live migration did not send notification for other componments

[Guangya Liu]

Currently, if live migration failed, nova compute did not send notification, this caused other service such as nova-scheduler has no chance to do some operation based on the result of live migration.

It is better that we add notification logic for live_migration logic like other operations such as resize, run instance etc.

A proposed code diff would be as following, please feel free to show your comments if any. Thanks.

After some discussion with Joe, we work out a draft code diff for compute/manager.py. I did not get a chance to do some test, just append the logic here. Please show your comments if any.

root@liugya-ubuntu:~/src/nova-es/nova/nova/compute# git diff
diff --git a/nova/compute/manager.py b/nova/compute/manager.py
index 687594a..b8b99b6 100755
--- a/nova/compute/manager.py
+++ b/nova/compute/manager.py
@@ -3127,6 +3127,10 @@ class ComputeManager(manager.SchedulerDependentManager):
         return self.driver.check_can_live_migrate_source(ctxt, instance,
                                                          dest_check_data)

+ @exception.wrap_exception(notifier=notifier, publisher_id=publisher_id())
+ @reverts_task_state
+ @wrap_instance_event
+ @wrap_instance_fault
     def pre_live_migration(self, context, instance,
                            block_migration=False, disk=None,
                            migrate_data=None):
@@ -3146,6 +3150,10 @@ class ComputeManager(manager.SchedulerDependentManager):

         network_info = self._get_instance_nw_info(context, instance)

+ self._notify_about_instance_usage(
+ context, instance, "pre.live_migration.start",
+ network_info=network_info)
+
         # TODO(tr3buchet): figure out how on the earth this is necessary
         fixed_ips = network_info.fixed_ips()
         if not fixed_ips:
@@ -3173,7 +3181,14 @@ class ComputeManager(manager.SchedulerDependentManager):
         # Preparation for block migration
         if block_migration:
             self.driver.pre_block_migration(context, instance, disk)
+ self._notify_about_instance_usage(
+ context, instance, "pre.live_migration.end",
+ network_info=network_info)

+ @exception.wrap_exception(notifier=notifier, publisher_id=publisher_id())
+ @reverts_task_state
+ @wrap_instance_event
+ @wrap_instance_fault
     def live_migration(self, context, dest, instance,
                        block_migration=False, migrate_data=None):
         """Executing live migration.
@@ -3286,6 +3301,10 @@ class ComputeManager(manager.SchedulerDependentManager):
                    "This error can be safely ignored."),
                  instance=instance_ref)

+ @exception.wrap_exception(notifier=notifier, publisher_id=publisher_id())
+ @reverts_task_state
+ @wrap_instance_event
+ @wrap_instance_fault
     def post_live_migration_at_destination(self, context, instance,
                                            block_migration=False):
         """Post operations for live migration .
@@ -3311,6 +3330,9 @@ class ComputeManager(manager.SchedulerDependentManager):
                                                            migration)

         network_info = self._get_instance_nw_info(context, instance)
+ self._notify_about_instance_usage(
+ context, instance, "post.live_migration.start",
+ network_info=network_info)
         block_device_info = self._get_instance_volume_block_device_info(
                             context, instance)

@@ -3326,6 +3348,9 @@ class ComputeManager(manager.SchedulerDependentManager):

         # NOTE(vish): this is necessary to update dhcp
         self.network_api.setup_networks_on_host(context, instance, self.host)
+ self._notify_about_instance_usage(
+ context, instance, "post.live_migration.end",
+ network_info=network_info)

     def _rollback_live_migration(self, context, instance,
                                  dest, block_migration, migrate_data=None):
@@ -3369,6 +3394,10 @@ class ComputeManager(manager.SchedulerDependentManager):
             self.compute_rpcapi.rollback_live_migration_at_destination(context,
                     instance, dest)

+ @exception.wrap_exception(notifier=notifier, publisher_id=publisher_id())
+ @reverts_task_state
+ @wrap_instance_event
+ @wrap_instance_fault
     def rollback_live_migration_at_destination(self, context, instance):
         """Cleaning up image directory that is created pre_live_migration.

@@ -3377,6 +3406,9 @@ class ComputeManager(manager.SchedulerDependentManager):
         """
         network_info = self._get_instance_nw_info(context, instance)

+ self._notify_about_instance_usage(
+ context, instance, "rollback.live_migration.start",
+ network_info=network_info)
         # NOTE(tr3buchet): tear down networks on destination host
         self.network_api.setup_networks_on_host(context, instance,
                                                 self.host, teardown=True)
@@ -3387,6 +3419,9 @@ class ComputeManager(manager.SchedulerDependentManager):
                             context, instance)
         self.driver.destroy(instance, self._legacy_nw_info(network_info),
                             block_device_info)
+ self._notify_about_instance_usage(
+ context, instance, "rollback.live_migration.end",
+ network_info=network_info)

     @manager.periodic_task
     def _heal_instance_info_cache(self, context)

[John Garbutt (johngarbutt)]

Please submit patch into Gerrit for review:
https://wiki.openstack.org/wiki/Gerrit_Workflow

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/28871

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/28871
Committed: http://github.com/openstack/nova/commit/fe8bddc0f25c78af3cd98ba06b0768f2a99b9093
Submitter: Jenkins
Branch: master

commit fe8bddc0f25c78af3cd98ba06b0768f2a99b9093
Author: Jay Lau <email address hidden>
Date: Sat May 11 21:53:20 2013 +0800

    Add notification for live migration

    Fix bug 1167759

    Currently, if live migration failed, nova compute did not send
    notification, this causes other services such as nova-scheduler
    to have no chance to do some operations

    The fix add notification logic for live_migration like other
    VM instance operations such as resize, run instance etc.

    Change-Id: Iaf61f5268e07284fd7d69ff331ce3bacd380c02b