初始化host :
步骤：
- 加载whitelist （不知道这个是干什么用的）
- 检测 host placement(如果placement API不可用，则应用抛出异常。
- host compute driver层的初始化
- 通过InstanceList 类获取该host上的示例列表instaces
- 检测是否延迟IP绑定，如果时，通过driver配置
- init_virt_events ，不知道具体干什么的
- 调用_init_instance函数，初始化检测host上的每个instance.
def init_host(self):
"""Initialization for a standalone compute service."""
if CONF.pci.passthrough_whitelist:
# Simply loading the PCI passthrough whitelist will do a bunch of
# validation that would otherwise wait until the PciDevTracker is
# constructed when updating available resources for the compute
# node(s) in the resource tracker, effectively killing that task.
# So load up the whitelist when starting the compute service to
# flush any invalid configuration early so we can kill the service
# if the configuration is wrong.
whitelist.Whitelist(CONF.pci.passthrough_whitelist)
# NOTE(sbauza): We want the compute node to hard fail if it can't be
# able to provide its resources to the placement API, or it would not
# be able to be eligible as a destination.
if CONF.placement.os_region_name is None:
raise exception.PlacementNotConfigured()
self.driver.init_host(host=self.host)
context = nova.context.get_admin_context()
instances = objects.InstanceList.get_by_host(
context, self.host, expected_attrs=['info_cache', 'metadata'])
if CONF.defer_iptables_apply:
self.driver.filter_defer_apply_on()
self.init_virt_events()
try:
# checking that instance was not already evacuated to other host
self._destroy_evacuated_instances(context)
for instance in instances:
self._init_instance(context, instance)
finally:
if CONF.defer_iptables_apply:
self.driver.filter_defer_apply_off()
self._update_scheduler_instance_info(context, instances)
_init_instance 函数执行步骤：
1. 如果该instance 被转移了，则记录log后直接返回，否则继续执行；
2. 如果该instance 被关闭了或是处于错误状态（可能是由于在迁移中或是删除中），则记录log后直接返回，否则继续执行；
3. 如果instance处于DELETED状态中，则
def _init_instance(self, context, instance):
'''Initialize this instance during service init.'''
if instance.host != self.host:
LOG.warning(_LW('Instance %(uuid)s appears to not be owned '
'by this host, but by %(host)s. Startup '
'processing is being skipped.'),
{'uuid': instance.uuid,
'host': instance.host})
return
if (instance.vm_state == vm_states.SOFT_DELETED or
(instance.vm_state == vm_states.ERROR and
instance.task_state not in
(task_states.RESIZE_MIGRATING, task_states.DELETING))):
LOG.debug("Instance is in %s state.",
instance.vm_state, instance=instance)
return
if instance.vm_state == vm_states.DELETED:
try:
self._complete_partial_deletion(context, instance)
except Exception:
# we don't want that an exception blocks the init_host
msg = _LE('Failed to complete a deletion')
LOG.exception(msg, instance=instance)
return
if (instance.vm_state == vm_states.BUILDING or
instance.task_state in [task_states.SCHEDULING,
task_states.BLOCK_DEVICE_MAPPING,
task_states.NETWORKING,
task_states.SPAWNING]):
# NOTE(dave-mcnally) compute stopped before instance was fully
# spawned so set to ERROR state. This is safe to do as the state
# may be set by the api but the host is not so if we get here the
# instance has already been scheduled to this particular host.
LOG.debug("Instance failed to spawn correctly, "
"setting to ERROR state", instance=instance)
instance.task_state = None
instance.vm_state = vm_states.ERROR
instance.save()
return
if (instance.vm_state in [vm_states.ACTIVE, vm_states.STOPPED] and
instance.task_state in [task_states.REBUILDING,
task_states.REBUILD_BLOCK_DEVICE_MAPPING,
task_states.REBUILD_SPAWNING]):
# NOTE(jichenjc) compute stopped before instance was fully
# spawned so set to ERROR state. This is consistent to BUILD
LOG.debug("Instance failed to rebuild correctly, "
"setting to ERROR state", instance=instance)
instance.task_state = None
instance.vm_state = vm_states.ERROR
instance.save()
return
if (instance.vm_state != vm_states.ERROR and
instance.task_state in [task_states.IMAGE_SNAPSHOT_PENDING,
task_states.IMAGE_PENDING_UPLOAD,
task_states.IMAGE_UPLOADING,
task_states.IMAGE_SNAPSHOT]):
LOG.debug("Instance in transitional state %s at start-up "
"clearing task state",
instance.task_state, instance=instance)
try:
self._post_interrupted_snapshot_cleanup(context, instance)
except Exception:
# we don't want that an exception blocks the init_host
msg = _LE('Failed to cleanup snapshot.')
LOG.exception(msg, instance=instance)
instance.task_state = None
instance.save()
if (instance.vm_state != vm_states.ERROR and
instance.task_state in [task_states.RESIZE_PREP]):
LOG.debug("Instance in transitional state %s at start-up "
"clearing task state",
instance['task_state'], instance=instance)
instance.task_state = None
instance.save()
if instance.task_state == task_states.DELETING:
try:
LOG.info(_LI('Service started deleting the instance during '
'the previous run, but did not finish. Restarting'
' the deletion now.'), instance=instance)
instance.obj_load_attr('metadata')
instance.obj_load_attr('system_metadata')
bdms = objects.BlockDeviceMappingList.get_by_instance_uuid(
context, instance.uuid)
project_id, user_id = objects.quotas.ids_from_instance(
context, instance)
quotas = self._create_reservations(context, instance,
project_id, user_id)
self._delete_instance(context, instance, bdms, quotas)
except Exception:
# we don't want that an exception blocks the init_host
msg = _LE('Failed to complete a deletion')
LOG.exception(msg, instance=instance)
self._set_instance_obj_error_state(context, instance)
return
current_power_state = self._get_power_state(context, instance)
try_reboot, reboot_type = self._retry_reboot(context, instance,
current_power_state)
if try_reboot:
LOG.debug("Instance in transitional state (%(task_state)s) at "
"start-up and power state is (%(power_state)s), "
"triggering reboot",
{'task_state': instance.task_state,
'power_state': current_power_state},
instance=instance)
# NOTE(mikal): if the instance was doing a soft reboot that got as
# far as shutting down the instance but not as far as starting it
# again, then we've just become a hard reboot. That means the
# task state for the instance needs to change so that we're in one
# of the expected task states for a hard reboot.
soft_types = [task_states.REBOOT_STARTED,
task_states.REBOOT_PENDING,
task_states.REBOOTING]
if instance.task_state in soft_types and reboot_type == 'HARD':
instance.task_state = task_states.REBOOT_PENDING_HARD
instance.save()
self.reboot_instance(context, instance, block_device_info=None,
reboot_type=reboot_type)
return
elif (current_power_state == power_state.RUNNING and
instance.task_state in [task_states.REBOOT_STARTED,
task_states.REBOOT_STARTED_HARD,
task_states.PAUSING,
task_states.UNPAUSING]):
LOG.warning(_LW("Instance in transitional state "
"(%(task_state)s) at start-up and power state "
"is (%(power_state)s), clearing task state"),
{'task_state': instance.task_state,
'power_state': current_power_state},
instance=instance)
instance.task_state = None
instance.vm_state = vm_states.ACTIVE
instance.save()
elif (current_power_state == power_state.PAUSED and
instance.task_state == task_states.UNPAUSING):
LOG.warning(_LW("Instance in transitional state "
"(%(task_state)s) at start-up and power state "
"is (%(power_state)s), clearing task state "
"and unpausing the instance"),
{'task_state': instance.task_state,
'power_state': current_power_state},
instance=instance)
try:
self.unpause_instance(context, instance)
except NotImplementedError:
# Some virt driver didn't support pause and unpause
pass
except Exception:
LOG.exception(_LE('Failed to unpause instance'),
instance=instance)
return
if instance.task_state == task_states.POWERING_OFF:
try:
LOG.debug("Instance in transitional state %s at start-up "
"retrying stop request",
instance.task_state, instance=instance)
self.stop_instance(context, instance, True)
except Exception:
# we don't want that an exception blocks the init_host
msg = _LE('Failed to stop instance')
LOG.exception(msg, instance=instance)
return
if instance.task_state == task_states.POWERING_ON:
try:
LOG.debug("Instance in transitional state %s at start-up "
"retrying start request",
instance.task_state, instance=instance)
self.start_instance(context, instance)
except Exception:
# we don't want that an exception blocks the init_host
msg = _LE('Failed to start instance')
LOG.exception(msg, instance=instance)
return
net_info = compute_utils.get_nw_info_for_instance(instance)
try:
self.driver.plug_vifs(instance, net_info)
except NotImplementedError as e:
LOG.debug(e, instance=instance)
except exception.VirtualInterfacePlugException:
# we don't want an exception to block the init_host
LOG.exception(_LE("Vifs plug failed"), instance=instance)
self._set_instance_obj_error_state(context, instance)
return
if instance.task_state == task_states.RESIZE_MIGRATING:
# We crashed during resize/migration, so roll back for safety
try:
# NOTE(mriedem): check old_vm_state for STOPPED here, if it's
# not in system_metadata we default to True for backwards
# compatibility
power_on = (instance.system_metadata.get('old_vm_state') !=
vm_states.STOPPED)
block_dev_info = self._get_instance_block_device_info(context,
instance)
self.driver.finish_revert_migration(context,
instance, net_info, block_dev_info, power_on)
except Exception:
LOG.exception(_LE('Failed to revert crashed migration'),
instance=instance)
finally:
LOG.info(_LI('Instance found in migrating state during '
'startup. Resetting task_state'),
instance=instance)
instance.task_state = None
instance.save()
if instance.task_state == task_states.MIGRATING:
# Live migration did not complete, but instance is on this
# host, so reset the state.
instance.task_state = None
instance.save(expected_task_state=[task_states.MIGRATING])
db_state = instance.power_state
drv_state = self._get_power_state(context, instance)
expect_running = (db_state == power_state.RUNNING and
drv_state != db_state)
LOG.debug('Current state is %(drv_state)s, state in DB is '
'%(db_state)s.',
{'drv_state': drv_state, 'db_state': db_state},
instance=instance)
if expect_running and CONF.resume_guests_state_on_host_boot:
LOG.info(_LI('Rebooting instance after nova-compute restart.'),
instance=instance)
block_device_info = \
self._get_instance_block_device_info(context, instance)
try:
self.driver.resume_state_on_host_boot(
context, instance, net_info, block_device_info)
except NotImplementedError:
LOG.warning(_LW('Hypervisor driver does not support '
'resume guests'), instance=instance)
except Exception:
# NOTE(vish): The instance failed to resume, so we set the
# instance to error and attempt to continue.
LOG.warning(_LW('Failed to resume instance'),
instance=instance)
self._set_instance_obj_error_state(context, instance)
elif drv_state == power_state.RUNNING:
# VMwareAPI drivers will raise an exception
try:
self.driver.ensure_filtering_rules_for_instance(
instance, net_info)
except NotImplementedError:
LOG.debug('Hypervisor driver does not support '
'firewall rules', instance=instance)