starlingx/nfv
Code Issues Proposed changes

Fix guest heartbeat post notification not sent for resize-revert

Browse Source 
Issue is Post Notification is not sent after resize-revert if
resize-revert is moving the instance to a new host.
Introduced max_resize_wait_in_secs and handling of the potential host
change event.

Also fixed a few VIM scenario tests failures due to
nfv_vim_instance_guest_services_post_notify_heartbeat not found.

Change-Id: I89064a6856bc90d18e5d0f51f9eb3a41a951b1a8
This commit is contained in:
Jack Ding 2018-04-02 11:10:20 -04:00 committed by Al Bailey
parent da93ce42c7
commit 1370e9d948
5 changed files with 94 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
nfv/nfv-vim/nfv_vim/config.ini
View File

@ -38,6 +38,7 @@ max_live_migrate_wait_in_secs=800
max_live_migrate_wait_in_secs_min=120
max_live_migrate_wait_in_secs_max=800
max_cold_migrate_wait_in_secs=900
max_resize_wait_in_secs=900
max_evacuate_wait_in_secs=900
max_cold_migrate_local_image_disk_gb=60
max_cold_migrate_local_lvm_disk_gb=20

5
nfv/nfv-vim/nfv_vim/host_fsm/_host_task_work.py
View File

@ -1021,9 +1021,12 @@ class NotifyInstancesHostDisablingTaskWork(state_machine.StateTaskWork):
int(section.get('max_live_migrate_wait_in_secs_max', 800))
max_cold_migrate_wait_in_secs = \
int(section.get('max_cold_migrate_wait_in_secs', 900))
max_resize_wait_in_secs = \
int(section.get('max_resize_wait_in_secs', 900))
max_migrate_wait_in_secs = max(
max_live_migrate_wait_in_secs_max,
max_cold_migrate_wait_in_secs)
max_cold_migrate_wait_in_secs,
max_resize_wait_in_secs)
else:
max_migrate_wait_in_secs = 900
# Add 60s to ensure the migration will time out before task

1
nfv/nfv-vim/nfv_vim/instance_fsm/_instance_defs.py
View File

@ -61,6 +61,7 @@ class EventNames(object):
GUEST_ACTION_PROCEED = Constant('guest-action-proceed')
GUEST_COMMUNICATION_ESTABLISHED = Constant('guest-communication-established')
LIVE_MIGRATE_ROLLBACK = Constant('live-migrate-rollback')
RESIZE_REVERT_COMPLETED = Constant('resize-revert-completed')
AUDIT = Constant('audit')
TASK_START = Constant('task-start')
TASK_STOP = Constant('task-stop')

46
nfv/nfv-vim/nfv_vim/instance_fsm/_instance_task_work.py
View File

@ -556,6 +556,7 @@ class ResizeRevertTaskWork(state_machine.StateTaskWork):
'resize-revert-instance_%s' % instance.name, task,
force_pass=force_pass, timeout_in_secs=60)
self._instance_reference = weakref.ref(instance)
self._from_host_name = instance.host_name
@property
def _instance(self):
@ -575,9 +576,18 @@ class ResizeRevertTaskWork(state_machine.StateTaskWork):
DLOG.debug("Resize-Revert-Instance callback for %s, "
"response=%s." % (self._instance.name, response))
if response['completed']:
self.task.task_work_complete(
state_machine.STATE_TASK_WORK_RESULT.SUCCESS,
empty_reason)
# A resize revert might causes a movement of the instance back
# to the original host. Need to wait for this movement to
# complete.
if 0 == self._instance.max_resize_wait_in_secs:
DLOG.verbose("Resize-Revert-Instance instance has a "
"timeout of zero, not waiting.")
self.task.task_work_complete(
state_machine.STATE_TASK_WORK_RESULT.SUCCESS,
empty_reason)
else:
self.extend_timeout(
self._instance.max_resize_wait_in_secs)
else:
if self.force_pass:
DLOG.info("Resize-Revert-Instance callback for %s, "
@ -612,6 +622,32 @@ class ResizeRevertTaskWork(state_machine.StateTaskWork):
return state_machine.STATE_TASK_WORK_RESULT.WAIT, empty_reason
def handle_event(self, event, event_data=None):
"""
Handle instance action proceed notifications
"""
handled = False
if INSTANCE_EVENT.NFVI_HOST_CHANGED == event:
if self._from_host_name != self._instance.host_name:
DLOG.debug("Resize-Revert-Instance for %s has moved from "
"host %s to host %s." % (self._instance.name,
self._from_host_name,
self._instance.host_name))
self.task.task_work_complete(
state_machine.STATE_TASK_WORK_RESULT.SUCCESS,
empty_reason)
handled = True
elif INSTANCE_EVENT.RESIZE_REVERT_COMPLETED == event:
DLOG.debug("Resize-Revert-Instance for %s completed"
% self._instance.name)
self.task.task_work_complete(
state_machine.STATE_TASK_WORK_RESULT.SUCCESS,
empty_reason)
handled = True
return handled
class EvacuateTaskWork(state_machine.StateTaskWork):
"""
@ -2225,6 +2261,10 @@ class GuestServicesPostNotifyTaskWork(state_machine.StateTaskWork):
self._action_type))
if guest_services.guest_communication_established():
# this log is needed for nfv_scenario_tests
DLOG.debug("Guest-Services-Post-Notify for %s, guest "
"communication re-established." % self._instance.name)
DLOG.debug("Guest-Services-Post-Notify for %s, action_type=%s."
% (self._instance.name, self._action_type))

51
nfv/nfv-vim/nfv_vim/objects/_instance.py
View File

@ -812,6 +812,7 @@ class Instance(ObjectData):
self._max_live_migrate_wait_in_secs = None
self._max_live_migration_downtime_in_ms = None
self._max_cold_migrate_wait_in_secs = None
self._max_resize_wait_in_secs = None
self._max_evacuate_wait_in_secs = None
self._deleted = False
self._fail_reason = None
@ -1401,6 +1402,27 @@ class Instance(ObjectData):
% (self._max_cold_migrate_wait_in_secs, self.name))
return self._max_cold_migrate_wait_in_secs
@property
def max_resize_wait_in_secs(self):
"""
Returns the resize timeout value for this instance
"""
if self._max_resize_wait_in_secs is not None:
DLOG.debug("Resize timeout is %s secs for %s."
% (self._max_resize_wait_in_secs, self.name))
return self._max_resize_wait_in_secs
if config.section_exists('instance-configuration'):
section = config.CONF['instance-configuration']
self._max_resize_wait_in_secs = \
int(section.get('max_resize_wait_in_secs', 900))
else:
self._max_resize_wait_in_secs = 900
DLOG.debug("Resize timeout set to %s secs for %s."
% (self._max_resize_wait_in_secs, self.name))
return self._max_resize_wait_in_secs
@property
def max_evacuate_wait_in_secs(self):
"""
@ -2786,6 +2808,16 @@ class Instance(ObjectData):
# There is not an action in progress, mark action as completed.
self._action_data.set_action_completed()
elif self._action_data.is_completed():
if INSTANCE_ACTION_TYPE.REVERT_RESIZE == action_type:
DLOG.debug("Resize-Revert-Instance for instance %s completed."
% self.name)
self._action_fsm.handle_event(
instance_fsm.INSTANCE_EVENT.RESIZE_REVERT_COMPLETED)
else:
DLOG.info("Ignoring action for instance %s, action_type=%s, "
"action-state %s." % (self.name, action_type,
action_state))
else:
DLOG.info("Ignoring action for instance %s, action_type=%s, "
"action-state %s." % (self.name, action_type,
@ -2804,13 +2836,25 @@ class Instance(ObjectData):
if not self._action_data.is_inprogress():
return
if INSTANCE_ACTION_TYPE.LIVE_MIGRATE_ROLLBACK == nfvi_action_type:
if nfvi.objects.v1.INSTANCE_ACTION_TYPE.LIVE_MIGRATE_ROLLBACK \
== nfvi_action_type:
self._action_data.nfvi_action_data_change(nfvi_action_type,
nfvi_action_state, reason)
self._persist()
self._nfvi_instance_handle_action_change()
return
elif nfvi.objects.v1.INSTANCE_ACTION_TYPE.RESIZE == nfvi_action_type:
if INSTANCE_ACTION_TYPE.REVERT_RESIZE \
== self._action_data.action_type:
nfvi_action_type \
= nfvi.objects.v1.INSTANCE_ACTION_TYPE.REVERT_RESIZE
self._action_data.nfvi_action_data_change(
nfvi_action_type, nfvi_action_state, reason)
self._persist()
self._nfvi_instance_handle_action_change()
return
if not self.guest_services.are_provisioned():
return
@ -2835,11 +2879,6 @@ class Instance(ObjectData):
nfvi_action_type \
= nfvi.objects.v1.INSTANCE_ACTION_TYPE.CONFIRM_RESIZE
elif INSTANCE_ACTION_TYPE.REVERT_RESIZE \
== self._action_data.action_type:
nfvi_action_type \
= nfvi.objects.v1.INSTANCE_ACTION_TYPE.REVERT_RESIZE
self._action_data.nfvi_action_data_change(nfvi_action_type,
nfvi_action_state, reason)
self._persist()