From 09bd503a9842857480bd4703d27335e83dd30571 Mon Sep 17 00:00:00 2001 From: Kei Masumoto Date: Sat, 11 Jun 2011 19:48:48 +0900 Subject: block migration feature added --- nova/compute/manager.py | 55 +++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 46 insertions(+), 9 deletions(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index 245958de7..f7604a253 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -921,11 +921,13 @@ class ComputeManager(manager.SchedulerDependentManager): """ return self.driver.update_available_resource(context, self.host) - def pre_live_migration(self, context, instance_id, time=None): + def pre_live_migration(self, context, instance_id, time=None, + block_migration=False, **kwargs): """Preparations for live migration at dest host. :param context: security context :param instance_id: nova.db.sqlalchemy.models.Instance.Id + :param block_migration: if true, prepare for block migration """ if not time: @@ -977,17 +979,24 @@ class ComputeManager(manager.SchedulerDependentManager): # onto destination host. self.driver.ensure_filtering_rules_for_instance(instance_ref) - def live_migration(self, context, instance_id, dest): + # Preparation for block migration + if block_migration: + self.driver.pre_block_migration(context, + instance_ref, + kwargs.get('disk')) + + def live_migration(self, context, instance_id, + dest, block_migration=False): """Executing live migration. :param context: security context :param instance_id: nova.db.sqlalchemy.models.Instance.Id :param dest: destination host + :param block_migration: if true, do block migration """ # Get instance for error handling. instance_ref = self.db.instance_get(context, instance_id) - i_name = instance_ref.name try: # Checking volume node is working correctly when any volumes @@ -998,13 +1007,20 @@ class ComputeManager(manager.SchedulerDependentManager): {"method": "check_for_export", "args": {'instance_id': instance_id}}) - # Asking dest host to preparing live migration. + args = {} + args['instance_id'] = instance_id + if block_migration: + args['block_migration'] = block_migration + args['disk'] = \ + self.driver.get_instance_disk_info(context, instance_ref) + rpc.call(context, self.db.queue_get_for(context, FLAGS.compute_topic, dest), {"method": "pre_live_migration", - "args": {'instance_id': instance_id}}) + "args": args}) except Exception: + i_name = instance_ref.name msg = _("Pre live migration for %(i_name)s failed at %(dest)s") LOG.error(msg % locals()) self.recover_live_migration(context, instance_ref) @@ -1015,9 +1031,11 @@ class ComputeManager(manager.SchedulerDependentManager): # nothing must be recovered in this version. self.driver.live_migration(context, instance_ref, dest, self.post_live_migration, - self.recover_live_migration) + self.recover_live_migration, + block_migration) - def post_live_migration(self, ctxt, instance_ref, dest): + def post_live_migration(self, ctxt, instance_ref, + dest, block_migration=False): """Post operations for live migration. This method is called from live_migration @@ -1068,6 +1086,10 @@ class ComputeManager(manager.SchedulerDependentManager): # Restore instance/volume state self.recover_live_migration(ctxt, instance_ref, dest) + # No instance booting at source host, but instance dir + # must be deleted for preparing next block migration + if block_migration: + self.driver.destroy(instance_ref) LOG.info(_('Migrating %(i_name)s to %(dest)s finished successfully.') % locals()) @@ -1075,14 +1097,20 @@ class ComputeManager(manager.SchedulerDependentManager): "Domain not found: no domain with matching name.\" " "This error can be safely ignored.")) - def recover_live_migration(self, ctxt, instance_ref, host=None, dest=None): + def recover_live_migration(self, ctxt, instance_ref, host=None, + dest=None, delete=True): """Recovers Instance/volume state from migrating -> running. :param ctxt: security context :param instance_id: nova.db.sqlalchemy.models.Instance.Id :param host: DB column value is updated by this hostname. If none, the host instance currently running is selected. - + :param dest: + This method is called from live migration src host. + This param specifies destination host. + :param delete: + If true, ask destination host to remove instance dir, + since empty disk image was created for block migration """ if not host: host = instance_ref['host'] @@ -1101,6 +1129,15 @@ class ComputeManager(manager.SchedulerDependentManager): if dest: volume_api.remove_from_compute(ctxt, volume_id, dest) + # TODO: Block migration needs empty image at destination host + # before migration starts, so if any failure occurs, + # any empty images has to be deleted. but not sure adding below + # method is appropreate here. for now, admin has to delete manually. + # rpc.call(ctxt, + # self.db.queue_get_for(ctxt, FLAGS.compute_topic, dest), + # {"method": "self.driver.destroy", + # "args": {'instance':instance_ref}) + def periodic_tasks(self, context=None): """Tasks to be run at a periodic interval.""" error_list = super(ComputeManager, self).periodic_tasks(context) -- cgit From 9b52343f792d83647978c7edbfe700258e3ddae2 Mon Sep 17 00:00:00 2001 From: Kei Masumoto Date: Mon, 20 Jun 2011 08:51:25 +0900 Subject: fix pep8 check --- nova/compute/manager.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index f7604a253..4410ff27e 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1133,7 +1133,7 @@ class ComputeManager(manager.SchedulerDependentManager): # before migration starts, so if any failure occurs, # any empty images has to be deleted. but not sure adding below # method is appropreate here. for now, admin has to delete manually. - # rpc.call(ctxt, + # rpc.cast(ctxt, # self.db.queue_get_for(ctxt, FLAGS.compute_topic, dest), # {"method": "self.driver.destroy", # "args": {'instance':instance_ref}) -- cgit From a6d527646184889863de5ab1082695a29f70988a Mon Sep 17 00:00:00 2001 From: Kei Masumoto Date: Mon, 20 Jun 2011 09:43:34 +0900 Subject: nova.virt.libvirt.connection._live_migration is changed --- nova/compute/manager.py | 20 +++++++++----------- 1 file changed, 9 insertions(+), 11 deletions(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index 402d5829e..f32dfe6ab 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1220,8 +1220,7 @@ class ComputeManager(manager.SchedulerDependentManager): "Domain not found: no domain with matching name.\" " "This error can be safely ignored.")) - def recover_live_migration(self, ctxt, instance_ref, host=None, - dest=None, delete=True): + def recover_live_migration(self, ctxt, instance_ref, host=None, dest=None): """Recovers Instance/volume state from migrating -> running. :param ctxt: security context @@ -1231,9 +1230,6 @@ class ComputeManager(manager.SchedulerDependentManager): :param dest: This method is called from live migration src host. This param specifies destination host. - :param delete: - If true, ask destination host to remove instance dir, - since empty disk image was created for block migration """ if not host: host = instance_ref['host'] @@ -1254,12 +1250,14 @@ class ComputeManager(manager.SchedulerDependentManager): # TODO: Block migration needs empty image at destination host # before migration starts, so if any failure occurs, - # any empty images has to be deleted. but not sure adding below - # method is appropreate here. for now, admin has to delete manually. - # rpc.cast(ctxt, - # self.db.queue_get_for(ctxt, FLAGS.compute_topic, dest), - # {"method": "self.driver.destroy", - # "args": {'instance':instance_ref}) + # any empty images has to be deleted. + # In current version argument dest != None means this method is + # called for error recovering + #if dest: + # rpc.cast(ctxt, + # self.db.queue_get_for(ctxt, FLAGS.compute_topic, dest), + # {"method": "self.driver.destroy", + # "args": {'instance':instance_ref}) def periodic_tasks(self, context=None): """Tasks to be run at a periodic interval.""" -- cgit From c184fa5d03f3f8d7faaff7b583404874de409aa6 Mon Sep 17 00:00:00 2001 From: Kei Masumoto Date: Tue, 21 Jun 2011 20:51:07 +0900 Subject: fixed reviewer's comment. 1. adding dest-instance-dir deleting operation to nova.compute.manager, 2. fix invalid raise statement --- nova/compute/manager.py | 36 +++++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 13 deletions(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index f32dfe6ab..eac543251 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1045,7 +1045,7 @@ class ComputeManager(manager.SchedulerDependentManager): return self.driver.update_available_resource(context, self.host) def pre_live_migration(self, context, instance_id, time=None, - block_migration=False, **kwargs): + block_migration=False, disk=None): """Preparations for live migration at dest host. :param context: security context @@ -1106,7 +1106,7 @@ class ComputeManager(manager.SchedulerDependentManager): if block_migration: self.driver.pre_block_migration(context, instance_ref, - kwargs.get('disk')) + disk) def live_migration(self, context, instance_id, dest, block_migration=False): @@ -1130,17 +1130,18 @@ class ComputeManager(manager.SchedulerDependentManager): {"method": "check_for_export", "args": {'instance_id': instance_id}}) - args = {} - args['instance_id'] = instance_id if block_migration: - args['block_migration'] = block_migration - args['disk'] = \ - self.driver.get_instance_disk_info(context, instance_ref) + disk = self.driver.get_instance_disk_info(context, + instance_ref) + else: + disk = None rpc.call(context, self.db.queue_get_for(context, FLAGS.compute_topic, dest), {"method": "pre_live_migration", - "args": args}) + "args": {'instance_id': instance_id, + 'block_migration': block_migration, + 'disk': disk}}) except Exception: i_name = instance_ref.name @@ -1253,11 +1254,20 @@ class ComputeManager(manager.SchedulerDependentManager): # any empty images has to be deleted. # In current version argument dest != None means this method is # called for error recovering - #if dest: - # rpc.cast(ctxt, - # self.db.queue_get_for(ctxt, FLAGS.compute_topic, dest), - # {"method": "self.driver.destroy", - # "args": {'instance':instance_ref}) + if dest: + rpc.cast(ctxt, + self.db.queue_get_for(ctxt, FLAGS.compute_topic, dest), + {"method": "cleanup", + "args": {'instance_id': instance_ref['id']}}) + + def cleanup(self, ctxt, instance_id): + """ Cleaning up image directory that is created pre_live_migration. + + :param ctxt: security context + :param instance_id: nova.db.sqlalchemy.models.Instance.Id + """ + instances_ref = self.db.instance_get(ctxt, instance_id) + self.driver.cleanup(instance_ref) def periodic_tasks(self, context=None): """Tasks to be run at a periodic interval.""" -- cgit From e01848ed64c4523bb9e375da07e962b5ea1ea6ee Mon Sep 17 00:00:00 2001 From: Kei Masumoto Date: Tue, 21 Jun 2011 21:10:08 +0900 Subject: erase unnecessary TODO: statement --- nova/compute/manager.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index eac543251..16dbda47d 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1249,7 +1249,7 @@ class ComputeManager(manager.SchedulerDependentManager): if dest: volume_api.remove_from_compute(ctxt, volume_id, dest) - # TODO: Block migration needs empty image at destination host + # Block migration needs empty image at destination host # before migration starts, so if any failure occurs, # any empty images has to be deleted. # In current version argument dest != None means this method is -- cgit From 4001ee488420589e345dc42001e6cab9c68a5e12 Mon Sep 17 00:00:00 2001 From: Kei Masumoto Date: Mon, 11 Jul 2011 16:31:31 +0900 Subject: fix comments --- nova/compute/manager.py | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index 7d89dde80..1fbbe0065 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1341,10 +1341,9 @@ class ComputeManager(manager.SchedulerDependentManager): instance_id, block_migration=False): """Post operations for live migration . - :param ctxt: security context + :param context: security context :param instance_id: nova.db.sqlalchemy.models.Instance.Id :param block_migration: block_migration - :param xml: libvirt.xml """ instance_ref = self.db.instance_get(context, instance_id) -- cgit From 79283cbb13d91e3c25e42af765f9da627813a6d8 Mon Sep 17 00:00:00 2001 From: Kei masumoto Date: Fri, 29 Jul 2011 20:03:23 +0900 Subject: merged trunk and fixed post_live_migratioin_at_destination to get nw_info --- nova/compute/manager.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'nova/compute') diff --git a/nova/compute/manager.py b/nova/compute/manager.py index e80394fbd..8aee456fc 100644 --- a/nova/compute/manager.py +++ b/nova/compute/manager.py @@ -1389,7 +1389,7 @@ class ComputeManager(manager.SchedulerDependentManager): # No instance booting at source host, but instance dir # must be deleted for preparing next block migration if block_migration: - self.driver.destroy(instance_ref) + self.driver.destroy(instance_ref, network_info) LOG.info(_('Migrating %(i_name)s to %(dest)s finished successfully.') % locals()) @@ -1409,8 +1409,10 @@ class ComputeManager(manager.SchedulerDependentManager): instance_ref = self.db.instance_get(context, instance_id) LOG.info(_('Post operation of migraton started for %s .') % instance_ref.name) + network_info = self._get_instance_nw_info(context, instance_ref) self.driver.post_live_migration_at_destination(context, instance_ref, + network_info, block_migration) def rollback_live_migration(self, context, instance_ref, -- cgit