# # Licensed under the Apache License, Version 2.0 (the "License"); you may # not use this file except in compliance with the License. You may obtain # a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the # License for the specific language governing permissions and limitations # under the License. import copy from oslo_config import cfg from oslo_log import log as logging import six from heat.common import exception from heat.common.i18n import _ from heat.common.i18n import _LI from heat.engine import attributes from heat.engine import constraints from heat.engine import properties from heat.engine import resource from heat.engine import scheduler from heat.engine import volume_tasks as vol_task cfg.CONF.import_opt('instance_user', 'heat.common.config') cfg.CONF.import_opt('stack_scheduler_hints', 'heat.common.config') LOG = logging.getLogger(__name__) class Instance(resource.Resource): PROPERTIES = ( IMAGE_ID, INSTANCE_TYPE, KEY_NAME, AVAILABILITY_ZONE, DISABLE_API_TERMINATION, KERNEL_ID, MONITORING, PLACEMENT_GROUP_NAME, PRIVATE_IP_ADDRESS, RAM_DISK_ID, SECURITY_GROUPS, SECURITY_GROUP_IDS, NETWORK_INTERFACES, SOURCE_DEST_CHECK, SUBNET_ID, TAGS, NOVA_SCHEDULER_HINTS, TENANCY, USER_DATA, VOLUMES, BLOCK_DEVICE_MAPPINGS ) = ( 'ImageId', 'InstanceType', 'KeyName', 'AvailabilityZone', 'DisableApiTermination', 'KernelId', 'Monitoring', 'PlacementGroupName', 'PrivateIpAddress', 'RamDiskId', 'SecurityGroups', 'SecurityGroupIds', 'NetworkInterfaces', 'SourceDestCheck', 'SubnetId', 'Tags', 'NovaSchedulerHints', 'Tenancy', 'UserData', 'Volumes', 'BlockDeviceMappings' ) _TAG_KEYS = ( TAG_KEY, TAG_VALUE, ) = ( 'Key', 'Value', ) _NOVA_SCHEDULER_HINT_KEYS = ( NOVA_SCHEDULER_HINT_KEY, NOVA_SCHEDULER_HINT_VALUE, ) = ( 'Key', 'Value', ) _VOLUME_KEYS = ( VOLUME_DEVICE, VOLUME_ID, ) = ( 'Device', 'VolumeId', ) _BLOCK_DEVICE_MAPPINGS_KEYS = ( DEVICE_NAME, EBS, NO_DEVICE, VIRTUAL_NAME, ) = ( 'DeviceName', 'Ebs', 'NoDevice', 'VirtualName', ) _EBS_KEYS = ( DELETE_ON_TERMINATION, IOPS, SNAPSHOT_ID, VOLUME_SIZE, VOLUME_TYPE, ) = ( 'DeleteOnTermination', 'Iops', 'SnapshotId', 'VolumeSize', 'VolumeType' ) ATTRIBUTES = ( AVAILABILITY_ZONE_ATTR, PRIVATE_DNS_NAME, PUBLIC_DNS_NAME, PRIVATE_IP, PUBLIC_IP, ) = ( 'AvailabilityZone', 'PrivateDnsName', 'PublicDnsName', 'PrivateIp', 'PublicIp', ) properties_schema = { IMAGE_ID: properties.Schema( properties.Schema.STRING, _('Glance image ID or name.'), constraints=[ constraints.CustomConstraint('glance.image') ], required=True ), # AWS does not require InstanceType but Heat does because the nova # create api call requires a flavor INSTANCE_TYPE: properties.Schema( properties.Schema.STRING, _('Nova instance type (flavor).'), required=True, update_allowed=True, constraints=[ constraints.CustomConstraint('nova.flavor') ] ), KEY_NAME: properties.Schema( properties.Schema.STRING, _('Optional Nova keypair name.'), constraints=[ constraints.CustomConstraint("nova.keypair") ] ), AVAILABILITY_ZONE: properties.Schema( properties.Schema.STRING, _('Availability zone to launch the instance in.') ), DISABLE_API_TERMINATION: properties.Schema( properties.Schema.STRING, _('Not Implemented.'), implemented=False ), KERNEL_ID: properties.Schema( properties.Schema.STRING, _('Not Implemented.'), implemented=False ), MONITORING: properties.Schema( properties.Schema.BOOLEAN, _('Not Implemented.'), implemented=False ), PLACEMENT_GROUP_NAME: properties.Schema( properties.Schema.STRING, _('Not Implemented.'), implemented=False ), PRIVATE_IP_ADDRESS: properties.Schema( properties.Schema.STRING, _('Not Implemented.'), implemented=False ), RAM_DISK_ID: properties.Schema( properties.Schema.STRING, _('Not Implemented.'), implemented=False ), SECURITY_GROUPS: properties.Schema( properties.Schema.LIST, _('Security group names to assign.') ), SECURITY_GROUP_IDS: properties.Schema( properties.Schema.LIST, _('Security group IDs to assign.') ), NETWORK_INTERFACES: properties.Schema( properties.Schema.LIST, _('Network interfaces to associate with instance.'), update_allowed=True ), SOURCE_DEST_CHECK: properties.Schema( properties.Schema.BOOLEAN, _('Not Implemented.'), implemented=False ), SUBNET_ID: properties.Schema( properties.Schema.STRING, _('Subnet ID to launch instance in.'), update_allowed=True ), TAGS: properties.Schema( properties.Schema.LIST, _('Tags to attach to instance.'), schema=properties.Schema( properties.Schema.MAP, schema={ TAG_KEY: properties.Schema( properties.Schema.STRING, required=True ), TAG_VALUE: properties.Schema( properties.Schema.STRING, required=True ), }, ), update_allowed=True ), NOVA_SCHEDULER_HINTS: properties.Schema( properties.Schema.LIST, _('Scheduler hints to pass to Nova (Heat extension).'), schema=properties.Schema( properties.Schema.MAP, schema={ NOVA_SCHEDULER_HINT_KEY: properties.Schema( properties.Schema.STRING, required=True ), NOVA_SCHEDULER_HINT_VALUE: properties.Schema( properties.Schema.STRING, required=True ), }, ) ), TENANCY: properties.Schema( properties.Schema.STRING, _('Not Implemented.'), constraints=[ constraints.AllowedValues(['dedicated', 'default']), ], implemented=False ), USER_DATA: properties.Schema( properties.Schema.STRING, _('User data to pass to instance.') ), VOLUMES: properties.Schema( properties.Schema.LIST, _('Volumes to attach to instance.'), default=[], schema=properties.Schema( properties.Schema.MAP, schema={ VOLUME_DEVICE: properties.Schema( properties.Schema.STRING, _('The device where the volume is exposed on the ' 'instance. This assignment may not be honored and ' 'it is advised that the path ' '/dev/disk/by-id/virtio- be used ' 'instead.'), required=True ), VOLUME_ID: properties.Schema( properties.Schema.STRING, _('The ID of the volume to be attached.'), required=True, constraints=[ constraints.CustomConstraint('cinder.volume') ] ), } ) ), BLOCK_DEVICE_MAPPINGS: properties.Schema( properties.Schema.LIST, _('Block device mappings to attach to instance.'), schema=properties.Schema( properties.Schema.MAP, schema={ DEVICE_NAME: properties.Schema( properties.Schema.STRING, _('A device name where the volume will be ' 'attached in the system at /dev/device_name.' 'e.g. vdb'), required=True, ), EBS: properties.Schema( properties.Schema.MAP, _('The ebs volume to attach to the instance.'), schema={ DELETE_ON_TERMINATION: properties.Schema( properties.Schema.BOOLEAN, _('Indicate whether the volume should be ' 'deleted when the instance is terminated.'), default=True ), IOPS: properties.Schema( properties.Schema.NUMBER, _('The number of I/O operations per second ' 'that the volume supports.'), implemented=False ), SNAPSHOT_ID: properties.Schema( properties.Schema.STRING, _('The ID of the snapshot to create ' 'a volume from.'), constraints=[ constraints.CustomConstraint( 'cinder.snapshot') ] ), VOLUME_SIZE: properties.Schema( properties.Schema.STRING, _('The size of the volume, in GB. Must be ' 'equal or greater than the size of the ' 'snapshot. It is safe to leave this blank ' 'and have the Compute service infer ' 'the size.'), ), VOLUME_TYPE: properties.Schema( properties.Schema.STRING, _('The volume type.'), implemented=False ), }, ), NO_DEVICE: properties.Schema( properties.Schema.MAP, _('The can be used to unmap a defined device.'), implemented=False ), VIRTUAL_NAME: properties.Schema( properties.Schema.STRING, _('The name of the virtual device. The name must be ' 'in the form ephemeralX where X is a number ' 'starting from zero (0); for example, ephemeral0.'), implemented=False ), }, ), ), } attributes_schema = { AVAILABILITY_ZONE_ATTR: attributes.Schema( _('The Availability Zone where the specified instance is ' 'launched.') ), PRIVATE_DNS_NAME: attributes.Schema( _('Private DNS name of the specified instance.') ), PUBLIC_DNS_NAME: attributes.Schema( _('Public DNS name of the specified instance.') ), PRIVATE_IP: attributes.Schema( _('Private IP address of the specified instance.') ), PUBLIC_IP: attributes.Schema( _('Public IP address of the specified instance.') ), } # Server host name limit to 53 characters by due to typical default # linux HOST_NAME_MAX of 64, minus the .novalocal appended to the name physical_resource_name_limit = 53 default_client_name = 'nova' def __init__(self, name, json_snippet, stack): super(Instance, self).__init__(name, json_snippet, stack) self.ipaddress = None def _set_ipaddress(self, networks): ''' Read the server's IP address from a list of networks provided by Nova ''' # Just record the first ipaddress for n in networks: if len(networks[n]) > 0: self.ipaddress = networks[n][0] break def _ipaddress(self): ''' Return the server's IP address, fetching it from Nova if necessary ''' if self.ipaddress is None: self.ipaddress = self.client_plugin().server_to_ipaddress( self.resource_id) return self.ipaddress or '0.0.0.0' def _availability_zone(self): ''' Return Server's Availability Zone, fetching it from Nova if necessary. ''' availability_zone = self.properties[self.AVAILABILITY_ZONE] if availability_zone is None: try: server = self.nova().servers.get(self.resource_id) availability_zone = getattr(server, 'OS-EXT-AZ:availability_zone') except Exception as e: self.client_plugin().ignore_not_found(e) return return availability_zone def _resolve_attribute(self, name): res = None if name == self.AVAILABILITY_ZONE_ATTR: res = self._availability_zone() elif name in self.ATTRIBUTES[1:]: res = self._ipaddress() LOG.info(_LI('%(name)s._resolve_attribute(%(attname)s) == %(res)s'), {'name': self.name, 'attname': name, 'res': res}) return six.text_type(res) if res else None def _port_data_delete(self): # delete the port data which implicit-created port_id = self.data().get('port_id') if port_id: try: self.neutron().delete_port(port_id) except Exception as ex: self.client_plugin('neutron').ignore_not_found(ex) self.data_delete('port_id') def _build_nics(self, network_interfaces, security_groups=None, subnet_id=None): nics = None if network_interfaces: unsorted_nics = [] for entry in network_interfaces: nic = (entry if not isinstance(entry, six.string_types) else {'NetworkInterfaceId': entry, 'DeviceIndex': len(unsorted_nics)}) unsorted_nics.append(nic) sorted_nics = sorted(unsorted_nics, key=lambda nic: int(nic['DeviceIndex'])) nics = [{'port-id': snic['NetworkInterfaceId']} for snic in sorted_nics] else: # if SubnetId property in Instance, ensure subnet exists if subnet_id: neutronclient = self.neutron() network_id = self.client_plugin( 'neutron').network_id_from_subnet_id(subnet_id) # if subnet verified, create a port to use this subnet # if port is not created explicitly, nova will choose # the first subnet in the given network. if network_id: fixed_ip = {'subnet_id': subnet_id} props = { 'admin_state_up': True, 'network_id': network_id, 'fixed_ips': [fixed_ip] } if security_groups: props['security_groups'] = self.client_plugin( 'neutron').get_secgroup_uuids(security_groups) port = neutronclient.create_port({'port': props})['port'] # after create the port, set the port-id to # resource data, so that the port can be deleted on # instance delete. self.data_set('port_id', port['id']) nics = [{'port-id': port['id']}] return nics def _get_security_groups(self): security_groups = [] for key in (self.SECURITY_GROUPS, self.SECURITY_GROUP_IDS): if self.properties.get(key) is not None: for sg in self.properties.get(key): security_groups.append(sg) if not security_groups: security_groups = None return security_groups def _build_block_device_mapping(self, bdm): if not bdm: return None bdm_dict = {} for mapping in bdm: device_name = mapping.get(self.DEVICE_NAME) ebs = mapping.get(self.EBS) if ebs: mapping_parts = [] snapshot_id = ebs.get(self.SNAPSHOT_ID) volume_size = ebs.get(self.VOLUME_SIZE) delete = ebs.get(self.DELETE_ON_TERMINATION) if snapshot_id: mapping_parts.append(snapshot_id) mapping_parts.append('snap') if volume_size: mapping_parts.append(str(volume_size)) else: mapping_parts.append('') if delete is not None: mapping_parts.append(str(delete)) bdm_dict[device_name] = ':'.join(mapping_parts) return bdm_dict def _get_nova_metadata(self, properties): if properties is None or properties.get(self.TAGS) is None: return None return dict((tm[self.TAG_KEY], tm[self.TAG_VALUE]) for tm in properties[self.TAGS]) def handle_create(self): security_groups = self._get_security_groups() userdata = self.properties[self.USER_DATA] or '' flavor = self.properties[self.INSTANCE_TYPE] availability_zone = self.properties[self.AVAILABILITY_ZONE] image_name = self.properties[self.IMAGE_ID] image_id = self.client_plugin('glance').get_image_id(image_name) flavor_id = self.client_plugin().get_flavor_id(flavor) scheduler_hints = {} if self.properties[self.NOVA_SCHEDULER_HINTS]: for tm in self.properties[self.NOVA_SCHEDULER_HINTS]: # adopted from novaclient shell hint = tm[self.NOVA_SCHEDULER_HINT_KEY] hint_value = tm[self.NOVA_SCHEDULER_HINT_VALUE] if hint in scheduler_hints: if isinstance(scheduler_hints[hint], six.string_types): scheduler_hints[hint] = [scheduler_hints[hint]] scheduler_hints[hint].append(hint_value) else: scheduler_hints[hint] = hint_value else: scheduler_hints = None if cfg.CONF.stack_scheduler_hints: if scheduler_hints is None: scheduler_hints = {} scheduler_hints['heat_root_stack_id'] = self.stack.root_stack.id scheduler_hints['heat_stack_id'] = self.stack.id scheduler_hints['heat_stack_name'] = self.stack.name scheduler_hints['heat_path_in_stack'] = self.stack.path_in_stack() scheduler_hints['heat_resource_name'] = self.name nics = self._build_nics(self.properties[self.NETWORK_INTERFACES], security_groups=security_groups, subnet_id=self.properties[self.SUBNET_ID]) block_device_mapping = self._build_block_device_mapping( self.properties.get(self.BLOCK_DEVICE_MAPPINGS)) server = None # FIXME(shadower): the instance_user config option is deprecated. Once # it's gone, we should always use ec2-user for compatibility with # CloudFormation. if cfg.CONF.instance_user: instance_user = cfg.CONF.instance_user else: instance_user = 'ec2-user' try: server = self.nova().servers.create( name=self.physical_resource_name(), image=image_id, flavor=flavor_id, key_name=self.properties[self.KEY_NAME], security_groups=security_groups, userdata=self.client_plugin().build_userdata( self.metadata_get(), userdata, instance_user), meta=self._get_nova_metadata(self.properties), scheduler_hints=scheduler_hints, nics=nics, availability_zone=availability_zone, block_device_mapping=block_device_mapping) finally: # Avoid a race condition where the thread could be cancelled # before the ID is stored if server is not None: self.resource_id_set(server.id) return server, scheduler.TaskRunner(self._attach_volumes_task()) def _attach_volumes_task(self): attach_tasks = (vol_task.VolumeAttachTask(self.stack, self.resource_id, volume_id, device) for volume_id, device in self.volumes()) return scheduler.PollingTaskGroup(attach_tasks) def check_create_complete(self, cookie): server, volume_attach_task = cookie return (self._check_active(server) and self._check_volume_attached(server, volume_attach_task)) def _check_volume_attached(self, server, volume_attach_task): if not volume_attach_task.started(): self._set_ipaddress(server.networks) volume_attach_task.start() return volume_attach_task.done() else: return volume_attach_task.step() def _check_active(self, server): cp = self.client_plugin() status = cp.get_status(server) if status != 'ACTIVE': cp.refresh_server(server) status = cp.get_status(server) if status == 'ACTIVE': return True if status in cp.deferred_server_statuses: return False if status == 'ERROR': fault = getattr(server, 'fault', {}) raise resource.ResourceInError( resource_status=status, status_reason=_("Message: %(message)s, Code: %(code)s") % { 'message': fault.get('message', _('Unknown')), 'code': fault.get('code', _('Unknown')) }) raise resource.ResourceUnknownStatus( resource_status=server.status, result=_('Instance is not active')) def volumes(self): """ Return an iterator over (volume_id, device) tuples for all volumes that should be attached to this instance. """ volumes = self.properties[self.VOLUMES] return ((vol[self.VOLUME_ID], vol[self.VOLUME_DEVICE]) for vol in volumes) def _remove_matched_ifaces(self, old_network_ifaces, new_network_ifaces): # find matches and remove them from old and new ifaces old_network_ifaces_copy = copy.deepcopy(old_network_ifaces) for iface in old_network_ifaces_copy: if iface in new_network_ifaces: new_network_ifaces.remove(iface) old_network_ifaces.remove(iface) def handle_check(self): server = self.nova().servers.get(self.resource_id) if not self._check_active(server): raise exception.Error(_("Instance is not ACTIVE (was: %s)") % server.status.strip()) def handle_update(self, json_snippet, tmpl_diff, prop_diff): if 'Metadata' in tmpl_diff: self.metadata_set(tmpl_diff['Metadata']) checkers = [] server = None if self.TAGS in prop_diff: server = self.nova().servers.get(self.resource_id) self.client_plugin().meta_update( server, self._get_nova_metadata(prop_diff)) if self.INSTANCE_TYPE in prop_diff: flavor = prop_diff[self.INSTANCE_TYPE] flavor_id = self.client_plugin().get_flavor_id(flavor) if not server: server = self.nova().servers.get(self.resource_id) checker = scheduler.TaskRunner(self.client_plugin().resize, server, flavor, flavor_id) checkers.append(checker) if self.NETWORK_INTERFACES in prop_diff: new_network_ifaces = prop_diff.get(self.NETWORK_INTERFACES) old_network_ifaces = self.properties.get(self.NETWORK_INTERFACES) subnet_id = ( prop_diff.get(self.SUBNET_ID) or self.properties.get(self.SUBNET_ID)) security_groups = self._get_security_groups() if not server: server = self.nova().servers.get(self.resource_id) # if there is entrys in old_network_ifaces and new_network_ifaces, # remove the same entrys from old and new ifaces if old_network_ifaces and new_network_ifaces: # there are four situations: # 1.old includes new, such as: old = 2,3, new = 2 # 2.new includes old, such as: old = 2,3, new = 1,2,3 # 3.has overlaps, such as: old = 2,3, new = 1,2 # 4.different, such as: old = 2,3, new = 1,4 # detach unmatched ones in old, attach unmatched ones in new self._remove_matched_ifaces(old_network_ifaces, new_network_ifaces) if old_network_ifaces: old_nics = self._build_nics(old_network_ifaces) for nic in old_nics: checker = scheduler.TaskRunner( server.interface_detach, nic['port-id']) checkers.append(checker) if new_network_ifaces: new_nics = self._build_nics(new_network_ifaces) for nic in new_nics: checker = scheduler.TaskRunner( server.interface_attach, nic['port-id'], None, None) checkers.append(checker) # if the interfaces not come from property 'NetworkInterfaces', # the situation is somewhat complex, so to detach the old ifaces, # and then attach the new ones. else: interfaces = server.interface_list() for iface in interfaces: checker = scheduler.TaskRunner(server.interface_detach, iface.port_id) checkers.append(checker) # first to delete the port which implicit-created by heat self._port_data_delete() nics = self._build_nics(new_network_ifaces, security_groups=security_groups, subnet_id=subnet_id) # 'SubnetId' property is empty(or None) and # 'NetworkInterfaces' property is empty(or None), # _build_nics() will return nics = None,we should attach # first free port, according to similar behavior during # instance creation if not nics: checker = scheduler.TaskRunner(server.interface_attach, None, None, None) checkers.append(checker) else: for nic in nics: checker = scheduler.TaskRunner( server.interface_attach, nic['port-id'], None, None) checkers.append(checker) if checkers: checkers[0].start() return checkers def check_update_complete(self, checkers): '''Push all checkers to completion in list order.''' for checker in checkers: if not checker.started(): checker.start() if not checker.step(): return False return True def metadata_update(self, new_metadata=None): ''' Refresh the metadata if new_metadata is None ''' if new_metadata is None: self.metadata_set(self.t.metadata()) def validate(self): ''' Validate any of the provided params ''' res = super(Instance, self).validate() if res: return res # check validity of security groups vs. network interfaces security_groups = self._get_security_groups() if security_groups and self.properties.get(self.NETWORK_INTERFACES): raise exception.ResourcePropertyConflict( '/'.join([self.SECURITY_GROUPS, self.SECURITY_GROUP_IDS]), self.NETWORK_INTERFACES) # check bdm property # now we don't support without snapshot_id in bdm bdm = self.properties.get(self.BLOCK_DEVICE_MAPPINGS) if bdm: for mapping in bdm: ebs = mapping.get(self.EBS) if ebs: snapshot_id = ebs.get(self.SNAPSHOT_ID) if not snapshot_id: msg = _("SnapshotId is missing, this is required " "when specifying BlockDeviceMappings.") raise exception.StackValidationFailed(message=msg) else: msg = _("Ebs is missing, this is required " "when specifying BlockDeviceMappings.") raise exception.StackValidationFailed(message=msg) def _detach_volumes_task(self): ''' Detach volumes from the instance ''' detach_tasks = (vol_task.VolumeDetachTask(self.stack, self.resource_id, volume_id) for volume_id, device in self.volumes()) return scheduler.PollingTaskGroup(detach_tasks) def handle_delete(self): # make sure to delete the port which implicit-created by heat self._port_data_delete() if self.resource_id is None: return try: server = self.nova().servers.get(self.resource_id) except Exception as e: self.client_plugin().ignore_not_found(e) return deleters = ( scheduler.TaskRunner(self._detach_volumes_task()), scheduler.TaskRunner(self.client_plugin().delete_server, server)) deleters[0].start() return deleters def check_delete_complete(self, deleters): # if the resource was already deleted, deleters will be None if deleters: for deleter in deleters: if not deleter.started(): deleter.start() if not deleter.step(): return False return True def handle_suspend(self): ''' Suspend an instance - note we do not wait for the SUSPENDED state, this is polled for by check_suspend_complete in a similar way to the create logic so we can take advantage of coroutines ''' if self.resource_id is None: raise exception.Error(_('Cannot suspend %s, resource_id not set') % self.name) try: server = self.nova().servers.get(self.resource_id) except Exception as e: if self.client_plugin().is_not_found(e): raise exception.NotFound(_('Failed to find instance %s') % self.resource_id) else: LOG.debug("suspending instance %s" % self.resource_id) # We want the server.suspend to happen after the volume # detachement has finished, so pass both tasks and the server suspend_runner = scheduler.TaskRunner(server.suspend) volumes_runner = scheduler.TaskRunner(self._detach_volumes_task()) return server, suspend_runner, volumes_runner def check_suspend_complete(self, cookie): server, suspend_runner, volumes_runner = cookie if not volumes_runner.started(): volumes_runner.start() if volumes_runner.done(): if not suspend_runner.started(): suspend_runner.start() if suspend_runner.done(): if server.status == 'SUSPENDED': return True cp = self.client_plugin() cp.refresh_server(server) LOG.debug("%(name)s check_suspend_complete " "status = %(status)s", {'name': self.name, 'status': server.status}) if server.status in list(cp.deferred_server_statuses + ['ACTIVE']): return server.status == 'SUSPENDED' else: raise exception.Error(_(' nova reported unexpected ' 'instance[%(instance)s] ' 'status[%(status)s]') % {'instance': self.name, 'status': server.status}) else: suspend_runner.step() else: volumes_runner.step() def handle_resume(self): ''' Resume an instance - note we do not wait for the ACTIVE state, this is polled for by check_resume_complete in a similar way to the create logic so we can take advantage of coroutines ''' if self.resource_id is None: raise exception.Error(_('Cannot resume %s, resource_id not set') % self.name) try: server = self.nova().servers.get(self.resource_id) except Exception as e: if self.client_plugin().is_not_found(e): raise exception.NotFound(_('Failed to find instance %s') % self.resource_id) else: LOG.debug("resuming instance %s" % self.resource_id) server.resume() return server, scheduler.TaskRunner(self._attach_volumes_task()) def check_resume_complete(self, cookie): server, volume_attach_task = cookie return (self._check_active(server) and self._check_volume_attached(server, volume_attach_task)) def resource_mapping(): return { 'AWS::EC2::Instance': Instance, }