openstack
/
nova
Code Issues Proposed changes
OpenStack Compute (Nova)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
58895 Commits
9 Branches
36 Tags
580 MiB
 
 
 
 
 
Branch: master
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from 'master'
${ noResults }
nova/nova/compute/api.py

6531 lines
302 KiB
Raw Permalink Blame History

# 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 string
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 import api as db
from nova.db.sqlalchemy import api as db_api
from nova import exception
from nova import exception_wrapper
from nova.i18n import _
from nova.image import glance
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.pci import request as pci_request
from nova.policies import servers as servers_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
# 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.
host_status = self.get_instance_host_status(instance)
if host_status not in (fields_obj.HostStatus.UP,
fields_obj.HostStatus.MAINTENANCE):
# 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):
"""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()
):
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
@profiler.trace_cls("compute_api")
class API:
"""API for interacting with the compute manager."""
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._placementclient = None # Lazy-load on first access.
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 QuotaError 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))
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)
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)
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)
API._validate_flavor_image_mem_encryption(flavor, image_meta)
# validate PMU extra spec and image metadata
flavor_pmu = flavor.extra_specs.get('hw:pmu')
image_pmu = image_meta.properties.get('hw_pmu')
if (flavor_pmu is not None and image_pmu is not None and
image_pmu != strutils.bool_from_string(flavor_pmu)):
raise exception.ImagePMUConflict()
# Only validate values of flavor/image so the return results of
# following 'get' functions are not used.
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)
if validate_numa:
hardware.numa_get_constraints(flavor, image_meta)
if validate_pci:
pci_request.get_pci_requests_from_flavor(flavor)
@staticmethod
def _validate_flavor_image_mem_encryption(flavor, image):
"""Validate that the flavor and image don't make contradictory
requests regarding memory encryption.
:param flavor: Flavor object
:param image: an ImageMeta object
:raises: nova.exception.FlavorImageConflict
"""
# This library function will raise the exception for us if
# necessary; if not, we can ignore the result returned.
hardware.get_mem_encryption_constraint(flavor, image)
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 _validate_and_build_base_options(
self, context, flavor, boot_meta, image_href, image_id, kernel_id,
ramdisk_id, display_name, display_description, 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 = result
# 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()
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,
'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}
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
@db_api.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)
security_groups = security_group_api.populate_security_groups(
security_groups)
port_resource_requests = base_options.pop('port_resource_requests')
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)
if block_device_mapping:
# Record whether or not we are a BFV instance
root = block_device_mapping.root_bdm()
req_spec.is_bfv = bool(root and root.is_volume)
else:
# If we have no BDMs, we're clearly not BFV
req_spec.is_bfv = False
# 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)
except exception.OverQuota:
msg = _("Quota exceeded, too many servers in "
"group")
raise exception.QuotaError(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)
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.QuotaError(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 _create_instance(self, context, flavor,
image_href, kernel_id, ramdisk_id,
min_count, max_count,
display_name, display_description,
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,
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
block_device_mapping = self._check_and_transform_bdm(context,
base_options, flavor, boot_meta, min_count, max_count,
block_device_mapping, legacy_bdm)
# 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
# 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.
if num_instances == 1:
default_hostname = 'Server-%s' % instance.uuid
instance.hostname = utils.sanitize_hostname(
instance.display_name, default_hostname)
elif num_instances > 1:
old_display_name = instance.display_name
new_display_name = '%s-%d' % (old_display_name, index + 1)
if utils.sanitize_hostname(old_display_name) == "":
instance.hostname = 'Server-%s' % instance.uuid
else:
instance.hostname = utils.sanitize_hostname(
new_display_name)
instance.display_name = new_display_name
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,
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 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,
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)
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)
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.
# 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
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):
if self._placementclient is None:
self._placementclient = report.SchedulerReportClient()
return self._placementclient
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)
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