance
                # even if they didn't create it.
                key_pair = objects.KeyPair.get_by_name(context,
                                                       context.user_id,
                                                       key_name)
                instance.key_name = key_pair.name
                instance.key_data = key_pair.public_key
                instance.keypairs = objects.KeyPairList(objects=[key_pair])
            else:
                instance.key_name = None
                instance.key_data = None
                instance.keypairs = objects.KeyPairList(objects=[])

        # Use trusted_certs value from kwargs to create TrustedCerts object
        trusted_certs = None
        if 'trusted_certs' in kwargs:
            # Note that the user can set, change, or unset / reset trusted
            # certs. If they are explicitly specifying
            # trusted_image_certificates=None, that means we'll either unset
            # them on the instance *or* reset to use the defaults (if defaults
            # are configured).
            trusted_certs = kwargs.pop('trusted_certs')
            instance.trusted_certs = self._retrieve_trusted_certs_object(
                context, trusted_certs, rebuild=True)

        if 'hostname' in kwargs:
            instance.hostname = kwargs.pop('hostname')

        image_id, image = self._get_image(context, image_href)
        self._check_auto_disk_config(image=image,
                                     auto_disk_config=auto_disk_config)
        self._check_image_arch(image=image)

        flavor = instance.get_flavor()
        bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
            context, instance.uuid)
        root_bdm = compute_utils.get_root_bdm(context, instance, bdms)

        # Check to see if the image is changing and we have a volume-backed
        # server. The compute doesn't support changing the image in the
        # root disk of a volume-backed server, so we need to just fail fast.
        is_volume_backed = compute_utils.is_volume_backed_instance(
            context, instance, bdms)
        if is_volume_backed:
            if trusted_certs:
                # The only way we can get here is if the user tried to set
                # trusted certs or specified trusted_image_certificates=None
                # and default_trusted_certificate_ids is configured.
                msg = _("Image certificate validation is not supported "
                        "for volume-backed servers.")
                raise exception.CertificateValidationFailed(message=msg)

            # For boot from volume, instance.image_ref is empty, so we need to
            # query the image from the volume.
            if root_bdm is None:
                # This shouldn't happen and is an error, we need to fail. This
                # is not the users fault, it's an internal error. Without a
                # root BDM we have no way of knowing the backing volume (or
                # image in that volume) for this instance.
                raise exception.NovaException(
                    _('Unable to find root block device mapping for '
                      'volume-backed instance.'))

            volume = self.volume_api.get(context, root_bdm.volume_id)
            volume_image_metadata = volume.get('volume_image_metadata', {})
            orig_image_ref = volume_image_metadata.get('image_id')

            if orig_image_ref != image_href:
                # Leave a breadcrumb.
                LOG.debug('Requested to rebuild instance with a new image %s '
                          'for a volume-backed server with image %s in its '
                          'root volume which is not supported.', image_href,
                          orig_image_ref, instance=instance)
                msg = _('Unable to rebuild with a different image for a '
                        'volume-backed server.')
                raise exception.ImageUnacceptable(
                    image_id=image_href, reason=msg)
        else:
            orig_image_ref = instance.image_ref

        request_spec = objects.RequestSpec.get_by_instance_uuid(
            context, instance.uuid)

        self._checks_for_create_and_rebuild(context, image_id, image,
                flavor, metadata, files_to_inject, root_bdm)

        # Check the state of the volume. If it is not in-use, an exception
        # will occur when creating attachment during reconstruction,
        # resulting in the failure of reconstruction and the instance
        # turning into an error state.
        self._check_volume_status(context, bdms)

        # NOTE(sean-k-mooney): When we rebuild with a new image we need to
        # validate that the NUMA topology does not change as we do a NOOP claim
        # in resource tracker. As such we cannot allow the resource usage or
        # assignment to change as a result of a new image altering the
        # numa constraints.
        if orig_image_ref != image_href:
            self._validate_numa_rebuild(instance, image, flavor)

        kernel_id, ramdisk_id = self._handle_kernel_and_ramdisk(
                context, None, None, image)

        def _reset_image_metadata():
            """Remove old image properties that we're storing as instance
            system metadata.  These properties start with 'image_'.
            Then add the properties for the new image.
            """
            # FIXME(comstud): There's a race condition here in that if
            # the system_metadata for this instance is updated after
            # we do the previous save() and before we update.. those
            # other updates will be lost. Since this problem exists in
            # a lot of other places, I think it should be addressed in
            # a DB layer overhaul.

            orig_sys_metadata = dict(instance.system_metadata)
            # Remove the old keys
            for key in list(instance.system_metadata.keys()):
                if key.startswith(utils.SM_IMAGE_PROP_PREFIX):
                    del instance.system_metadata[key]

            # Add the new ones
            new_sys_metadata = utils.get_system_metadata_from_image(
                image, flavor)

            new_sys_metadata.update({'image_base_image_ref': image_id})

            instance.system_metadata.update(new_sys_metadata)
            instance.save()
            return orig_sys_metadata

        # Since image might have changed, we may have new values for
        # os_type, vm_mode, etc
        options_from_image = self._inherit_properties_from_image(
                image, auto_disk_config)
        instance.update(options_from_image)

        instance.task_state = task_states.REBUILDING
        # An empty instance.image_ref is currently used as an indication
        # of BFV.  Preserve that over a rebuild to not break users.
        if not is_volume_backed:
            instance.image_ref = image_href
        instance.kernel_id = kernel_id or ""
        instance.ramdisk_id = ramdisk_id or ""
        instance.progress = 0
        instance.update(kwargs)
        instance.save(expected_task_state=[None])

        # On a rebuild, since we're potentially changing images, we need to
        # wipe out the old image properties that we're storing as instance
        # system metadata... and copy in the properties for the new image.
        orig_sys_metadata = _reset_image_metadata()

        self._record_action_start(context, instance, instance_actions.REBUILD)

        # NOTE(sbauza): The migration script we provided in Newton should make
        # sure that all our instances are currently migrated to have an
        # attached RequestSpec object but let's consider that the operator only
        # half migrated all their instances in the meantime.
        host = instance.host
        # If a new image is provided on rebuild, we will need to run
        # through the scheduler again, but we want the instance to be
        # rebuilt on the same host it's already on.
        if orig_image_ref != image_href:
            # We have to modify the request spec that goes to the scheduler
            # to contain the new image. We persist this since we've already
            # changed the instance.image_ref above so we're being
            # consistent.
            request_spec.image = objects.ImageMeta.from_dict(image)
            request_spec.save()
            if 'scheduler_hints' not in request_spec:
                request_spec.scheduler_hints = {}
            # Nuke the id on this so we can't accidentally save
            # this hint hack later
            del request_spec.id

            # NOTE(danms): Passing host=None tells conductor to
            # call the scheduler. The _nova_check_type hint
            # requires that the scheduler returns only the same
            # host that we are currently on and only checks
            # rebuild-related filters.
            request_spec.scheduler_hints['_nova_check_type'] = ['rebuild']
            request_spec.force_hosts = [instance.host]
            request_spec.force_nodes = [instance.node]
            host = None

        self.compute_task_api.rebuild_instance(context, instance=instance,
                new_pass=admin_password, injected_files=files_to_inject,
                image_ref=image_href, orig_image_ref=orig_image_ref,
                orig_sys_metadata=orig_sys_metadata, bdms=bdms,
                preserve_ephemeral=preserve_ephemeral, host=host,
                request_spec=request_spec)

    def _check_volume_status(self, context, bdms):
        """Check whether the status of the volume is "in-use".

        :param context: A context.RequestContext
        :param bdms: BlockDeviceMappingList of BDMs for the instance
        """
        for bdm in bdms:
            if bdm.volume_id:
                vol = self.volume_api.get(context, bdm.volume_id)
                self.volume_api.check_attached(context, vol)

    @staticmethod
    def _validate_numa_rebuild(instance, image, flavor):
        """validates that the NUMA constraints do not change on rebuild.

        :param instance: nova.objects.instance.Instance object
        :param image: the new image the instance will be rebuilt with.
        :param flavor: the flavor of the current instance.
        :raises: nova.exception.ImageNUMATopologyRebuildConflict
        """

        # NOTE(sean-k-mooney): currently it is not possible to express
        # a PCI NUMA affinity policy via flavor or image but that will
        # change in the future. we pull out the image metadata into
        # separate variable to make future testing of this easier.
        old_image_meta = instance.image_meta
        new_image_meta = objects.ImageMeta.from_dict(image)
        old_constraints = hardware.numa_get_constraints(flavor, old_image_meta)
        new_constraints = hardware.numa_get_constraints(flavor, new_image_meta)

        # early out for non NUMA instances
        if old_constraints is None and new_constraints is None:
            return

        # if only one of the constraints are non-None (or 'set') then the
        # constraints changed so raise an exception.
        if old_constraints is None or new_constraints is None:
            action = "removing" if old_constraints else "introducing"
            LOG.debug("NUMA rebuild validation failed. The requested image "
                      "would alter the NUMA constraints by %s a NUMA "
                      "topology.", action, instance=instance)
            raise exception.ImageNUMATopologyRebuildConflict()

        # otherwise since both the old a new constraints are non none compare
        # them as dictionaries.
        old = old_constraints.obj_to_primitive()
        new = new_constraints.obj_to_primitive()
        if old != new:
            LOG.debug("NUMA rebuild validation failed. The requested image "
                      "conflicts with the existing NUMA constraints.",
                      instance=instance)
            raise exception.ImageNUMATopologyRebuildConflict()
        # TODO(sean-k-mooney): add PCI NUMA affinity policy check.

    @staticmethod
    def _check_quota_for_upsize(context, instance, current_flavor,
                                new_flavor, is_bfv, is_revert):
        project_id, user_id = quotas_obj.ids_from_instance(context,
                                                           instance)
        # NOTE(johngarbutt) for resize, check for sum of existing usage
        # plus the usage from new flavor, as it will be claimed in
        # placement that way, even if there is no change in flavor
        # But for revert resize, we are just removing claims in placement
        # so we can ignore the quota check
        if not is_revert:
            placement_limits.enforce_num_instances_and_flavor(context,
                                                              project_id,
                                                              new_flavor,
                                                              is_bfv, 1, 1)

        # Old quota system only looks at the change in size.
        # Deltas will be empty if the resize is not an upsize.
        deltas = compute_utils.upsize_quota_delta(new_flavor,
                                                  current_flavor)
        if deltas:
            try:
                res_deltas = {'cores': deltas.get('cores', 0),
                              'ram': deltas.get('ram', 0)}
                objects.Quotas.check_deltas(context, res_deltas,
                                            project_id, user_id=user_id,
                                            check_project_id=project_id,
                                            check_user_id=user_id)
            except exception.OverQuota as exc:
                quotas = exc.kwargs['quotas']
                overs = exc.kwargs['overs']
                usages = exc.kwargs['usages']
                headroom = compute_utils.get_headroom(quotas, usages,
                                                      deltas)
                (overs, reqs, total_alloweds,
                 useds) = compute_utils.get_over_quota_detail(headroom,
                                                              overs,
                                                              quotas,
                                                              deltas)
                LOG.info("%(overs)s quota exceeded for %(pid)s,"
                         " tried to resize instance.",
                         {'overs': overs, 'pid': context.project_id})
                raise exception.TooManyInstances(overs=overs,
                                                 req=reqs,
                                                 used=useds,
                                                 allowed=total_alloweds)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.RESIZED])
    def revert_resize(self, context, instance):
        """Reverts a resize or cold migration, deleting the 'new' instance in
        the process.
        """
        elevated = context.elevated()
        migration = objects.Migration.get_by_instance_and_status(
            elevated, instance.uuid, 'finished')

        # If this is a resize down, a revert might go over quota.
        reqspec = objects.RequestSpec.get_by_instance_uuid(
            context, instance.uuid)
        self._check_quota_for_upsize(context, instance, instance.flavor,
                                     instance.old_flavor, reqspec.is_bfv,
                                     is_revert=True)

        # The AZ for the server may have changed when it was migrated so while
        # we are in the API and have access to the API DB, update the
        # instance.availability_zone before casting off to the compute service.
        # Note that we do this in the API to avoid an "up-call" from the
        # compute service to the API DB. This is not great in case something
        # fails during revert before the instance.host is updated to the
        # original source host, but it is good enough for now. Long-term we
        # could consider passing the AZ down to compute so it can set it when
        # the instance.host value is set in finish_revert_resize.
        instance.availability_zone = (
            availability_zones.get_host_availability_zone(
                context, migration.source_compute))

        # If this was a resize, the conductor may have updated the
        # RequestSpec.flavor field (to point at the new flavor) and the
        # RequestSpec.numa_topology field (to reflect the new flavor's extra
        # specs) during the initial resize operation, so we need to update the
        # RequestSpec to point back at the original flavor and reflect the NUMA
        # settings of this flavor, otherwise subsequent move operations through
        # the scheduler will be using the wrong values. There's no need to do
        # this if the flavor hasn't changed though and we're migrating rather
        # than resizing.
        if reqspec.flavor['id'] != instance.old_flavor['id']:
            reqspec.flavor = instance.old_flavor
            reqspec.numa_topology = hardware.numa_get_constraints(
                instance.old_flavor, instance.image_meta)
            reqspec.save()

        # NOTE(gibi): This is a performance optimization. If the network info
        # cache does not have ports with allocations in the binding profile
        # then we can skip reading port resource request from neutron below.
        # If a port has resource request then that would have already caused
        # that the finish_resize call put allocation in the binding profile
        # during the resize.
        if instance.get_network_info().has_port_with_allocation():
            # TODO(gibi): do not directly overwrite the
            # RequestSpec.requested_resources and
            # RequestSpec.request_level_paramsas others like cyborg might added
            # to things there already
            # NOTE(gibi): We need to collect the requested resource again as it
            # is intentionally not persisted in nova. Note that this needs to
            # be done here as the nova API code directly calls revert on the
            # dest compute service skipping the conductor.
            port_res_req, req_lvl_params = (
                self.network_api.get_requested_resource_for_instance(
                    context, instance.uuid))
            reqspec.requested_resources = port_res_req
            reqspec.request_level_params = req_lvl_params

        instance.task_state = task_states.RESIZE_REVERTING
        instance.save(expected_task_state=[None])

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

        self._record_action_start(context, instance,
                                  instance_actions.REVERT_RESIZE)

        if migration.cross_cell_move:
            # RPC cast to conductor to orchestrate the revert of the cross-cell
            # resize.
            self.compute_task_api.revert_snapshot_based_resize(
                context, instance, migration)
        else:
            # TODO(melwitt): We're not rechecking for strict quota here to
            # guard against going over quota during a race at this time because
            # the resource consumption for this operation is written to the
            # database by compute.
            self.compute_rpcapi.revert_resize(context, instance,
                                              migration,
                                              migration.dest_compute,
                                              reqspec)

    @staticmethod
    def _get_source_compute_service(context, migration):
        """Find the source compute Service object given the Migration.

        :param context: nova auth RequestContext target at the destination
            compute cell
        :param migration: Migration object for the move operation
        :return: Service object representing the source host nova-compute
        """
        if migration.cross_cell_move:
            # The source compute could be in another cell so look up the
            # HostMapping to determine the source cell.
            hm = objects.HostMapping.get_by_host(
                context, migration.source_compute)
            with nova_context.target_cell(context, hm.cell_mapping) as cctxt:
                return objects.Service.get_by_compute_host(
                    cctxt, migration.source_compute)
        # Same-cell migration so just use the context we have.
        return objects.Service.get_by_compute_host(
            context, migration.source_compute)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.RESIZED])
    def confirm_resize(self, context, instance, migration=None):
        """Confirms a migration/resize and deletes the 'old' instance.

        :param context: nova auth RequestContext
        :param instance: Instance object to confirm the resize
        :param migration: Migration object; provided if called from the
            _poll_unconfirmed_resizes periodic task on the dest compute.
        :raises: MigrationNotFound if migration is not provided and a migration
            cannot be found for the instance with status "finished".
        :raises: ServiceUnavailable if the source compute service is down.
        """
        elevated = context.elevated()
        # NOTE(melwitt): We're not checking quota here because there isn't a
        # change in resource usage when confirming a resize. Resource
        # consumption for resizes are written to the database by compute, so
        # a confirm resize is just a clean up of the migration objects and a
        # state change in compute.
        if migration is None:
            migration = objects.Migration.get_by_instance_and_status(
                elevated, instance.uuid, 'finished')

        # Check if the source compute service is up before modifying the
        # migration record because once we do we cannot come back through this
        # method since it will be looking for a "finished" status migration.
        source_svc = self._get_source_compute_service(context, migration)
        if not self.servicegroup_api.service_is_up(source_svc):
            raise exception.ServiceUnavailable()

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

        self._record_action_start(context, instance,
                                  instance_actions.CONFIRM_RESIZE)

        # Check to see if this was a cross-cell resize, in which case the
        # resized instance is in the target cell (the migration and instance
        # came from the target cell DB in this case), and we need to cleanup
        # the source host and source cell database records.
        if migration.cross_cell_move:
            self.compute_task_api.confirm_snapshot_based_resize(
                context, instance, migration)
        else:
            # It's a traditional resize within a single cell, so RPC cast to
            # the source compute host to cleanup the host since the instance
            # is already on the target host.
            self.compute_rpcapi.confirm_resize(context,
                                               instance,
                                               migration,
                                               migration.source_compute)

    def _allow_cross_cell_resize(self, context, instance):
        """Determine if the request can perform a cross-cell resize on this
        instance.

        :param context: nova auth request context for the resize operation
        :param instance: Instance object being resized
        :returns: True if cross-cell resize is allowed, False otherwise
        """
        # First check to see if the requesting project/user is allowed by
        # policy to perform cross-cell resize.
        allowed = context.can(
            servers_policies.CROSS_CELL_RESIZE,
            target={'user_id': instance.user_id,
                    'project_id': instance.project_id},
            fatal=False)
        # If the user is allowed by policy, check to make sure the deployment
        # is upgraded to the point of supporting cross-cell resize on all
        # compute services.
        if allowed:
            # TODO(mriedem): We can remove this minimum compute version check
            # in the 22.0.0 "V" release.
            min_compute_version = (
                objects.service.get_minimum_version_all_cells(
                    context, ['nova-compute']))
            if min_compute_version < MIN_COMPUTE_CROSS_CELL_RESIZE:
                LOG.debug('Request is allowed by policy to perform cross-cell '
                          'resize but the minimum nova-compute service '
                          'version in the deployment %s is less than %s so '
                          'cross-cell resize is not allowed at this time.',
                          min_compute_version, MIN_COMPUTE_CROSS_CELL_RESIZE)
                return False

            res_req, req_lvl_params = (
                self.network_api.get_requested_resource_for_instance(
                    context, instance.uuid)
            )
            if res_req:
                LOG.info(
                    'Request is allowed by policy to perform cross-cell '
                    'resize but the instance has ports with resource request '
                    'and cross-cell resize is not supported with such ports.',
                    instance=instance)
                return False

        return allowed

    @staticmethod
    def _validate_host_for_cold_migrate(
            context, instance, host_name, allow_cross_cell_resize):
        """Validates a host specified for cold migration.

        :param context: nova auth request context for the cold migration
        :param instance: Instance object being cold migrated
        :param host_name: User-specified compute service hostname for the
            desired destination of the instance during the cold migration
        :param allow_cross_cell_resize: If True, cross-cell resize is allowed
            for this operation and the host could be in a different cell from
            the one that the instance is currently in. If False, the speciifed
            host must be in the same cell as the instance.
        :returns: ComputeNode object of the requested host
        :raises: CannotMigrateToSameHost if the host is the same as the
            current instance.host
        :raises: ComputeHostNotFound if the specified host cannot be found
        """
        # Cannot migrate to the host where the instance exists
        # because it is useless.
        if host_name == instance.host:
            raise exception.CannotMigrateToSameHost()

        # Check whether host exists or not. If a cross-cell resize is
        # allowed, the host could be in another cell from the one the
        # instance is currently in, so we need to lookup the HostMapping
        # to get the cell and lookup the ComputeNode in that cell.
        if allow_cross_cell_resize:
            try:
                hm = objects.HostMapping.get_by_host(context, host_name)
            except exception.HostMappingNotFound:
                LOG.info('HostMapping not found for host: %s', host_name)
                raise exception.ComputeHostNotFound(host=host_name)

            with nova_context.target_cell(context, hm.cell_mapping) as cctxt:
                node = objects.ComputeNode.\
                    get_first_node_by_host_for_old_compat(
                        cctxt, host_name, use_slave=True)
        else:
            node = objects.ComputeNode.get_first_node_by_host_for_old_compat(
                context, host_name, use_slave=True)

        return node

    # TODO(stephenfin): This logic would be so much easier to grok if we
    # finally split resize and cold migration into separate code paths
    @block_extended_resource_request
    @block_port_accelerators()
    @block_accelerators()
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED])
    @check_instance_host(check_is_up=True)
    def resize(self, context, instance, flavor_id=None, clean_shutdown=True,
               host_name=None, auto_disk_config=None):
        """Resize (ie, migrate) a running instance.

        If flavor_id is None, the process is considered a migration, keeping
        the original flavor_id. If flavor_id is not None, the instance should
        be migrated to a new host and resized to the new flavor_id.
        host_name is always None in the resize case.
        host_name can be set in the cold migration case only.
        """

        allow_cross_cell_resize = self._allow_cross_cell_resize(
            context, instance)

        if host_name is not None:
            node = self._validate_host_for_cold_migrate(
                context, instance, host_name, allow_cross_cell_resize)

        self._check_auto_disk_config(
            instance, auto_disk_config=auto_disk_config)

        current_flavor = instance.get_flavor()

        # NOTE(aarents): Ensure image_base_image_ref is present as it will be
        # needed during finish_resize/cross_cell_resize. Instances upgraded
        # from an older nova release may not have this property because of
        # a rebuild bug Bug/1893618.
        instance.system_metadata.update(
            {'image_base_image_ref': instance.image_ref}
        )

        # If flavor_id is not provided, only migrate the instance.
        volume_backed = None
        if not flavor_id:
            LOG.debug("flavor_id is None. Assuming migration.",
                      instance=instance)
            new_flavor = current_flavor
        else:
            new_flavor = flavors.get_flavor_by_flavor_id(
                flavor_id, read_deleted="no")
            # NOTE(wenping): We use this instead of the 'block_accelerator'
            # decorator since the operation can differ depending on args,
            # and for resize we have two flavors to worry about, we should
            # reject resize with new flavor with accelerator.
            if new_flavor.extra_specs.get('accel:device_profile'):
                raise exception.ForbiddenWithAccelerators()
            # Check to see if we're resizing to a zero-disk flavor which is
            # only supported with volume-backed servers.
            if (new_flavor.get('root_gb') == 0 and
                    current_flavor.get('root_gb') != 0):
                volume_backed = compute_utils.is_volume_backed_instance(
                        context, instance)
                if not volume_backed:
                    reason = _('Resize to zero disk flavor is not allowed.')
                    raise exception.CannotResizeDisk(reason=reason)

        current_flavor_name = current_flavor['name']
        new_flavor_name = new_flavor['name']
        LOG.debug("Old instance type %(current_flavor_name)s, "
                  "new instance type %(new_flavor_name)s",
                  {'current_flavor_name': current_flavor_name,
                   'new_flavor_name': new_flavor_name},
                  instance=instance)

        same_flavor = current_flavor['id'] == new_flavor['id']

        # NOTE(sirp): We don't want to force a customer to change their flavor
        # when Ops is migrating off of a failed host.
        if not same_flavor and new_flavor.get('disabled'):
            raise exception.FlavorNotFound(flavor_id=flavor_id)

        if same_flavor and flavor_id:
            raise exception.CannotResizeToSameFlavor()

        # ensure there is sufficient headroom for upsizes
        if flavor_id:
            # Figure out if the instance is volume-backed but only if we didn't
            # already figure that out above (avoid the extra db hit).
            if volume_backed is None:
                # TODO(johngarbutt) should we just use the request spec?
                volume_backed = compute_utils.is_volume_backed_instance(
                    context, instance)
            self._check_quota_for_upsize(context, instance,
                                         current_flavor,
                                         new_flavor, volume_backed,
                                         is_revert=False)

        if not same_flavor:
            image = utils.get_image_from_system_metadata(
                instance.system_metadata)
            # Figure out if the instance is volume-backed but only if we didn't
            # already figure that out above (avoid the extra db hit).
            if volume_backed is None:
                volume_backed = compute_utils.is_volume_backed_instance(
                    context, instance)
            # If the server is volume-backed, we still want to validate numa
            # and pci information in the new flavor, but we don't call
            # _validate_flavor_image_nostatus because how it handles checking
            # disk size validation was not intended for a volume-backed
            # resize case.
            if volume_backed:
                self._validate_flavor_image_numa_pci(
                    image, new_flavor, validate_pci=True)
            else:
                self._validate_flavor_image_nostatus(
                    context, image, new_flavor, root_bdm=None,
                    validate_pci=True)

        filter_properties = {'ignore_hosts': []}
        if not self._allow_resize_to_same_host(same_flavor, instance):
            filter_properties['ignore_hosts'].append(instance.host)

        request_spec = objects.RequestSpec.get_by_instance_uuid(
            context, instance.uuid)
        request_spec.ignore_hosts = filter_properties['ignore_hosts']

        # don't recalculate the NUMA topology unless the flavor has changed
        if not same_flavor:
            request_spec.numa_topology = hardware.numa_get_constraints(
                new_flavor, instance.image_meta)
            # if the flavor is changed then we need to recalculate the
            # pci_requests as well because the new flavor might request
            # different pci_aliases
            new_pci_requests = pci_request.get_pci_requests_from_flavor(
                new_flavor)
            new_pci_requests.instance_uuid = instance.uuid
            # The neutron based InstancePCIRequest cannot change during resize,
            # so we just need to copy them from the old request
            for request in request_spec.pci_requests.requests or []:
                if request.source == objects.InstancePCIRequest.NEUTRON_PORT:
                    new_pci_requests.requests.append(request)
            request_spec.pci_requests = new_pci_requests

            # TODO(huaqiang): Remove in Wallaby
            # check nova-compute nodes have been updated to Victoria to resize
            # instance to a new mixed instance from a dedicated or shared
            # instance.
            self._check_compute_service_for_mixed_instance(
                request_spec.numa_topology)

        instance.task_state = task_states.RESIZE_PREP
        instance.progress = 0
        instance.auto_disk_config = auto_disk_config or False
        instance.save(expected_task_state=[None])

        if not flavor_id:
            self._record_action_start(context, instance,
                                      instance_actions.MIGRATE)
        else:
            self._record_action_start(context, instance,
                                      instance_actions.RESIZE)

        # TODO(melwitt): We're not rechecking for strict quota here to guard
        # against going over quota during a race at this time because the
        # resource consumption for this operation is written to the database
        # by compute.
        scheduler_hint = {'filter_properties': filter_properties}

        if host_name is None:
            # If 'host_name' is not specified,
            # clear the 'requested_destination' field of the RequestSpec
            # except set the allow_cross_cell_move flag since conductor uses
            # it prior to scheduling.
            request_spec.requested_destination = objects.Destination(
                allow_cross_cell_move=allow_cross_cell_resize)
        else:
            # Set the host and the node so that the scheduler will
            # validate them.
            request_spec.requested_destination = objects.Destination(
                host=node.host, node=node.hypervisor_hostname,
                allow_cross_cell_move=allow_cross_cell_resize)

        # Asynchronously RPC cast to conductor so the response is not blocked
        # during scheduling. If something fails the user can find out via
        # instance actions.
        self.compute_task_api.resize_instance(
            context, instance,
            scheduler_hint=scheduler_hint,
            flavor=new_flavor,
            clean_shutdown=clean_shutdown,
            request_spec=request_spec,
            do_cast=True)

    def _allow_resize_to_same_host(self, cold_migrate, instance):
        """Contains logic for excluding the instance.host on resize/migrate.

        If performing a cold migration and the compute node resource provider
        reports the COMPUTE_SAME_HOST_COLD_MIGRATE trait then same-host cold
        migration is allowed otherwise it is not and the current instance.host
        should be excluded as a scheduling candidate.

        :param cold_migrate: true if performing a cold migration, false
            for resize
        :param instance: Instance object being resized or cold migrated
        :returns: True if same-host resize/cold migrate is allowed, False
            otherwise
        """
        if cold_migrate:
            # Check to see if the compute node resource provider on which the
            # instance is running has the COMPUTE_SAME_HOST_COLD_MIGRATE
            # trait.
            # Note that we check this here in the API since we cannot
            # pre-filter allocation candidates in the scheduler using this
            # trait as it would not work. For example, libvirt nodes will not
            # report the trait but using it as a forbidden trait filter when
            # getting allocation candidates would still return libvirt nodes
            # which means we could attempt to cold migrate to the same libvirt
            # node, which would fail.
            ctxt = instance._context
            cn = objects.ComputeNode.get_by_host_and_nodename(
                ctxt, instance.host, instance.node)
            traits = self.placementclient.get_provider_traits(
                ctxt, cn.uuid).traits
            # If the provider has the trait it is (1) new enough to report that
            # trait and (2) supports cold migration on the same host.
            if os_traits.COMPUTE_SAME_HOST_COLD_MIGRATE in traits:
                allow_same_host = True
            else:
                # TODO(mriedem): Remove this compatibility code after one
                # release. If the compute is old we will not know if it
                # supports same-host cold migration so we fallback to config.
                service = objects.Service.get_by_compute_host(ctxt, cn.host)
                if service.version >= MIN_COMPUTE_SAME_HOST_COLD_MIGRATE:
                    # The compute is new enough to report the trait but does
                    # not so same-host cold migration is not allowed.
                    allow_same_host = False
                else:
                    # The compute is not new enough to report the trait so we
                    # fallback to config.
                    allow_same_host = CONF.allow_resize_to_same_host
        else:
            allow_same_host = CONF.allow_resize_to_same_host
        return allow_same_host

    @block_port_accelerators()
    @reject_vtpm_instances(instance_actions.SHELVE)
    @block_accelerators(until_service=54)
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.PAUSED, vm_states.SUSPENDED])
    def shelve(self, context, instance, clean_shutdown=True):
        """Shelve an instance.

        Shuts down an instance and frees it up to be removed from the
        hypervisor.
        """
        instance.task_state = task_states.SHELVING

        # NOTE(aarents): Ensure image_base_image_ref is present as it will be
        # needed during unshelve and instance rebuild done before Bug/1893618
        # Fix dropped it.
        instance.system_metadata.update(
                {'image_base_image_ref': instance.image_ref}
        )
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance, instance_actions.SHELVE)

        accel_uuids = []
        if instance.flavor.extra_specs.get('accel:device_profile'):
            cyclient = cyborg.get_client(context)
            accel_uuids = cyclient.get_arq_uuids_for_instance(instance)

        if not compute_utils.is_volume_backed_instance(context, instance):
            name = '%s-shelved' % instance.display_name
            image_meta = compute_utils.create_image(
                context, instance, name, 'snapshot', self.image_api)
            image_id = image_meta['id']
            self.compute_rpcapi.shelve_instance(context, instance=instance,
                image_id=image_id, clean_shutdown=clean_shutdown,
                accel_uuids=accel_uuids)
        else:
            self.compute_rpcapi.shelve_offload_instance(
                context, instance=instance, clean_shutdown=clean_shutdown,
                accel_uuids=accel_uuids)

    @block_port_accelerators()
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.SHELVED])
    def shelve_offload(self, context, instance, clean_shutdown=True):
        """Remove a shelved instance from the hypervisor."""
        instance.task_state = task_states.SHELVING_OFFLOADING
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance,
                                  instance_actions.SHELVE_OFFLOAD)

        accel_uuids = []
        if instance.flavor.extra_specs.get('accel:device_profile'):
            cyclient = cyborg.get_client(context)
            accel_uuids = cyclient.get_arq_uuids_for_instance(instance)

        self.compute_rpcapi.shelve_offload_instance(
            context, instance=instance,
            clean_shutdown=clean_shutdown, accel_uuids=accel_uuids)

    def _check_offloaded(self, context, instance):
        """Check if the status of an instance is SHELVE_OFFLOADED,
        if not raise an exception.
        """
        if instance.vm_state != vm_states.SHELVED_OFFLOADED:
            # NOTE(brinzhang): If the server status is 'SHELVED', it still
            # belongs to a host, the availability_zone should not change.
            # Unshelving a shelved offloaded server will go through the
            # scheduler to find a new host.
            raise exception.UnshelveInstanceInvalidState(
                state=instance.vm_state, instance_uuid=instance.uuid)

    def _ensure_host_in_az(self, context, host, availability_zone):
        """Ensure the host provided belongs to the availability zone,
        if not raise an exception.
        """
        if availability_zone is not None:
            host_az = availability_zones.get_host_availability_zone(
                context,
                host
            )
            if host_az != availability_zone:
                raise exception.UnshelveHostNotInAZ(
                    host=host, availability_zone=availability_zone)

    def _validate_unshelve_az(self, context, instance, availability_zone):
        """Verify the specified availability_zone during unshelve.

        Verifies the AZ exists and if [cinder]/cross_az_attach=False, that
        any attached volumes are in the same AZ.

        :param context: nova auth RequestContext for the unshelve action
        :param instance: Instance object for the server being unshelved
        :param availability_zone: The user-requested availability zone in
            which to unshelve the server.
        :raises: InvalidRequest if the requested AZ does not exist
        :raises: MismatchVolumeAZException if [cinder]/cross_az_attach=False
            and any attached volumes are not in the requested AZ
        """
        available_zones = availability_zones.get_availability_zones(
            context, self.host_api, get_only_available=True)
        if availability_zone not in available_zones:
            msg = _('The requested availability zone is not available')
            raise exception.InvalidRequest(msg)

        # NOTE(brinzhang): When specifying a availability zone to unshelve
        # a shelved offloaded server, and conf cross_az_attach=False, need
        # to determine if attached volume AZ matches the user-specified AZ.
        if not CONF.cinder.cross_az_attach:
            bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
                context, instance.uuid)
            for bdm in bdms:
                if bdm.is_volume and bdm.volume_id:
                    volume = self.volume_api.get(context, bdm.volume_id)
                    if availability_zone != volume['availability_zone']:
                        msg = _("The specified availability zone does not "
                                "match the volume %(vol_id)s attached to the "
                                "server. Specified availability zone is "
                                "%(az)s. Volume is in %(vol_zone)s.") % {
                            "vol_id": volume['id'],
                            "az": availability_zone,
                            "vol_zone": volume['availability_zone']}
                        raise exception.MismatchVolumeAZException(reason=msg)

    @block_extended_resource_request
    @check_instance_lock
    @check_instance_state(
            vm_state=[vm_states.SHELVED, vm_states.SHELVED_OFFLOADED])
    def unshelve(
            self, context, instance, new_az=_sentinel, host=None):
        """Restore a shelved instance.

        :param context: the nova request context
        :param instance: nova.objects.instance.Instance object
        :param new_az: (optional) target AZ.
                       If None is provided then the current AZ restriction
                       will be removed from the instance.
                       If the parameter is not provided then the current
                       AZ restriction will not be changed.
        :param host: (optional) a host to target
        """
        # Unshelving a shelved offloaded server will go through the
        # scheduler to pick a new host, so we update the
        # RequestSpec.availability_zone here. Note that if scheduling
        # fails the RequestSpec will remain updated, which is not great.
        # Bug open to track this https://bugs.launchpad.net/nova/+bug/1978573

        az_passed = new_az is not self._sentinel

        request_spec = objects.RequestSpec.get_by_instance_uuid(
            context, instance.uuid)

        # We need to check a list of preconditions and validate inputs first

        # Ensure instance is shelve offloaded
        if az_passed or host:
            self._check_offloaded(context, instance)

        if az_passed and new_az:
            # we have to ensure that new AZ is valid
            self._validate_unshelve_az(context, instance, new_az)
        # This will be the AZ of the instance after the unshelve. It can be
        # None indicating that the instance is not pinned to any AZ after the
        # unshelve
        expected_az_after_unshelve = (
            request_spec.availability_zone
            if not az_passed else new_az
        )
        # host is requested, so we have to see if it exists and does not
        # contradict with the AZ of the instance
        if host:
            # Make sure only admin can unshelve to a specific host.
            context.can(
                shelve_policies.POLICY_ROOT % 'unshelve_to_host',
                target={
                    'user_id': instance.user_id,
                    'project_id': instance.project_id
                }
            )
            # Ensure that the requested host exists otherwise raise
            # a ComputeHostNotFound exception
            objects.ComputeNode.get_first_node_by_host_for_old_compat(
                context, host, use_slave=True)
            # A specific host is requested so we need to make sure that it is
            # not contradicts with the AZ of the instance
            self._ensure_host_in_az(
                context, host, expected_az_after_unshelve)

        if new_az is None:
            LOG.debug(
                'Unpin instance from AZ "%(old_az)s".',
                {'old_az': request_spec.availability_zone},
                instance=instance
            )

        LOG.debug(
            'Unshelving instance with old availability_zone "%(old_az)s" to '
            'new availability_zone "%(new_az)s" and host "%(host)s".',
            {
                'old_az': request_spec.availability_zone,
                'new_az': '%s' %
                          new_az if az_passed
                                 else 'not provided',
                'host': host,
             },
            instance=instance,
        )
        # OK every precondition checks out, we just need to tell the scheduler
        # where to put the instance
        # We have the expected AZ already calculated. So we just need to
        # set it in the request_spec to drive the scheduling
        request_spec.availability_zone = expected_az_after_unshelve
        # if host is requested we also need to tell the scheduler that
        if host:
            request_spec.requested_destination = objects.Destination(host=host)
        request_spec.save()

        instance.task_state = task_states.UNSHELVING
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance, instance_actions.UNSHELVE)

        self.compute_task_api.unshelve_instance(context, instance,
                                                request_spec)

    @check_instance_lock
    def add_fixed_ip(self, context, instance, network_id):
        """Add fixed_ip from specified network to given instance."""
        self.compute_rpcapi.add_fixed_ip_to_instance(context,
                instance=instance, network_id=network_id)

    @check_instance_lock
    def remove_fixed_ip(self, context, instance, address):
        """Remove fixed_ip from specified network to given instance."""
        self.compute_rpcapi.remove_fixed_ip_from_instance(context,
                instance=instance, address=address)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE])
    def pause(self, context, instance):
        """Pause the given instance."""
        instance.task_state = task_states.PAUSING
        instance.save(expected_task_state=[None])
        self._record_action_start(context, instance, instance_actions.PAUSE)
        self.compute_rpcapi.pause_instance(context, instance)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.PAUSED])
    def unpause(self, context, instance):
        """Unpause the given instance."""
        instance.task_state = task_states.UNPAUSING
        instance.save(expected_task_state=[None])
        self._record_action_start(context, instance, instance_actions.UNPAUSE)
        self.compute_rpcapi.unpause_instance(context, instance)

    @check_instance_host()
    def get_diagnostics(self, context, instance):
        """Retrieve diagnostics for the given instance."""
        return self.compute_rpcapi.get_diagnostics(context, instance=instance)

    @check_instance_host()
    def get_instance_diagnostics(self, context, instance):
        """Retrieve diagnostics for the given instance."""
        return self.compute_rpcapi.get_instance_diagnostics(context,
                                                            instance=instance)

    @block_port_accelerators()
    @reject_vdpa_instances(
        instance_actions.SUSPEND, until=MIN_COMPUTE_VDPA_HOTPLUG_LIVE_MIGRATION
    )
    @block_accelerators()
    @reject_sev_instances(instance_actions.SUSPEND)
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE])
    def suspend(self, context, instance):
        """Suspend the given instance."""
        instance.task_state = task_states.SUSPENDING
        instance.save(expected_task_state=[None])
        self._record_action_start(context, instance, instance_actions.SUSPEND)
        self.compute_rpcapi.suspend_instance(context, instance)

    @check_instance_lock
    @reject_vdpa_instances(
        instance_actions.RESUME, until=MIN_COMPUTE_VDPA_HOTPLUG_LIVE_MIGRATION
    )
    @check_instance_state(vm_state=[vm_states.SUSPENDED])
    def resume(self, context, instance):
        """Resume the given instance."""
        instance.task_state = task_states.RESUMING
        instance.save(expected_task_state=[None])
        self._record_action_start(context, instance, instance_actions.RESUME)
        self.compute_rpcapi.resume_instance(context, instance)

    @reject_vtpm_instances(instance_actions.RESCUE)
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.ERROR])
    def rescue(self, context, instance, rescue_password=None,
               rescue_image_ref=None, clean_shutdown=True,
               allow_bfv_rescue=False):
        """Rescue the given instance."""

        if rescue_image_ref:
            try:
                image_meta = image_meta_obj.ImageMeta.from_image_ref(
                    context, self.image_api, rescue_image_ref)
            except (exception.ImageNotFound, exception.ImageBadRequest):
                LOG.warning("Failed to fetch rescue image metadata using "
                            "image_ref %(image_ref)s",
                            {'image_ref': rescue_image_ref})
                raise exception.UnsupportedRescueImage(
                    image=rescue_image_ref)

            # FIXME(lyarwood): There is currently no support for rescuing
            # instances using a volume snapshot so fail here before we cast to
            # the compute.
            if image_meta.properties.get('img_block_device_mapping'):
                LOG.warning("Unable to rescue an instance using a volume "
                            "snapshot image with img_block_device_mapping "
                            "image properties set")
                raise exception.UnsupportedRescueImage(
                    image=rescue_image_ref)

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

        volume_backed = compute_utils.is_volume_backed_instance(
            context, instance, bdms)

        if volume_backed and allow_bfv_rescue:
            cn = objects.ComputeNode.get_by_host_and_nodename(
                context, instance.host, instance.node)
            traits = self.placementclient.get_provider_traits(
                context, cn.uuid).traits
            if os_traits.COMPUTE_RESCUE_BFV not in traits:
                reason = _("Host unable to rescue a volume-backed instance")
                raise exception.InstanceNotRescuable(instance_id=instance.uuid,
                                                     reason=reason)
        elif volume_backed:
            reason = _("Cannot rescue a volume-backed instance")
            raise exception.InstanceNotRescuable(instance_id=instance.uuid,
                                                 reason=reason)

        instance.task_state = task_states.RESCUING
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance, instance_actions.RESCUE)

        self.compute_rpcapi.rescue_instance(context, instance=instance,
            rescue_password=rescue_password, rescue_image_ref=rescue_image_ref,
            clean_shutdown=clean_shutdown)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.RESCUED])
    def unrescue(self, context, instance):
        """Unrescue the given instance."""
        instance.task_state = task_states.UNRESCUING
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance, instance_actions.UNRESCUE)

        self.compute_rpcapi.unrescue_instance(context, instance=instance)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE])
    def set_admin_password(self, context, instance, password):
        """Set the root/admin password for the given instance.

        @param context: Nova auth context.
        @param instance: Nova instance object.
        @param password: The admin password for the instance.
        """
        instance.task_state = task_states.UPDATING_PASSWORD
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance,
                                  instance_actions.CHANGE_PASSWORD)

        self.compute_rpcapi.set_admin_password(context,
                                               instance=instance,
                                               new_pass=password)

    @check_instance_host()
    @reject_instance_state(
        task_state=[task_states.DELETING, task_states.MIGRATING])
    def get_vnc_console(self, context, instance, console_type):
        """Get a url to an instance Console."""
        connect_info = self.compute_rpcapi.get_vnc_console(context,
                instance=instance, console_type=console_type)
        return {'url': connect_info['access_url']}

    @check_instance_host()
    @reject_instance_state(
        task_state=[task_states.DELETING, task_states.MIGRATING])
    def get_spice_console(self, context, instance, console_type):
        """Get a url to an instance Console."""
        connect_info = self.compute_rpcapi.get_spice_console(context,
                instance=instance, console_type=console_type)
        return {'url': connect_info['access_url']}

    @check_instance_host()
    @reject_instance_state(
        task_state=[task_states.DELETING, task_states.MIGRATING])
    def get_rdp_console(self, context, instance, console_type):
        """Get a url to an instance Console."""
        connect_info = self.compute_rpcapi.get_rdp_console(context,
                instance=instance, console_type=console_type)
        return {'url': connect_info['access_url']}

    @check_instance_host()
    @reject_instance_state(
        task_state=[task_states.DELETING, task_states.MIGRATING])
    def get_serial_console(self, context, instance, console_type):
        """Get a url to a serial console."""
        connect_info = self.compute_rpcapi.get_serial_console(context,
                instance=instance, console_type=console_type)
        return {'url': connect_info['access_url']}

    @check_instance_host()
    @reject_instance_state(
        task_state=[task_states.DELETING, task_states.MIGRATING])
    def get_mks_console(self, context, instance, console_type):
        """Get a url to a MKS console."""
        connect_info = self.compute_rpcapi.get_mks_console(context,
                instance=instance, console_type=console_type)
        return {'url': connect_info['access_url']}

    @check_instance_host()
    def get_console_output(self, context, instance, tail_length=None):
        """Get console output for an instance."""
        return self.compute_rpcapi.get_console_output(context,
                instance=instance, tail_length=tail_length)

    def lock(self, context, instance, reason=None):
        """Lock the given instance."""
        # Only update the lock if we are an admin (non-owner)
        is_owner = instance.project_id == context.project_id
        if instance.locked and is_owner:
            return

        context = context.elevated()
        self._record_action_start(context, instance,
                                  instance_actions.LOCK)

        @wrap_instance_event(prefix='api')
        def lock(self, context, instance, reason=None):
            LOG.debug('Locking', instance=instance)
            instance.locked = True
            instance.locked_by = 'owner' if is_owner else 'admin'
            if reason:
                instance.system_metadata['locked_reason'] = reason
            instance.save()

        lock(self, context, instance, reason=reason)
        compute_utils.notify_about_instance_action(
            context, instance, CONF.host,
            action=fields_obj.NotificationAction.LOCK,
            source=fields_obj.NotificationSource.API)

    def is_expected_locked_by(self, context, instance):
        is_owner = instance.project_id == context.project_id
        expect_locked_by = 'owner' if is_owner else 'admin'
        locked_by = instance.locked_by
        if locked_by and locked_by != expect_locked_by:
            return False
        return True

    def unlock(self, context, instance):
        """Unlock the given instance."""
        context = context.elevated()
        self._record_action_start(context, instance,
                                  instance_actions.UNLOCK)

        @wrap_instance_event(prefix='api')
        def unlock(self, context, instance):
            LOG.debug('Unlocking', instance=instance)
            instance.locked = False
            instance.locked_by = None
            instance.system_metadata.pop('locked_reason', None)
            instance.save()

        unlock(self, context, instance)
        compute_utils.notify_about_instance_action(
            context, instance, CONF.host,
            action=fields_obj.NotificationAction.UNLOCK,
            source=fields_obj.NotificationSource.API)

    @check_instance_lock
    def inject_network_info(self, context, instance):
        """Inject network info for the instance."""
        self.compute_rpcapi.inject_network_info(context, instance=instance)

    def _create_volume_bdm(self, context, instance, device, volume,
                           disk_bus, device_type, is_local_creation=False,
                           tag=None, delete_on_termination=False):
        volume_id = volume['id']
        if is_local_creation:
            # when the creation is done locally we can't specify the device
            # name as we do not have a way to check that the name specified is
            # a valid one.
            # We leave the setting of that value when the actual attach
            # happens on the compute manager
            # NOTE(artom) Local attach (to a shelved-offload instance) cannot
            # support device tagging because we have no way to call the compute
            # manager to check that it supports device tagging. In fact, we
            # don't even know which computer manager the instance will
            # eventually end up on when it's unshelved.
            volume_bdm = objects.BlockDeviceMapping(
                context=context,
                source_type='volume', destination_type='volume',
                instance_uuid=instance.uuid, boot_index=None,
                volume_id=volume_id,
                device_name=None, guest_format=None,
                disk_bus=disk_bus, device_type=device_type,
                delete_on_termination=delete_on_termination)
            volume_bdm.create()
        else:
            # NOTE(vish): This is done on the compute host because we want
            #             to avoid a race where two devices are requested at
            #             the same time. When db access is removed from
            #             compute, the bdm will be created here and we will
            #             have to make sure that they are assigned atomically.
            volume_bdm = self.compute_rpcapi.reserve_block_device_name(
                context, instance, device, volume_id, disk_bus=disk_bus,
                device_type=device_type, tag=tag,
                multiattach=volume['multiattach'])
            volume_bdm.delete_on_termination = delete_on_termination
            volume_bdm.save()
        return volume_bdm

    def _check_volume_already_attached(
        self,
        context: nova_context.RequestContext,
        instance: objects.Instance,
        volume: ty.Mapping[str, ty.Any],
    ):
        """Avoid duplicate volume attachments.

        Since the 3.44 Cinder microversion, Cinder allows us to attach the same
        volume to the same instance twice. This is ostensibly to enable live
        migration, but it's not something we want to occur outside of this
        particular code path.

        In addition, we also need to ensure that non-multiattached volumes are
        not attached to multiple instances. This check is also carried out
        later by c-api itself but it can however be circumvented by admins
        resetting the state of an attached volume to available. As a result we
        also need to perform a check within Nova before creating a new BDM for
        the attachment.

        :param context: nova auth RequestContext
        :param instance: Instance object
        :param volume: volume dict from cinder
        """
        # Fetch a list of active bdms for the volume, return if none are found.
        try:
            bdms = objects.BlockDeviceMappingList.get_by_volume(
                context, volume['id'])
        except exception.VolumeBDMNotFound:
            return

        # Fail if the volume isn't multiattach but BDMs already exist
        if not volume.get('multiattach'):
            instance_uuids = ' '.join(f"{b.instance_uuid}" for b in bdms)
            msg = _(
                "volume %(volume_id)s is already attached to instances: "
                "%(instance_uuids)s"
            ) % {
                'volume_id': volume['id'],
                'instance_uuids': instance_uuids
            }
            raise exception.InvalidVolume(reason=msg)

        # Fail if the volume is already attached to our instance
        if any(b for b in bdms if b.instance_uuid == instance.uuid):
            msg = _("volume %s already attached") % volume['id']
            raise exception.InvalidVolume(reason=msg)

    def _check_attach_and_reserve_volume(self, context, volume, instance,
                                         bdm, supports_multiattach=False,
                                         validate_az=True):
        """Perform checks against the instance and volume before attaching.

        If validation succeeds, the bdm is updated with an attachment_id which
        effectively reserves it during the attach process in cinder.

        :param context: nova auth RequestContext
        :param volume: volume dict from cinder
        :param instance: Instance object
        :param bdm: BlockDeviceMapping object
        :param supports_multiattach: True if the request supports multiattach
            volumes, i.e. microversion >= 2.60, False otherwise
        :param validate_az: True if the instance and volume availability zones
            should be validated for cross_az_attach, False to not validate AZ
        """
        volume_id = volume['id']
        if validate_az:
            self.volume_api.check_availability_zone(context, volume,
                                                    instance=instance)
        # If volume.multiattach=True and the microversion to
        # support multiattach is not used, fail the request.
        if volume['multiattach'] and not supports_multiattach:
            raise exception.MultiattachNotSupportedOldMicroversion()

        attachment_id = self.volume_api.attachment_create(
            context, volume_id, instance.uuid)['id']
        bdm.attachment_id = attachment_id
        # NOTE(ildikov): In case of boot from volume the BDM at this
        # point is not yet created in a cell database, so we can't
        # call save().  When attaching a volume to an existing
        # instance, the instance is already in a cell and the BDM has
        # been created in that same cell so updating here in that case
        # is "ok".
        if bdm.obj_attr_is_set('id'):
            bdm.save()

    # TODO(stephenfin): Fold this back in now that cells v1 no longer needs to
    # override it.
    def _attach_volume(self, context, instance, volume, device,
                       disk_bus, device_type, tag=None,
                       supports_multiattach=False,
                       delete_on_termination=False):
        """Attach an existing volume to an existing instance.

        This method is separated to make it possible for cells version
        to override it.
        """
        try:
            volume_bdm = self._create_volume_bdm(
                context, instance, device, volume, disk_bus=disk_bus,
                device_type=device_type, tag=tag,
                delete_on_termination=delete_on_termination)
        except oslo_exceptions.MessagingTimeout:
            # The compute node might have already created the attachment but
            # we never received the answer. In this case it is safe to delete
            # the attachment as nobody will ever pick it up again.
            with excutils.save_and_reraise_exception():
                try:
                    objects.BlockDeviceMapping.get_by_volume_and_instance(
                        context, volume['id'], instance.uuid).destroy()
                    LOG.debug("Delete BDM after compute did not respond to "
                              f"attachment request for volume {volume['id']}")
                except exception.VolumeBDMNotFound:
                    LOG.debug("BDM not found, ignoring removal. "
                              f"Error attaching volume {volume['id']}")
        try:
            self._check_attach_and_reserve_volume(context, volume, instance,
                                                  volume_bdm,
                                                  supports_multiattach)
            self._record_action_start(
                context, instance, instance_actions.ATTACH_VOLUME)
            self.compute_rpcapi.attach_volume(context, instance, volume_bdm)
        except Exception:
            with excutils.save_and_reraise_exception():
                volume_bdm.destroy()

        return volume_bdm.device_name

    def _attach_volume_shelved_offloaded(self, context, instance, volume,
                                         device, disk_bus, device_type,
                                         delete_on_termination):
        """Attach an existing volume to an instance in shelved offloaded state.

        Attaching a volume for an instance in shelved offloaded state requires
        to perform the regular check to see if we can attach and reserve the
        volume then we need to call the attach method on the volume API
        to mark the volume as 'in-use'.
        The instance at this stage is not managed by a compute manager
        therefore the actual attachment will be performed once the
        instance will be unshelved.
        """
        volume_id = volume['id']

        @wrap_instance_event(prefix='api')
        def attach_volume(self, context, v_id, instance, dev, attachment_id):
            if attachment_id:
                # Normally we wouldn't complete an attachment without a host
                # connector, but we do this to make the volume status change
                # to "in-use" to maintain the API semantics with the old flow.
                # When unshelving the instance, the compute service will deal
                # with this disconnected attachment.
                self.volume_api.attachment_complete(context, attachment_id)
            else:
                self.volume_api.attach(context,
                                       v_id,
                                       instance.uuid,
                                       dev)

        volume_bdm = self._create_volume_bdm(
            context, instance, device, volume, disk_bus=disk_bus,
            device_type=device_type, is_local_creation=True,
            delete_on_termination=delete_on_termination)
        try:
            self._check_attach_and_reserve_volume(context, volume, instance,
                                                  volume_bdm)
            self._record_action_start(
                context, instance,
                instance_actions.ATTACH_VOLUME)
            attach_volume(self, context, volume_id, instance, device,
                          volume_bdm.attachment_id)
        except Exception:
            with excutils.save_and_reraise_exception():
                volume_bdm.destroy()

        return volume_bdm.device_name

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.PAUSED,
                                    vm_states.STOPPED, vm_states.RESIZED,
                                    vm_states.SOFT_DELETED, vm_states.SHELVED,
                                    vm_states.SHELVED_OFFLOADED])
    def attach_volume(self, context, instance, volume_id, device=None,
                      disk_bus=None, device_type=None, tag=None,
                      supports_multiattach=False,
                      delete_on_termination=False):
        """Attach an existing volume to an existing instance."""
        # NOTE(vish): Fail fast if the device is not going to pass. This
        #             will need to be removed along with the test if we
        #             change the logic in the manager for what constitutes
        #             a valid device.
        if device and not block_device.match_device(device):
            raise exception.InvalidDevicePath(path=device)

        # Make sure the volume isn't already attached to this instance
        # because we'll use the v3.44 attachment flow in
        # _check_attach_and_reserve_volume and Cinder will allow multiple
        # attachments between the same volume and instance but the old flow
        # API semantics don't allow that so we enforce it here.
        # NOTE(lyarwood): Ensure that non multiattach volumes don't already
        # have active block device mappings present in Nova.
        volume = self.volume_api.get(context, volume_id)
        self._check_volume_already_attached(context, instance, volume)

        is_shelved_offloaded = instance.vm_state == vm_states.SHELVED_OFFLOADED
        if is_shelved_offloaded:
            if tag:
                # NOTE(artom) Local attach (to a shelved-offload instance)
                # cannot support device tagging because we have no way to call
                # the compute manager to check that it supports device tagging.
                # In fact, we don't even know which computer manager the
                # instance will eventually end up on when it's unshelved.
                raise exception.VolumeTaggedAttachToShelvedNotSupported()
            if volume['multiattach']:
                # NOTE(mriedem): Similar to tagged attach, we don't support
                # attaching a multiattach volume to shelved offloaded instances
                # because we can't tell if the compute host (since there isn't
                # one) supports it. This could possibly be supported in the
                # future if the scheduler was made aware of which computes
                # support multiattach volumes.
                raise exception.MultiattachToShelvedNotSupported()
            return self._attach_volume_shelved_offloaded(context,
                                                         instance,
                                                         volume,
                                                         device,
                                                         disk_bus,
                                                         device_type,
                                                         delete_on_termination)

        return self._attach_volume(context, instance, volume, device,
                                   disk_bus, device_type, tag=tag,
                                   supports_multiattach=supports_multiattach,
                                   delete_on_termination=delete_on_termination)

    def _detach_volume_shelved_offloaded(self, context, instance, volume):
        """Detach a volume from an instance in shelved offloaded state.

        If the instance is shelved offloaded we just need to cleanup volume
        calling the volume api detach, the volume api terminate_connection
        and delete the bdm record.
        If the volume has delete_on_termination option set then we call the
        volume api delete as well.
        """
        @wrap_instance_event(prefix='api')
        def detach_volume(self, context, instance, bdms):
            self._local_cleanup_bdm_volumes(bdms, instance, context)

        bdms = [objects.BlockDeviceMapping.get_by_volume_id(
                context, volume['id'], instance.uuid)]
        # The begin_detaching() call only works with in-use volumes,
        # which will not be the case for volumes attached to a shelved
        # offloaded server via the attachments API since those volumes
        # will have `reserved` status.
        if not bdms[0].attachment_id:
            try:
                self.volume_api.begin_detaching(context, volume['id'])
            except exception.InvalidInput as exc:
                raise exception.InvalidVolume(reason=exc.format_message())
        self._record_action_start(
            context, instance,
            instance_actions.DETACH_VOLUME)
        detach_volume(self, context, instance, bdms)

    @check_instance_host(check_is_up=True)
    def _detach_volume(self, context, instance, volume):
        try:
            self.volume_api.begin_detaching(context, volume['id'])
        except exception.InvalidInput as exc:
            raise exception.InvalidVolume(reason=exc.format_message())
        attachments = volume.get('attachments', {})
        attachment_id = None
        if attachments and instance.uuid in attachments:
            attachment_id = attachments[instance.uuid]['attachment_id']
        self._record_action_start(
            context, instance, instance_actions.DETACH_VOLUME)
        self.compute_rpcapi.detach_volume(context, instance=instance,
                volume_id=volume['id'], attachment_id=attachment_id)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.PAUSED,
                                    vm_states.STOPPED, vm_states.RESIZED,
                                    vm_states.SOFT_DELETED, vm_states.SHELVED,
                                    vm_states.SHELVED_OFFLOADED])
    def detach_volume(self, context, instance, volume):
        """Detach a volume from an instance."""
        if instance.vm_state == vm_states.SHELVED_OFFLOADED:
            self._detach_volume_shelved_offloaded(context, instance, volume)
        else:
            self._detach_volume(context, instance, volume)

    def _count_attachments_for_swap(self, ctxt, volume):
        """Counts the number of attachments for a swap-related volume.

        Attempts to only count read/write attachments if the volume attachment
        records exist, otherwise simply just counts the number of attachments
        regardless of attach mode.

        :param ctxt: nova.context.RequestContext - user request context
        :param volume: nova-translated volume dict from nova.volume.cinder.
        :returns: count of attachments for the volume
        """
        # This is a dict, keyed by server ID, to a dict of attachment_id and
        # mountpoint.
        attachments = volume.get('attachments', {})
        # Multiattach volumes can have more than one attachment, so if there
        # is more than one attachment, attempt to count the read/write
        # attachments.
        if len(attachments) > 1:
            count = 0
            for attachment in attachments.values():
                attachment_id = attachment['attachment_id']
                # Get the attachment record for this attachment so we can
                # get the attach_mode.
                # TODO(mriedem): This could be optimized if we had
                # GET /attachments/detail?volume_id=volume['id'] in Cinder.
                try:
                    attachment_record = self.volume_api.attachment_get(
                        ctxt, attachment_id)
                    # Note that the attachment record from Cinder has
                    # attach_mode in the top-level of the resource but the
                    # nova.volume.cinder code translates it and puts the
                    # attach_mode in the connection_info for some legacy
                    # reason...
                    if attachment_record['attach_mode'] == 'rw':
                        count += 1
                except exception.VolumeAttachmentNotFound:
                    # attachments are read/write by default so count it
                    count += 1
        else:
            count = len(attachments)

        return count

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.PAUSED,
                                    vm_states.RESIZED])
    def swap_volume(self, context, instance, old_volume, new_volume):
        """Swap volume attached to an instance."""
        # The caller likely got the instance from volume['attachments']
        # in the first place, but let's sanity check.
        if not old_volume.get('attachments', {}).get(instance.uuid):
            msg = _("Old volume is attached to a different instance.")
            raise exception.InvalidVolume(reason=msg)
        if new_volume['attach_status'] == 'attached':
            msg = _("New volume must be detached in order to swap.")
            raise exception.InvalidVolume(reason=msg)
        if int(new_volume['size']) < int(old_volume['size']):
            msg = _("New volume must be the same size or larger.")
            raise exception.InvalidVolume(reason=msg)
        self.volume_api.check_availability_zone(context, new_volume,
                                                instance=instance)
        try:
            self.volume_api.begin_detaching(context, old_volume['id'])
        except exception.InvalidInput as exc:
            raise exception.InvalidVolume(reason=exc.format_message())

        # Disallow swapping from multiattach volumes that have more than one
        # read/write attachment. We know the old_volume has at least one
        # attachment since it's attached to this server. The new_volume
        # can't have any attachments because of the attach_status check above.
        # We do this count after calling "begin_detaching" to lock against
        # concurrent attachments being made while we're counting.
        try:
            if self._count_attachments_for_swap(context, old_volume) > 1:
                raise exception.MultiattachSwapVolumeNotSupported()
        except Exception:  # This is generic to handle failures while counting
            # We need to reset the detaching status before raising.
            with excutils.save_and_reraise_exception():
                self.volume_api.roll_detaching(context, old_volume['id'])

        # Get the BDM for the attached (old) volume so we can tell if it was
        # attached with the new-style Cinder 3.44 API.
        bdm = objects.BlockDeviceMapping.get_by_volume_and_instance(
            context, old_volume['id'], instance.uuid)
        new_attachment_id = None
        if bdm.attachment_id is None:
            # This is an old-style attachment so reserve the new volume before
            # we cast to the compute host.
            self.volume_api.reserve_volume(context, new_volume['id'])
        else:
            try:
                self._check_volume_already_attached(
                    context, instance, new_volume)
            except exception.InvalidVolume:
                with excutils.save_and_reraise_exception():
                    self.volume_api.roll_detaching(context, old_volume['id'])

            # This is a new-style attachment so for the volume that we are
            # going to swap to, create a new volume attachment.
            new_attachment_id = self.volume_api.attachment_create(
                context, new_volume['id'], instance.uuid)['id']

        self._record_action_start(
            context, instance, instance_actions.SWAP_VOLUME)

        try:
            self.compute_rpcapi.swap_volume(
                    context, instance=instance,
                    old_volume_id=old_volume['id'],
                    new_volume_id=new_volume['id'],
                    new_attachment_id=new_attachment_id)
        except Exception:
            with excutils.save_and_reraise_exception():
                self.volume_api.roll_detaching(context, old_volume['id'])
                if new_attachment_id is None:
                    self.volume_api.unreserve_volume(context, new_volume['id'])
                else:
                    self.volume_api.attachment_delete(
                        context, new_attachment_id)

    def ensure_compute_version_for_resource_request(
        self, context, instance, port
    ):
        """Checks that the compute service version is new enough for the
        resource request of the port.
       """
        if self.network_api.has_extended_resource_request_extension(
            context
        ):
            # TODO(gibi): Remove this check in Y where we can be sure that
            # the compute is already upgraded to X.
            res_req = port.get(constants.RESOURCE_REQUEST) or {}
            groups = res_req.get('request_groups', [])
            if groups:
                svc = objects.Service.get_by_host_and_binary(
                    context, instance.host, 'nova-compute')
                if svc.version < MIN_COMPUTE_INT_ATTACH_WITH_EXTENDED_RES_REQ:
                    raise exception.ExtendedResourceRequestOldCompute()

        else:
            # NOTE(gibi): Checking if the requested port has resource request
            # as such ports are only supported if the compute service version
            # is >= 55.
            # TODO(gibi): Remove this check in X as there we can be sure
            # that all computes are new enough.
            if port.get(constants.RESOURCE_REQUEST):
                svc = objects.Service.get_by_host_and_binary(
                    context, instance.host, 'nova-compute')
                if svc.version < 55:
                    raise exception.AttachInterfaceWithQoSPolicyNotSupported(
                        instance_uuid=instance.uuid)

    @check_instance_lock
    @reject_vdpa_instances(
        instance_actions.ATTACH_INTERFACE, until=MIN_COMPUTE_VDPA_ATTACH_DETACH
    )
    @check_instance_state(
        vm_state=[
            vm_states.ACTIVE, vm_states.PAUSED, vm_states.STOPPED
        ], task_state=[None]
    )
    def attach_interface(self, context, instance, network_id, port_id,
                         requested_ip, tag=None):
        """Use hotplug to add an network adapter to an instance."""
        self._record_action_start(
            context, instance, instance_actions.ATTACH_INTERFACE)

        if port_id:
            # We need to query the port with admin context as
            # ensure_compute_version_for_resource_request depends on the
            # port.resource_request field which only returned for admins
            port = self.network_api.show_port(
                context.elevated(), port_id)['port']

            if port.get('binding:vnic_type', 'normal') in (
                network_model.VNIC_TYPE_ACCELERATOR_DIRECT,
                network_model.VNIC_TYPE_ACCELERATOR_DIRECT_PHYSICAL):
                raise exception.ForbiddenPortsWithAccelerator()

            if port.get('binding:vnic_type',
                        'normal') == network_model.VNIC_TYPE_REMOTE_MANAGED:
                self._check_vnic_remote_managed_min_version(context)

            self.ensure_compute_version_for_resource_request(
                context, instance, port)

        return self.compute_rpcapi.attach_interface(context,
            instance=instance, network_id=network_id, port_id=port_id,
            requested_ip=requested_ip, tag=tag)

    @check_instance_lock
    @reject_vdpa_instances(
        instance_actions.DETACH_INTERFACE, until=MIN_COMPUTE_VDPA_ATTACH_DETACH
    )
    @check_instance_state(
        vm_state=[
            vm_states.ACTIVE, vm_states.PAUSED, vm_states.STOPPED
        ], task_state=[None]
    )
    def detach_interface(self, context, instance, port_id):
        """Detach an network adapter from an instance."""
        for vif in instance.get_network_info():
            if vif['id'] == port_id:
                if vif['vnic_type'] in (
                    network_model.VNIC_TYPE_ACCELERATOR_DIRECT,
                    network_model.VNIC_TYPE_ACCELERATOR_DIRECT_PHYSICAL):
                    raise exception.ForbiddenPortsWithAccelerator()
                break

        self._record_action_start(
            context, instance, instance_actions.DETACH_INTERFACE)
        self.compute_rpcapi.detach_interface(context, instance=instance,
            port_id=port_id)

    def get_instance_metadata(self, context, instance):
        """Get all metadata associated with an instance."""
        return main_db_api.instance_metadata_get(context, instance.uuid)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.PAUSED,
                                    vm_states.SUSPENDED, vm_states.STOPPED],
                          task_state=None)
    def delete_instance_metadata(self, context, instance, key):
        """Delete the given metadata item from an instance."""
        instance.delete_metadata_key(key)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.PAUSED,
                                    vm_states.SUSPENDED, vm_states.STOPPED],
                          task_state=None)
    def update_instance_metadata(self, context, instance,
                                 metadata, delete=False):
        """Updates or creates instance metadata.

        If delete is True, metadata items that are not specified in the
        `metadata` argument will be deleted.

        """
        if delete:
            _metadata = metadata
        else:
            _metadata = dict(instance.metadata)
            _metadata.update(metadata)

        self._check_metadata_properties_quota(context, _metadata)
        instance.metadata = _metadata
        instance.save()

        return _metadata

    @block_extended_resource_request
    @block_port_accelerators()
    @reject_vdpa_instances(
        instance_actions.LIVE_MIGRATION,
        until=MIN_COMPUTE_VDPA_HOTPLUG_LIVE_MIGRATION
    )
    @block_accelerators()
    @reject_vtpm_instances(instance_actions.LIVE_MIGRATION)
    @reject_sev_instances(instance_actions.LIVE_MIGRATION)
    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.PAUSED])
    def live_migrate(self, context, instance, block_migration,
                     disk_over_commit, host_name, force=None, async_=False):
        """Migrate a server lively to a new host."""
        LOG.debug("Going to try to live migrate instance to %s",
                  host_name or "another host", instance=instance)

        if host_name:
            # Validate the specified host before changing the instance task
            # state.
            nodes = objects.ComputeNodeList.get_all_by_host(context, host_name)

        request_spec = objects.RequestSpec.get_by_instance_uuid(
            context, instance.uuid)

        instance.task_state = task_states.MIGRATING
        instance.save(expected_task_state=[None])

        self._record_action_start(context, instance,
                                  instance_actions.LIVE_MIGRATION)

        # NOTE(sbauza): Force is a boolean by the new related API version
        if force is False and host_name:
            # Unset the host to make sure we call the scheduler
            # from the conductor LiveMigrationTask. Yes this is tightly-coupled
            # to behavior in conductor and not great.
            host_name = None
            # FIXME(sbauza): Since only Ironic driver uses more than one
            # compute per service but doesn't support live migrations,
            # let's provide the first one.
            target = nodes[0]
            destination = objects.Destination(
                host=target.host,
                node=target.hypervisor_hostname
            )
            # This is essentially a hint to the scheduler to only consider
            # the specified host but still run it through the filters.
            request_spec.requested_destination = destination

        try:
            self.compute_task_api.live_migrate_instance(context, instance,
                host_name, block_migration=block_migration,
                disk_over_commit=disk_over_commit,
                request_spec=request_spec, async_=async_)
        except oslo_exceptions.MessagingTimeout as messaging_timeout:
            with excutils.save_and_reraise_exception():
                # NOTE(pkoniszewski): It is possible that MessagingTimeout
                # occurs, but LM will still be in progress, so write
                # instance fault to database
                compute_utils.add_instance_fault_from_exc(context,
                                                          instance,
                                                          messaging_timeout)

    @check_instance_lock
    @check_instance_state(vm_state=[vm_states.ACTIVE],
                          task_state=[task_states.MIGRATING])
    def live_migrate_force_complete(self, context, instance, migration_id):
        """Force live migration to complete.

        :param context: Security context
        :param instance: The instance that is being migrated
        :param migration_id: ID of ongoing migration

        """
        LOG.debug("Going to try to force live migration to complete",
                  instance=instance)

        # NOTE(pkoniszewski): Get migration object to check if there is ongoing
        # live migration for particular instance. Also pass migration id to
        # compute to double check and avoid possible race condition.
        migration = objects.Migration.get_by_id_and_instance(
            context, migration_id, instance.uuid)
        if migration.status != 'running':
            raise exception.InvalidMigrationState(migration_id=migration_id,
                                                  instance_uuid=instance.uuid,
                                                  state=migration.status,
                                                  method='force complete')

        self._record_action_start(
            context, instance, instance_actions.LIVE_MIGRATION_FORCE_COMPLETE)

        self.compute_rpcapi.live_migration_force_complete(
            context, instance, migration)

    @check_instance_lock
    @check_instance_state(task_state=[task_states.MIGRATING])
    def live_migrate_abort(self, context, instance, migration_id,
                           support_abort_in_queue=False):
        """Abort an in-progress live migration.

        :param context: Security context
        :param instance: The instance that is being migrated
        :param migration_id: ID of in-progress live migration
        :param support_abort_in_queue: Flag indicating whether we can support
            abort migrations in "queued" or "preparing" status.

        """
        migration = objects.Migration.get_by_id_and_instance(context,
                    migration_id, instance.uuid)
        LOG.debug("Going to cancel live migration %s",
                  migration.id, instance=instance)

        # If the microversion does not support abort migration in queue,
        # we are only be able to abort migrations with `running` status;
        # if it is supported, we are able to also abort migrations in
        # `queued` and `preparing` status.
        allowed_states = ['running']
        queued_states = ['queued', 'preparing']
        if support_abort_in_queue:
            # The user requested a microversion that supports aborting a queued
            # or preparing live migration. But we need to check that the
            # compute service hosting the instance is new enough to support
            # aborting a queued/preparing live migration, so we check the
            # service version here.
            allowed_states.extend(queued_states)

        if migration.status not in allowed_states:
            raise exception.InvalidMigrationState(migration_id=migration_id,
                    instance_uuid=instance.uuid,
                    state=migration.status,
                    method='abort live migration')
        self._record_action_start(context, instance,
                                  instance_actions.LIVE_MIGRATION_CANCEL)

        self.compute_rpcapi.live_migration_abort(context,
                instance, migration.id)

    @block_extended_resource_request
    @block_port_accelerators()
    @reject_vtpm_instances(instance_actions.EVACUATE)
    @block_accelerators(until_service=SUPPORT_ACCELERATOR_SERVICE_FOR_REBUILD)
    @check_instance_state(vm_state=[vm_states.ACTIVE, vm_states.STOPPED,
                                    vm_states.ERROR], task_state=None)
    def evacuate(self, context, instance, host, on_shared_storage,
                 admin_password=None, force=None):
        """Running evacuate to target host.

        Checking vm compute host state, if the host not in expected_state,
        raising an exception.

        :param instance: The instance to evacuate
        :param host: Target host. if not set, the scheduler will pick up one
        :param on_shared_storage: True if instance files on shared storage
        :param admin_password: password to set on rebuilt instance
        :param force: Force the evacuation to the specific host target

        """
        LOG.debug('vm evacuation scheduled', instance=instance)
        inst_host = instance.host
        service = objects.Service.get_by_compute_host(context, inst_host)
        if self.servicegroup_api.service_is_up(service):
            LOG.error('Instance compute service state on %s '
                      'expected to be down, but it was up.', inst_host)
            raise exception.ComputeServiceInUse(host=inst_host)

        request_spec = objects.RequestSpec.get_by_instance_uuid(
            context, instance.uuid)

        instance.task_state = task_states.REBUILDING
        instance.save(expected_task_state=None)
        self._record_action_start(context, instance, instance_actions.EVACUATE)

        # NOTE(danms): Create this as a tombstone for the source compute
        # to find and cleanup. No need to pass it anywhere else.
        migration = objects.Migration(
            context, source_compute=instance.host, source_node=instance.node,
            instance_uuid=instance.uuid, status='accepted',
            migration_type=fields_obj.MigrationType.EVACUATION)
        if host:
            migration.dest_compute = host
        migration.create()

        compute_utils.notify_about_instance_usage(
            self.notifier, context, instance, "evacuate")
        compute_utils.notify_about_instance_action(
            context, instance, CONF.host,
            action=fields_obj.NotificationAction.EVACUATE,
            source=fields_obj.NotificationSource.API)

        # NOTE(sbauza): Force is a boolean by the new related API version
        # TODO(stephenfin): Any reason we can't use 'not force' here to handle
        # the pre-v2.29 API microversion, which wouldn't set force
        if force is False and host:
            nodes = objects.ComputeNodeList.get_all_by_host(context, host)
            # NOTE(sbauza): Unset the host to make sure we call the scheduler
            host = None
            # FIXME(sbauza): Since only Ironic driver uses more than one
            # compute per service but doesn't support evacuations,
            # let's provide the first one.
            target = nodes[0]
            destination = objects.Destination(
                host=target.host,
                node=target.hypervisor_hostname
            )
            request_spec.requested_destination = destination

        return self.compute_task_api.rebuild_instance(context,
                       instance=instance,
                       new_pass=admin_password,
                       injected_files=None,
                       image_ref=None,
                       orig_image_ref=None,
                       orig_sys_metadata=None,
                       bdms=None,
                       recreate=True,
                       on_shared_storage=on_shared_storage,
                       host=host,
                       request_spec=request_spec,
                       )

    def get_migrations(self, context, filters):
        """Get all migrations for the given filters."""
        load_cells()

        migrations = []
        for cell in CELLS:
            if cell.uuid == objects.CellMapping.CELL0_UUID:
                continue
            with nova_context.target_cell(context, cell) as cctxt:
                migrations.extend(objects.MigrationList.get_by_filters(
                    cctxt, filters).objects)
        return objects.MigrationList(objects=migrations)

    def get_migrations_sorted(self, context, filters, sort_dirs=None,
                              sort_keys=None, limit=None, marker=None):
        """Get all migrations for the given parameters."""
        mig_objs = migration_list.get_migration_objects_sorted(
            context, filters, limit, marker, sort_keys, sort_dirs)
        # Due to cross-cell resize, we could have duplicate migration records
        # while the instance is in VERIFY_RESIZE state in the destination cell
        # but the original migration record still exists in the source cell.
        # Filter out duplicate migration records here based on which record
        # is newer (last updated).

        def _get_newer_obj(obj1, obj2):
            # created_at will always be set.
            created_at1 = obj1.created_at
            created_at2 = obj2.created_at
            # updated_at might be None
            updated_at1 = obj1.updated_at
            updated_at2 = obj2.updated_at
            # If both have updated_at, compare using that field.
            if updated_at1 and updated_at2:
                if updated_at1 > updated_at2:
                    return obj1
                return obj2
            # Compare created_at versus updated_at.
            if updated_at1:
                if updated_at1 > created_at2:
                    return obj1
                return obj2
            if updated_at2:
                if updated_at2 > created_at1:
                    return obj2
                return obj1
            # Compare created_at only.
            if created_at1 > created_at2:
                return obj1
            return obj2

        # TODO(mriedem): This could be easier if we leveraged the "hidden"
        # field on the Migration record and then just did like
        # _get_unique_filter_method in the get_all() method for instances.
        migrations_by_uuid = collections.OrderedDict()  # maintain sort order
        for migration in mig_objs:
            if migration.uuid not in migrations_by_uuid:
                migrations_by_uuid[migration.uuid] = migration
            else:
                # We have a collision, keep the newer record.
                # Note that using updated_at could be wrong if changes-since or
                # changes-before filters are being used but we have the same
                # issue in _get_unique_filter_method for instances.
                doppelganger = migrations_by_uuid[migration.uuid]
                newer = _get_newer_obj(doppelganger, migration)
                migrations_by_uuid[migration.uuid] = newer
        return objects.MigrationList(objects=list(migrations_by_uuid.values()))

    def get_migrations_in_progress_by_instance(self, context, instance_uuid,
                                               migration_type=None):
        """Get all migrations of an instance in progress."""
        return objects.MigrationList.get_in_progress_by_instance(
                context, instance_uuid, migration_type)

    def get_migration_by_id_and_instance(self, context,
                                         migration_id, instance_uuid):
        """Get the migration of an instance by id."""
        return objects.Migration.get_by_id_and_instance(
                context, migration_id, instance_uuid)

    def _get_bdm_by_volume_id(self, context, volume_id, expected_attrs=None):
        """Retrieve a BDM without knowing its cell.

        .. note:: The context will be targeted to the cell in which the
            BDM is found, if any.

        :param context: The API request context.
        :param volume_id: The ID of the volume.
        :param expected_attrs: list of any additional attributes that should
            be joined when the BDM is loaded from the database.
        :raises: nova.exception.VolumeBDMNotFound if not found in any cell
        """
        load_cells()
        for cell in CELLS:
            nova_context.set_target_cell(context, cell)
            try:
                return objects.BlockDeviceMapping.get_by_volume(
                    context, volume_id, expected_attrs=expected_attrs)
            except exception.NotFound:
                continue
        raise exception.VolumeBDMNotFound(volume_id=volume_id)

    def volume_snapshot_create(self, context, volume_id, create_info):
        bdm = self._get_bdm_by_volume_id(
            context, volume_id, expected_attrs=['instance'])

        # We allow creating the snapshot in any vm_state as long as there is
        # no task being performed on the instance and it has a host.
        @check_instance_host()
        @check_instance_state(vm_state=None)
        def do_volume_snapshot_create(self, context, instance):
            self.compute_rpcapi.volume_snapshot_create(context, instance,
                    volume_id, create_info)
            snapshot = {
                'snapshot': {
                    'id': create_info.get('id'),
                    'volumeId': volume_id
                }
            }
            return snapshot

        return do_volume_snapshot_create(self, context, bdm.instance)

    def volume_snapshot_delete(self, context, volume_id, snapshot_id,
                               delete_info):
        bdm = self._get_bdm_by_volume_id(
            context, volume_id, expected_attrs=['instance'])

        @check_instance_host()
        @check_instance_state(vm_state=None)
        def do_volume_snapshot_delete(self, context, instance):
            # FIXME(lyarwood): Avoid bug #1919487 by rejecting the request
            # to delete an intermediary volume snapshot offline as this isn't
            # currently implemented within the libvirt driver and will fail.
            # This should be fixed in a future release but as it is essentially
            # a new feature wouldn't be something we could backport. As such
            # reject the request here so n-api can respond correctly to c-vol.
            if (delete_info.get('merge_target_file') is not None and
                instance.vm_state != vm_states.ACTIVE
            ):
                raise exception.InstanceInvalidState(
                    attr='vm_state',
                    instance_uuid=instance.uuid,
                    state=instance.vm_state,
                    method='volume_snapshot_delete'
                )

            self.compute_rpcapi.volume_snapshot_delete(context, instance,
                    volume_id, snapshot_id, delete_info)

        do_volume_snapshot_delete(self, context, bdm.instance)

    def external_instance_event(self, api_context, instances, events):
        # NOTE(danms): The external API consumer just provides events,
        # but doesn't know where they go. We need to collate lists
        # by the host the affected instance is on and dispatch them
        # according to host
        instances_by_host = collections.defaultdict(list)
        events_by_host = collections.defaultdict(list)
        hosts_by_instance = collections.defaultdict(list)
        cell_contexts_by_host = {}
        for instance in instances:
            # instance._context is used here since it's already targeted to
            # the cell that the instance lives in, and we need to use that
            # cell context to lookup any migrations associated to the instance.
            hosts, cross_cell_move = self._get_relevant_hosts(
                instance._context, instance)
            for host in hosts:
                # NOTE(danms): All instances on a host must have the same
                # mapping, so just use that
                if host not in cell_contexts_by_host:
                    # NOTE(mriedem): If the instance is being migrated across
                    # cells then we have to get the host mapping to determine
                    # which cell a given host is in.
                    if cross_cell_move:
                        hm = objects.HostMapping.get_by_host(api_context, host)
                        ctxt = nova_context.get_admin_context()
                        nova_context.set_target_cell(ctxt, hm.cell_mapping)
                        cell_contexts_by_host[host] = ctxt
                    else:
                        # The instance is not migrating across cells so just
                        # use the cell-targeted context already in the
                        # instance since the host has to be in that same cell.
                        cell_contexts_by_host[host] = instance._context

                instances_by_host[host].append(instance)
                hosts_by_instance[instance.uuid].append(host)

        for event in events:
            if event.name == 'volume-extended':
                # Volume extend is a user-initiated operation starting in the
                # Block Storage service API. We record an instance action so
                # the user can monitor the operation to completion.
                host = hosts_by_instance[event.instance_uuid][0]
                cell_context = cell_contexts_by_host[host]
                objects.InstanceAction.action_start(
                    cell_context, event.instance_uuid,
                    instance_actions.EXTEND_VOLUME, want_result=False)
            elif event.name == 'power-update':
                host = hosts_by_instance[event.instance_uuid][0]
                cell_context = cell_contexts_by_host[host]
                if event.tag == external_event_obj.POWER_ON:
                    inst_action = instance_actions.START
                elif event.tag == external_event_obj.POWER_OFF:
                    inst_action = instance_actions.STOP
                else:
                    LOG.warning("Invalid power state %s. Cannot process "
                                "the event %s. Skipping it.", event.tag,
                                event)
                    continue
                objects.InstanceAction.action_start(
                    cell_context, event.instance_uuid, inst_action,
                    want_result=False)

            for host in hosts_by_instance[event.instance_uuid]:
                events_by_host[host].append(event)

        for host in instances_by_host:
            cell_context = cell_contexts_by_host[host]

            # TODO(salv-orlando): Handle exceptions raised by the rpc api layer
            # in order to ensure that a failure in processing events on a host
            # will not prevent processing events on other hosts
            self.compute_rpcapi.external_instance_event(
                cell_context, instances_by_host[host], events_by_host[host],
                host=host)

    def _get_relevant_hosts(self, context, instance):
        """Get the relevant hosts for an external server event on an instance.

        :param context: nova auth request context targeted at the same cell
            that the instance lives in
        :param instance: Instance object which is the target of an external
            server event
        :returns: 2-item tuple of:
            - set of at least one host (the host where the instance lives); if
              the instance is being migrated the source and dest compute
              hostnames are in the returned set
            - boolean indicating if the instance is being migrated across cells
        """
        hosts = set()
        hosts.add(instance.host)
        cross_cell_move = False
        if instance.migration_context is not None:
            migration_id = instance.migration_context.migration_id
            migration = objects.Migration.get_by_id(context, migration_id)
            cross_cell_move = migration.cross_cell_move
            hosts.add(migration.dest_compute)
            hosts.add(migration.source_compute)
            cells_msg = (
                'across cells' if cross_cell_move else 'within the same cell')
            LOG.debug('Instance %(instance)s is migrating %(cells_msg)s, '
                      'copying events to all relevant hosts: '
                      '%(hosts)s', {'cells_msg': cells_msg,
                                    'instance': instance.uuid,
                                    'hosts': hosts})
        return hosts, cross_cell_move

    def get_instance_host_status(self, instance):
        if instance.host:
            try:
                service = [service for service in instance.services if
                           service.binary == 'nova-compute'][0]
                if service.forced_down:
                    host_status = fields_obj.HostStatus.DOWN
                elif service.disabled:
                    host_status = fields_obj.HostStatus.MAINTENANCE
                else:
                    alive = self.servicegroup_api.service_is_up(service)
                    host_status = ((alive and fields_obj.HostStatus.UP) or
                                   fields_obj.HostStatus.UNKNOWN)
            except IndexError:
                host_status = fields_obj.HostStatus.NONE
        else:
            host_status = fields_obj.HostStatus.NONE
        return host_status

    def get_instances_host_statuses(self, instance_list):
        host_status_dict = dict()
        host_statuses = dict()
        for instance in instance_list:
            if instance.host:
                if instance.host not in host_status_dict:
                    host_status = self.get_instance_host_status(instance)
                    host_status_dict[instance.host] = host_status
                else:
                    host_status = host_status_dict[instance.host]
            else:
                host_status = fields_obj.HostStatus.NONE
            host_statuses[instance.uuid] = host_status
        return host_statuses


def target_host_cell(fn):
    """Target a host-based function to a cell.

    Expects to wrap a function of signature:

       func(self, context, host, ...)
    """

    @functools.wraps(fn)
    def targeted(self, context, host, *args, **kwargs):
        mapping = objects.HostMapping.get_by_host(context, host)
        nova_context.set_target_cell(context, mapping.cell_mapping)
        return fn(self, context, host, *args, **kwargs)
    return targeted


def _get_service_in_cell_by_host(context, host_name):
    # validates the host; ComputeHostNotFound is raised if invalid
    try:
        mapping = objects.HostMapping.get_by_host(context, host_name)
        nova_context.set_target_cell(context, mapping.cell_mapping)
        service = objects.Service.get_by_compute_host(context, host_name)
    except exception.HostMappingNotFound:
        try:
            # NOTE(danms): This targets our cell
            service = _find_service_in_cell(context, service_host=host_name)
        except exception.NotFound:
            raise exception.ComputeHostNotFound(host=host_name)
    return service


def _find_service_in_cell(context, service_id=None, service_host=None):
    """Find a service by id or hostname by searching all cells.

    If one matching service is found, return it. If none or multiple
    are found, raise an exception.

    :param context: A context.RequestContext
    :param service_id: If not none, the DB ID of the service to find
    :param service_host: If not None, the hostname of the service to find
    :returns: An objects.Service
    :raises: ServiceNotUnique if multiple matching IDs are found
    :raises: NotFound if no matches are found
    :raises: NovaException if called with neither search option
    """

    load_cells()
    service = None
    found_in_cell = None

    is_uuid = False
    if service_id is not None:
        is_uuid = uuidutils.is_uuid_like(service_id)
        if is_uuid:
            lookup_fn = lambda c: objects.Service.get_by_uuid(c, service_id)
        else:
            lookup_fn = lambda c: objects.Service.get_by_id(c, service_id)
    elif service_host is not None:
        lookup_fn = lambda c: (
            objects.Service.get_by_compute_host(c, service_host))
    else:
        LOG.exception('_find_service_in_cell called with no search parameters')
        # This is intentionally cryptic so we don't leak implementation details
        # out of the API.
        raise exception.NovaException()

    for cell in CELLS:
        # NOTE(danms): Services can be in cell0, so don't skip it here
        try:
            with nova_context.target_cell(context, cell) as cctxt:
                cell_service = lookup_fn(cctxt)
        except exception.NotFound:
            # NOTE(danms): Keep looking in other cells
            continue
        if service and cell_service:
            raise exception.ServiceNotUnique()
        service = cell_service
        found_in_cell = cell
        if service and is_uuid:
            break

    if service:
        # NOTE(danms): Set the cell on the context so it remains
        # when we return to our caller
        nova_context.set_target_cell(context, found_in_cell)
        return service
    else:
        raise exception.NotFound()


class HostAPI:
    """Sub-set of the Compute Manager API for managing host operations."""

    def __init__(self, rpcapi=None, servicegroup_api=None):
        self.rpcapi = rpcapi or compute_rpcapi.ComputeAPI()
        self.servicegroup_api = servicegroup_api or servicegroup.API()

    def _assert_host_exists(self, context, host_name, must_be_up=False):
        """Raise HostNotFound if compute host doesn't exist."""
        service = objects.Service.get_by_compute_host(context, host_name)
        if not service:
            raise exception.HostNotFound(host=host_name)
        if must_be_up and not self.servicegroup_api.service_is_up(service):
            raise exception.ComputeServiceUnavailable(host=host_name)
        return service['host']

    @wrap_exception()
    @target_host_cell
    def set_host_enabled(self, context, host_name, enabled):
        """Sets the specified host's ability to accept new instances."""
        host_name = self._assert_host_exists(context, host_name)
        payload = {'host_name': host_name, 'enabled': enabled}
        compute_utils.notify_about_host_update(context,
                                               'set_enabled.start',
                                               payload)
        result = self.rpcapi.set_host_enabled(context, enabled=enabled,
                host=host_name)
        compute_utils.notify_about_host_update(context,
                                               'set_enabled.end',
                                               payload)
        return result

    @target_host_cell
    def get_host_uptime(self, context, host_name):
        """Returns the result of calling "uptime" on the target host."""
        host_name = self._assert_host_exists(context, host_name,
                         must_be_up=True)
        return self.rpcapi.get_host_uptime(context, host=host_name)

    @wrap_exception()
    @target_host_cell
    def host_power_action(self, context, host_name, action):
        """Reboots, shuts down or powers up the host."""
        host_name = self._assert_host_exists(context, host_name)
        payload = {'host_name': host_name, 'action': action}
        compute_utils.notify_about_host_update(context,
                                               'power_action.start',
                                               payload)
        result = self.rpcapi.host_power_action(context, action=action,
                host=host_name)
        compute_utils.notify_about_host_update(context,
                                               'power_action.end',
                                               payload)
        return result

    @wrap_exception()
    @target_host_cell
    def set_host_maintenance(self, context, host_name, mode):
        """Start/Stop host maintenance window. On start, it triggers
        guest VMs evacuation.
        """
        host_name = self._assert_host_exists(context, host_name)
        payload = {'host_name': host_name, 'mode': mode}
        compute_utils.notify_about_host_update(context,
                                               'set_maintenance.start',
                                               payload)
        result = self.rpcapi.host_maintenance_mode(context,
                host_param=host_name, mode=mode, host=host_name)
        compute_utils.notify_about_host_update(context,
                                               'set_maintenance.end',
                                               payload)
        return result

    def service_get_all(self, context, filters=None, set_zones=False,
                        all_cells=False, cell_down_support=False):
        """Returns a list of services, optionally filtering the results.

        If specified, 'filters' should be a dictionary containing services
        attributes and matching values.  Ie, to get a list of services for
        the 'compute' topic, use filters={'topic': 'compute'}.

        If all_cells=True, then scan all cells and merge the results.

        If cell_down_support=True then return minimal service records
        for cells that do not respond based on what we have in the
        host mappings. These will have only 'binary' and 'host' set.
        """
        if filters is None:
            filters = {}
        disabled = filters.pop('disabled', None)
        if 'availability_zone' in filters:
            set_zones = True

        # NOTE(danms): Eventually this all_cells nonsense should go away
        # and we should always iterate over the cells. However, certain
        # callers need the legacy behavior for now.
        if all_cells:
            services = []
            service_dict = nova_context.scatter_gather_all_cells(context,
                objects.ServiceList.get_all, disabled, set_zones=set_zones)
            for cell_uuid, service in service_dict.items():
                if not nova_context.is_cell_failure_sentinel(service):
                    services.extend(service)
                elif cell_down_support:
                    unavailable_services = objects.ServiceList()
                    cid = [cm.id for cm in nova_context.CELLS
                           if cm.uuid == cell_uuid]
                    # We know cid[0] is in the list because we are using the
                    # same list that scatter_gather_all_cells used
                    hms = objects.HostMappingList.get_by_cell_id(context,
                                                                 cid[0])
                    for hm in hms:
                        unavailable_services.objects.append(objects.Service(
                            binary='nova-compute', host=hm.host))
                    LOG.warning("Cell %s is not responding and hence only "
                                "partial results are available from this "
                                "cell.", cell_uuid)
                    services.extend(unavailable_services)
                else:
                    LOG.warning("Cell %s is not responding and hence skipped "
                                "from the results.", cell_uuid)
        else:
            services = objects.ServiceList.get_all(context, disabled,
                                                   set_zones=set_zones)
        ret_services = []
        for service in services:
            for key, val in filters.items():
                if service[key] != val:
                    break
            else:
                # All filters matched.
                ret_services.append(service)
        return ret_services

    def service_get_by_id(self, context, service_id):
        """Get service entry for the given service id or uuid."""
        try:
            return _find_service_in_cell(context, service_id=service_id)
        except exception.NotFound:
            raise exception.ServiceNotFound(service_id=service_id)

    @target_host_cell
    def service_get_by_compute_host(self, context, host_name):
        """Get service entry for the given compute hostname."""
        return objects.Service.get_by_compute_host(context, host_name)

    def _update_compute_provider_status(self, context, service):
        """Calls the compute service to sync the COMPUTE_STATUS_DISABLED trait.

        There are two cases where the API will not call the compute service:

        * The compute service is down. In this case the trait is synchronized
          when the compute service is restarted.
        * The compute service is old. In this case the trait is synchronized
          when the compute service is upgraded and restarted.

        :param context: nova auth RequestContext
        :param service: nova.objects.Service object which has been enabled
            or disabled (see ``service_update``).
        """
        # Make sure the service is up so we can make the RPC call.
        if not self.servicegroup_api.service_is_up(service):
            LOG.info('Compute service on host %s is down. The '
                     'COMPUTE_STATUS_DISABLED trait will be synchronized '
                     'when the service is restarted.', service.host)
            return

        # Make sure the compute service is new enough for the trait sync
        # behavior.
        # TODO(mriedem): Remove this compat check in the U release.
        if service.version < MIN_COMPUTE_SYNC_COMPUTE_STATUS_DISABLED:
            LOG.info('Compute service on host %s is too old to sync the '
                     'COMPUTE_STATUS_DISABLED trait in Placement. The '
                     'trait will be synchronized when the service is '
                     'upgraded and restarted.', service.host)
            return

        enabled = not service.disabled
        # Avoid leaking errors out of the API.
        try:
            LOG.debug('Calling the compute service on host %s to sync the '
                      'COMPUTE_STATUS_DISABLED trait.', service.host)
            self.rpcapi.set_host_enabled(context, service.host, enabled)
        except Exception:
            LOG.exception('An error occurred while updating the '
                          'COMPUTE_STATUS_DISABLED trait on compute node '
                          'resource providers managed by host %s. The trait '
                          'will be synchronized automatically by the compute '
                          'service when the update_available_resource '
                          'periodic task runs.', service.host)

    def service_update(self, context, service):
        """Performs the actual service update operation.

        If the "disabled" field is changed, potentially calls the compute
        service to sync the COMPUTE_STATUS_DISABLED trait on the compute node
        resource providers managed by this compute service.

        :param context: nova auth RequestContext
        :param service: nova.objects.Service object with changes already
            set on the object
        """
        # Before persisting changes and resetting the changed fields on the
        # Service object, determine if the disabled field changed.
        update_placement = 'disabled' in service.obj_what_changed()
        # Persist the Service object changes to the database.
        service.save()
        # If the disabled field changed, potentially call the compute service
        # to sync the COMPUTE_STATUS_DISABLED trait.
        if update_placement:
            self._update_compute_provider_status(context, service)
        return service

    @target_host_cell
    def service_update_by_host_and_binary(self, context, host_name, binary,
                                          params_to_update):
        """Enable / Disable a service.

        Determines the cell that the service is in using the HostMapping.

        For compute services, this stops new builds and migrations going to
        the host.

        See also ``service_update``.

        :param context: nova auth RequestContext
        :param host_name: hostname of the service
        :param binary: service binary (really only supports "nova-compute")
        :param params_to_update: dict of changes to make to the Service object
        :raises: HostMappingNotFound if the host is not mapped to a cell
        :raises: HostBinaryNotFound if a services table record is not found
            with the given host_name and binary
        """
        # TODO(mriedem): Service.get_by_args is deprecated; we should use
        # get_by_compute_host here (remember to update the "raises" docstring).
        service = objects.Service.get_by_args(context, host_name, binary)
        service.update(params_to_update)
        return self.service_update(context, service)

    @target_host_cell
    def instance_get_all_by_host(self, context, host_name):
        """Return all instances on the given host."""
        return objects.InstanceList.get_by_host(context, host_name)

    def task_log_get_all(self, context, task_name, period_beginning,
                         period_ending, host=None, state=None):
        """Return the task logs within a given range, optionally
        filtering by host and/or state.
        """
        return main_db_api.task_log_get_all(
            context, task_name, period_beginning, period_ending, host=host,
            state=state)

    def compute_node_get(self, context, compute_id):
        """Return compute node entry for particular integer ID or UUID."""
        load_cells()

        # NOTE(danms): Unfortunately this API exposes database identifiers
        # which means we really can't do something efficient here
        is_uuid = uuidutils.is_uuid_like(compute_id)
        for cell in CELLS:
            if cell.uuid == objects.CellMapping.CELL0_UUID:
                continue
            with nova_context.target_cell(context, cell) as cctxt:
                try:
                    if is_uuid:
                        return objects.ComputeNode.get_by_uuid(cctxt,
                                                               compute_id)
                    return objects.ComputeNode.get_by_id(cctxt,
                                                         int(compute_id))
                except exception.ComputeHostNotFound:
                    # NOTE(danms): Keep looking in other cells
                    continue

        raise exception.ComputeHostNotFound(host=compute_id)

    def compute_node_get_all(self, context, limit=None, marker=None):
        load_cells()

        computes = []
        uuid_marker = marker and uuidutils.is_uuid_like(marker)
        for cell in CELLS:
            if cell.uuid == objects.CellMapping.CELL0_UUID:
                continue
            with nova_context.target_cell(context, cell) as cctxt:

                # If we have a marker and it's a uuid, see if the compute node
                # is in this cell.
                if marker and uuid_marker:
                    try:
                        compute_marker = objects.ComputeNode.get_by_uuid(
                            cctxt, marker)
                        # we found the marker compute node, so use it's id
                        # for the actual marker for paging in this cell's db
                        marker = compute_marker.id
                    except exception.ComputeHostNotFound:
                        # The marker node isn't in this cell so keep looking.
                        continue

                try:
                    cell_computes = objects.ComputeNodeList.get_by_pagination(
                        cctxt, limit=limit, marker=marker)
                except exception.MarkerNotFound:
                    # NOTE(danms): Keep looking through cells
                    continue
                computes.extend(cell_computes)
                # NOTE(danms): We must have found the marker, so continue on
                # without one
                marker = None
                if limit:
                    limit -= len(cell_computes)
                    if limit <= 0:
                        break

        if marker is not None and len(computes) == 0:
            # NOTE(danms): If we did not find the marker in any cell,
            # mimic the db_api behavior here.
            raise exception.MarkerNotFound(marker=marker)

        return objects.ComputeNodeList(objects=computes)

    def compute_node_search_by_hypervisor(self, context, hypervisor_match):
        load_cells()

        computes = []
        for cell in CELLS:
            if cell.uuid == objects.CellMapping.CELL0_UUID:
                continue
            with nova_context.target_cell(context, cell) as cctxt:
                cell_computes = objects.ComputeNodeList.get_by_hypervisor(
                    cctxt, hypervisor_match)
            computes.extend(cell_computes)
        return objects.ComputeNodeList(objects=computes)

    def compute_node_statistics(self, context):
        load_cells()

        cell_stats = []
        for cell in CELLS:
            if cell.uuid == objects.CellMapping.CELL0_UUID:
                continue
            with nova_context.target_cell(context, cell) as cctxt:
                cell_stats.append(main_db_api.compute_node_statistics(cctxt))

        if cell_stats:
            keys = cell_stats[0].keys()
            return {k: sum(stats[k] for stats in cell_stats)
                    for k in keys}
        else:
            return {}


class InstanceActionAPI:
    """Sub-set of the Compute Manager API for managing instance actions."""

    def actions_get(self, context, instance, limit=None, marker=None,
                    filters=None):
        return objects.InstanceActionList.get_by_instance_uuid(
            context, instance.uuid, limit, marker, filters)

    def action_get_by_request_id(self, context, instance, request_id):
        return objects.InstanceAction.get_by_request_id(
            context, instance.uuid, request_id)

    def action_events_get(self, context, instance, action_id):
        return objects.InstanceActionEventList.get_by_action(
            context, action_id)


class AggregateAPI:
    """Sub-set of the Compute Manager API for managing host aggregates."""

    def __init__(self):
        self.compute_rpcapi = compute_rpcapi.ComputeAPI()
        self.query_client = query.SchedulerQueryClient()

    @property
    def placement_client(self):
        return report.report_client_singleton()

    @wrap_exception()
    def create_aggregate(self, context, aggregate_name, availability_zone):
        """Creates the model for the aggregate."""

        aggregate = objects.Aggregate(context=context)
        aggregate.name = aggregate_name
        if availability_zone:
            aggregate.metadata = {'availability_zone': availability_zone}
        aggregate.create()
        self.query_client.update_aggregates(context, [aggregate])
        return aggregate

    def get_aggregate(self, context, aggregate_id):
        """Get an aggregate by id."""
        return objects.Aggregate.get_by_id(context, aggregate_id)

    def get_aggregate_list(self, context):
        """Get all the aggregates."""
        return objects.AggregateList.get_all(context)

    def get_aggregates_by_host(self, context, compute_host):
        """Get all the aggregates where the given host is presented."""
        return objects.AggregateList.get_by_host(context, compute_host)

    @wrap_exception()
    def update_aggregate(self, context, aggregate_id, values):
        """Update the properties of an aggregate."""
        aggregate = objects.Aggregate.get_by_id(context, aggregate_id)
        if 'name' in values:
            aggregate.name = values.pop('name')
            aggregate.save()
        self.is_safe_to_update_az(context, values, aggregate=aggregate,
                                  action_name=AGGREGATE_ACTION_UPDATE,
                                  check_no_instances_in_az=True)
        if values:
            aggregate.update_metadata(values)
            aggregate.updated_at = timeutils.utcnow()
        self.query_client.update_aggregates(context, [aggregate])
        # If updated values include availability_zones, then the cache
        # which stored availability_zones and host need to be reset
        if values.get('availability_zone'):
            availability_zones.reset_cache()
        return aggregate

    @wrap_exception()
    def update_aggregate_metadata(self, context, aggregate_id, metadata):
        """Updates the aggregate metadata."""
        aggregate = objects.Aggregate.get_by_id(context, aggregate_id)
        self.is_safe_to_update_az(context, metadata, aggregate=aggregate,
                                  action_name=AGGREGATE_ACTION_UPDATE_META,
                                  check_no_instances_in_az=True)
        aggregate.update_metadata(metadata)
        self.query_client.update_aggregates(context, [aggregate])
        # If updated metadata include availability_zones, then the cache
        # which stored availability_zones and host need to be reset
        if metadata and metadata.get('availability_zone'):
            availability_zones.reset_cache()
        aggregate.updated_at = timeutils.utcnow()
        return aggregate

    @wrap_exception()
    def delete_aggregate(self, context, aggregate_id):
        """Deletes the aggregate."""
        aggregate_payload = {'aggregate_id': aggregate_id}
        compute_utils.notify_about_aggregate_update(context,
                                                    "delete.start",
                                                    aggregate_payload)
        aggregate = objects.Aggregate.get_by_id(context, aggregate_id)

        compute_utils.notify_about_aggregate_action(
            context=context,
            aggregate=aggregate,
            action=fields_obj.NotificationAction.DELETE,
            phase=fields_obj.NotificationPhase.START)

        if len(aggregate.hosts) > 0:
            msg = _("Host aggregate is not empty")
            raise exception.InvalidAggregateActionDelete(
                aggregate_id=aggregate_id, reason=msg)
        aggregate.destroy()
        self.query_client.delete_aggregate(context, aggregate)
        compute_utils.notify_about_aggregate_update(context,
                                                    "delete.end",
                                                    aggregate_payload)
        compute_utils.notify_about_aggregate_action(
            context=context,
            aggregate=aggregate,
            action=fields_obj.NotificationAction.DELETE,
            phase=fields_obj.NotificationPhase.END)

    def is_safe_to_update_az(self, context, metadata, aggregate,
                             hosts=None,
                             action_name=AGGREGATE_ACTION_ADD,
                             check_no_instances_in_az=False):
        """Determine if updates alter an aggregate's availability zone.

            :param context: local context
            :param metadata: Target metadata for updating aggregate
            :param aggregate: Aggregate to update
            :param hosts: Hosts to check. If None, aggregate.hosts is used
            :type hosts: list
            :param action_name: Calling method for logging purposes
            :param check_no_instances_in_az: if True, it checks
                there is no instances on any hosts of the aggregate

        """
        if 'availability_zone' in metadata:
            if not metadata['availability_zone']:
                msg = _("Aggregate %s does not support empty named "
                        "availability zone") % aggregate.name
                self._raise_invalid_aggregate_exc(action_name, aggregate.id,
                                                  msg)
            _hosts = hosts or aggregate.hosts
            host_aggregates = objects.AggregateList.get_by_metadata_key(
                context, 'availability_zone', hosts=_hosts)
            conflicting_azs = [
                agg.availability_zone for agg in host_aggregates
                if agg.availability_zone != metadata['availability_zone'] and
                agg.id != aggregate.id]
            if conflicting_azs:
                msg = _("One or more hosts already in availability zone(s) "
                        "%s") % conflicting_azs
                self._raise_invalid_aggregate_exc(action_name, aggregate.id,
                                                  msg)
            same_az_name = (aggregate.availability_zone ==
                            metadata['availability_zone'])
            if check_no_instances_in_az and not same_az_name:
                instance_count_by_cell = (
                    nova_context.scatter_gather_skip_cell0(
                        context,
                        objects.InstanceList.get_count_by_hosts,
                        _hosts))
                if any(cnt for cnt in instance_count_by_cell.values()):
                    msg = _("One or more hosts contain instances in this zone")
                    self._raise_invalid_aggregate_exc(
                        action_name, aggregate.id, msg)

    def _raise_invalid_aggregate_exc(self, action_name, aggregate_id, reason):
        if action_name == AGGREGATE_ACTION_ADD:
            raise exception.InvalidAggregateActionAdd(
                aggregate_id=aggregate_id, reason=reason)
        elif action_name == AGGREGATE_ACTION_UPDATE:
            raise exception.InvalidAggregateActionUpdate(
                aggregate_id=aggregate_id, reason=reason)
        elif action_name == AGGREGATE_ACTION_UPDATE_META:
            raise exception.InvalidAggregateActionUpdateMeta(
                aggregate_id=aggregate_id, reason=reason)
        elif action_name == AGGREGATE_ACTION_DELETE:
            raise exception.InvalidAggregateActionDelete(
                aggregate_id=aggregate_id, reason=reason)

        raise exception.NovaException(
            _("Unexpected aggregate action %s") % action_name)

    def _update_az_cache_for_host(self, context, host_name, aggregate_meta):
        # Update the availability_zone cache to avoid getting wrong
        # availability_zone in cache retention time when add/remove
        # host to/from aggregate.
        if aggregate_meta and aggregate_meta.get('availability_zone'):
            availability_zones.update_host_availability_zone_cache(context,
                                                                   host_name)

    @wrap_exception()
    def add_host_to_aggregate(self, context, aggregate_id, host_name):
        """Adds the host to an aggregate."""
        aggregate_payload = {'aggregate_id': aggregate_id,
                             'host_name': host_name}
        compute_utils.notify_about_aggregate_update(context,
                                                    "addhost.start",
                                                    aggregate_payload)

        service = _get_service_in_cell_by_host(context, host_name)
        if service.host != host_name:
            # NOTE(danms): If we found a service but it is not an
            # exact match, we may have a case-insensitive backend
            # database (like mysql) which will end up with us
            # adding the host-aggregate mapping with a
            # non-matching hostname.
            raise exception.ComputeHostNotFound(host=host_name)

        aggregate = objects.Aggregate.get_by_id(context, aggregate_id)

        compute_utils.notify_about_aggregate_action(
            context=context,
            aggregate=aggregate,
            action=fields_obj.NotificationAction.ADD_HOST,
            phase=fields_obj.NotificationPhase.START)

        self.is_safe_to_update_az(context, aggregate.metadata,
                                  hosts=[host_name], aggregate=aggregate)

        aggregate.add_host(host_name)
        self.query_client.update_aggregates(context, [aggregate])
        nodes = objects.ComputeNodeList.get_all_by_host(context, host_name)
        node_name = nodes[0].hypervisor_hostname
        try:
            self.placement_client.aggregate_add_host(
                context, aggregate.uuid, host_name=node_name)
        except (exception.ResourceProviderNotFound,
                exception.ResourceProviderAggregateRetrievalFailed,
                exception.ResourceProviderUpdateFailed,
                exception.ResourceProviderUpdateConflict) as err:
            # NOTE(jaypipes): We don't want a failure perform the mirroring
            # action in the placement service to be returned to the user (they
            # probably don't know anything about the placement service and
            # would just be confused). So, we just log a warning here, noting
            # that on the next run of nova-manage placement sync_aggregates
            # things will go back to normal
            LOG.warning("Failed to associate %s with a placement "
                        "aggregate: %s. This may be corrected after running "
                        "nova-manage placement sync_aggregates.",
                        node_name, err)
        self._update_az_cache_for_host(context, host_name, aggregate.metadata)
        aggregate_payload.update({'name': aggregate.name})
        compute_utils.notify_about_aggregate_update(context,
                                                    "addhost.end",
                                                    aggregate_payload)
        compute_utils.notify_about_aggregate_action(
            context=context,
            aggregate=aggregate,
            action=fields_obj.NotificationAction.ADD_HOST,
            phase=fields_obj.NotificationPhase.END)

        return aggregate

    @wrap_exception()
    def remove_host_from_aggregate(self, context, aggregate_id, host_name):
        """Removes host from the aggregate."""
        aggregate_payload = {'aggregate_id': aggregate_id,
                             'host_name': host_name}
        compute_utils.notify_about_aggregate_update(context,
                                                    "removehost.start",
                                                    aggregate_payload)
        _get_service_in_cell_by_host(context, host_name)
        aggregate = objects.Aggregate.get_by_id(context, aggregate_id)

        compute_utils.notify_about_aggregate_action(
            context=context,
            aggregate=aggregate,
            action=fields_obj.NotificationAction.REMOVE_HOST,
            phase=fields_obj.NotificationPhase.START)

        # Remove the resource provider from the provider aggregate first before
        # we change anything on the nova side because if we did the nova stuff
        # first we can't re-attempt this from the compute API if cleaning up
        # placement fails.
        nodes = objects.ComputeNodeList.get_all_by_host(context, host_name)
        node_name = nodes[0].hypervisor_hostname
        try:
            # Anything else this raises is handled in the route handler as
            # either a 409 (ResourceProviderUpdateConflict) or 500.
            self.placement_client.aggregate_remove_host(
                context, aggregate.uuid, node_name)
        except exception.ResourceProviderNotFound as err:
            # If the resource provider is not found then it's likely not part
            # of the aggregate anymore anyway since provider aggregates are
            # not resources themselves with metadata like nova aggregates, they
            # are just a grouping concept around resource providers. Log and
            # continue.
            LOG.warning("Failed to remove association of %s with a placement "
                        "aggregate: %s.", node_name, err)

        aggregate.delete_host(host_name)
        self.query_client.update_aggregates(context, [aggregate])
        self._update_az_cache_for_host(context, host_name, aggregate.metadata)
        compute_utils.notify_about_aggregate_update(context,
                                                    "removehost.end",
                                                    aggregate_payload)
        compute_utils.notify_about_aggregate_action(
            context=context,
            aggregate=aggregate,
            action=fields_obj.NotificationAction.REMOVE_HOST,
            phase=fields_obj.NotificationPhase.END)
        return aggregate


class KeypairAPI:
    """Subset of the Compute Manager API for managing key pairs."""

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

    def __init__(self):
        self.notifier = rpc.get_notifier('api')

    def _notify(self, context, event_suffix, keypair_name):
        payload = {
            'tenant_id': context.project_id,
            'user_id': context.user_id,
            'key_name': keypair_name,
        }
        self.notifier.info(context, 'keypair.%s' % event_suffix, payload)

    def _check_key_pair_quotas(self, context, user_id, key_name, key_type):
        try:
            objects.Quotas.check_deltas(context, {'key_pairs': 1}, user_id)
            local_limit.enforce_db_limit(context, local_limit.KEY_PAIRS,
                                         entity_scope=user_id, delta=1)
        except exception.KeypairLimitExceeded:
            raise
        except exception.OverQuota:
            raise exception.KeypairLimitExceeded()

    @wrap_exception()
    def import_key_pair(self, context, user_id, key_name, public_key,
                        key_type=keypair_obj.KEYPAIR_TYPE_SSH):
        """Import a key pair using an existing public key."""
        self._check_key_pair_quotas(context, user_id, key_name, key_type)

        self._notify(context, 'import.start', key_name)

        keypair = objects.KeyPair(context)
        keypair.user_id = user_id
        keypair.name = key_name
        keypair.type = key_type
        keypair.fingerprint = None
        keypair.public_key = public_key

        compute_utils.notify_about_keypair_action(
            context=context,
            keypair=keypair,
            action=fields_obj.NotificationAction.IMPORT,
            phase=fields_obj.NotificationPhase.START)

        fingerprint = self._generate_fingerprint(public_key, key_type)

        keypair.fingerprint = fingerprint
        keypair.create()

        compute_utils.notify_about_keypair_action(
            context=context,
            keypair=keypair,
            action=fields_obj.NotificationAction.IMPORT,
            phase=fields_obj.NotificationPhase.END)
        self._notify(context, 'import.end', key_name)

        return keypair

    @wrap_exception()
    def create_key_pair(self, context, user_id, key_name,
                        key_type=keypair_obj.KEYPAIR_TYPE_SSH):
        """Create a new key pair."""
        self._check_key_pair_quotas(context, user_id, key_name, key_type)

        keypair = objects.KeyPair(context)
        keypair.user_id = user_id
        keypair.name = key_name
        keypair.type = key_type
        keypair.fingerprint = None
        keypair.public_key = None

        self._notify(context, 'create.start', key_name)
        compute_utils.notify_about_keypair_action(
            context=context,
            keypair=keypair,
            action=fields_obj.NotificationAction.CREATE,
            phase=fields_obj.NotificationPhase.START)

        private_key, public_key, fingerprint = self._generate_key_pair(
            user_id, key_type)

        keypair.fingerprint = fingerprint
        keypair.public_key = public_key
        keypair.create()

        # NOTE(melwitt): We recheck the quota after creating the object to
        # prevent users from allocating more resources than their allowed quota
        # in the event of a race. This is configurable because it can be
        # expensive if strict quota limits are not required in a deployment.
        if CONF.quota.recheck_quota:
            try:
                objects.Quotas.check_deltas(context, {'key_pairs': 0}, user_id)
                # TODO(johngarbutt) do we really need this recheck?
                # The quota rechecking of limits is really just to protect
                # against denial of service attacks that aim to fill up the
                # database. Its usefulness could be debated.
                local_limit.enforce_db_limit(context, local_limit.KEY_PAIRS,
                                             entity_scope=user_id, delta=0)
            except exception.KeypairLimitExceeded:
                with excutils.save_and_reraise_exception():
                    keypair.destroy()
            except exception.OverQuota:
                keypair.destroy()
                raise exception.KeypairLimitExceeded()

        compute_utils.notify_about_keypair_action(
            context=context,
            keypair=keypair,
            action=fields_obj.NotificationAction.CREATE,
            phase=fields_obj.NotificationPhase.END)

        self._notify(context, 'create.end', key_name)

        return keypair, private_key

    def _generate_fingerprint(self, public_key, key_type):
        if key_type == keypair_obj.KEYPAIR_TYPE_SSH:
            return crypto.generate_fingerprint(public_key)
        elif key_type == keypair_obj.KEYPAIR_TYPE_X509:
            return crypto.generate_x509_fingerprint(public_key)

    def _generate_key_pair(self, user_id, key_type):
        if key_type == keypair_obj.KEYPAIR_TYPE_SSH:
            return crypto.generate_key_pair()
        elif key_type == keypair_obj.KEYPAIR_TYPE_X509:
            return crypto.generate_winrm_x509_cert(user_id)

    @wrap_exception()
    def delete_key_pair(self, context, user_id, key_name):
        """Delete a keypair by name."""
        self._notify(context, 'delete.start', key_name)
        keypair = self.get_key_pair(context, user_id, key_name)
        compute_utils.notify_about_keypair_action(
            context=context,
            keypair=keypair,
            action=fields_obj.NotificationAction.DELETE,
            phase=fields_obj.NotificationPhase.START)
        objects.KeyPair.destroy_by_name(context, user_id, key_name)
        compute_utils.notify_about_keypair_action(
            context=context,
            keypair=keypair,
            action=fields_obj.NotificationAction.DELETE,
            phase=fields_obj.NotificationPhase.END)
        self._notify(context, 'delete.end', key_name)

    def get_key_pairs(self, context, user_id, limit=None, marker=None):
        """List key pairs."""
        return objects.KeyPairList.get_by_user(
            context, user_id, limit=limit, marker=marker)

    def get_key_pair(self, context, user_id, key_name):
        """Get a keypair by name."""
        return objects.KeyPair.get_by_name(context, user_id, key_name)