Add support for resizes to resource tracker.

Keep track of additional resources required to resize an instance
to a new host.  Also hold resources for a revert resize to the original
host.

This fixes race conditions where the destination host could become
overscheduled.  (or the source host in the event of a revert)

bug 1065267

Change-Id: Ic565d4e2ab9bee40f25fe9f198e1217cdd92ca1b

nova/compute/api.py

@@ -1564,13 +1564,13 @@ def revert_resize(self, context, instance):
                                task_state=task_states.RESIZE_REVERTING,
                                expected_task_state=None)
 
+        self.db.migration_update(elevated, migration_ref['id'],
+                                 {'status': 'reverting'})
+
         self.compute_rpcapi.revert_resize(context,
                 instance=instance, migration=migration_ref,
                 host=migration_ref['dest_compute'], reservations=reservations)
 
-        self.db.migration_update(elevated, migration_ref['id'],
-                                 {'status': 'reverted'})
-
     @wrap_check_policy
     @check_instance_lock
     @check_instance_state(vm_state=[vm_states.RESIZED])
@@ -1588,14 +1588,14 @@ def confirm_resize(self, context, instance):
                                task_state=None,
                                expected_task_state=None)
 
+        self.db.migration_update(elevated, migration_ref['id'],
+                {'status': 'confirming'})
+
         self.compute_rpcapi.confirm_resize(context,
                 instance=instance, migration=migration_ref,
                 host=migration_ref['source_compute'],
                 reservations=reservations)
 
-        self.db.migration_update(elevated, migration_ref['id'],
-                {'status': 'confirmed'})
-
     @staticmethod
     def _resize_quota_delta(context, new_instance_type,
                             old_instance_type, sense, compare):

nova/compute/claims.py

@@ -29,6 +29,9 @@
 class NopClaim(object):
     """For use with compute drivers that do not support resource tracking"""
 
+    def __init__(self, migration=None):
+        self.migration = migration
+
     @property
     def disk_gb(self):
         return 0
@@ -184,3 +187,35 @@ def _test(self, type_, unit, total, used, requested, limit):
             LOG.info(msg, instance=self.instance)
 
         return can_claim
+
+
+class ResizeClaim(Claim):
+    """Claim used for holding resources for an incoming resize/migration
+    operation.
+    """
+    def __init__(self, instance, instance_type, tracker):
+        super(ResizeClaim, self).__init__(instance, tracker)
+        self.instance_type = instance_type
+        self.migration = None
+
+    @property
+    def disk_gb(self):
+        return (self.instance_type['root_gb'] +
+                self.instance_type['ephemeral_gb'])
+
+    @property
+    def memory_mb(self):
+        return self.instance_type['memory_mb']
+
+    @property
+    def vcpus(self):
+        return self.instance_type['vcpus']
+
+    @lockutils.synchronized(COMPUTE_RESOURCE_SEMAPHORE, 'nova-')
+    def abort(self):
+        """Compute operation requiring claimed resources has failed or
+        been aborted.
+        """
+        LOG.debug(_("Aborting claim: %s") % self, instance=self.instance)
+        self.tracker.abort_resize_claim(self.instance['uuid'],
+                self.instance_type)

nova/compute/manager.py

@@ -1593,6 +1593,9 @@ def confirm_resize(self, context, instance, reservations=None,
             self.driver.confirm_migration(migration, instance,
                                           self._legacy_nw_info(network_info))
 
+            rt = self._get_resource_tracker(instance.get('node'))
+            rt.confirm_resize(context, migration)
+
             self._notify_about_instance_usage(
                 context, instance, "resize.confirm.end",
                 network_info=network_info)
@@ -1637,6 +1640,9 @@ def revert_resize(self, context, instance, migration=None,
 
             self._terminate_volume_connections(context, instance)
 
+            rt = self._get_resource_tracker(instance.get('node'))
+            rt.revert_resize(context, migration, status='reverted_dest')
+
             self.compute_rpcapi.finish_revert_resize(context, instance,
                     migration, migration['source_compute'],
                     reservations)
@@ -1707,8 +1713,8 @@ def finish_revert_resize(self, context, instance, reservations=None,
                                   vm_state=vm_states.ACTIVE,
                                   task_state=None)
 
-            self.db.migration_update(elevated, migration['id'],
-                    {'status': 'reverted'})
+            rt = self._get_resource_tracker(instance.get('node'))
+            rt.revert_resize(context, migration)
 
             self._notify_about_instance_usage(
                     context, instance, "resize.revert.end")
@@ -1725,6 +1731,29 @@ def _quota_rollback(context, reservations):
         if reservations:
             QUOTAS.rollback(context, reservations)
 
+    def _prep_resize(self, context, image, instance, instance_type,
+            reservations, request_spec, filter_properties):
+
+        if not filter_properties:
+            filter_properties = {}
+
+        same_host = instance['host'] == self.host
+        if same_host and not CONF.allow_resize_to_same_host:
+            self._set_instance_error_state(context, instance['uuid'])
+            msg = _('destination same as source!')
+            raise exception.MigrationError(msg)
+
+        limits = filter_properties.get('limits', {})
+        rt = self._get_resource_tracker(instance.get('node'))
+        with rt.resize_claim(context, instance, instance_type, limits=limits) \
+                as claim:
+            migration_ref = claim.migration
+
+            LOG.audit(_('Migrating'), context=context,
+                    instance=instance)
+            self.compute_rpcapi.resize_instance(context, instance,
+                    migration_ref, image, instance_type, reservations)
+
     @exception.wrap_exception(notifier=notifier, publisher_id=publisher_id())
     @reverts_task_state
     @wrap_instance_fault
@@ -1742,30 +1771,9 @@ def prep_resize(self, context, image, instance, instance_type,
                     context, instance, current_period=True)
             self._notify_about_instance_usage(
                     context, instance, "resize.prep.start")
-
             try:
-                same_host = instance['host'] == self.host
-                if same_host and not CONF.allow_resize_to_same_host:
-                    self._set_instance_error_state(context, instance['uuid'])
-                    msg = _('destination same as source!')
-                    raise exception.MigrationError(msg)
-
-                old_instance_type = instance['instance_type']
-
-                migration_ref = self.db.migration_create(context.elevated(),
-                        {'instance_uuid': instance['uuid'],
-                         'source_compute': instance['host'],
-                         'dest_compute': self.host,
-                         'dest_host': self.driver.get_host_ip_addr(),
-                         'old_instance_type_id': old_instance_type['id'],
-                         'new_instance_type_id': instance_type['id'],
-                         'status': 'pre-migrating'})
-
-                LOG.audit(_('Migrating'), context=context,
-                        instance=instance)
-                self.compute_rpcapi.resize_instance(context, instance,
-                        migration_ref, image, instance_type, reservations)
-
+                self._prep_resize(context, image, instance, instance_type,
+                        reservations, request_spec, filter_properties)
             except Exception:
                 # try to re-schedule the resize elsewhere:
                 self._reschedule_resize_or_reraise(context, image, instance,