Change the comments of SOFT_DELETED race condition

_reclaim_queued_deletes will only find instances in SOFT_DELETED
state and delete them if they are only old enough. The quotas will be
committed when the instance was soft-deleted so that resource can
be used right after the soft-delete operation.

There are some concern about the quota inconsistency for the operation,
The only case that the quota might be inconsistent is
the compute node died between set instance state to SOFT_DELETED
and quota commit to DB; when compute node start again
it will have no idea the reservation is committed or not or even
expired, since it's a rare case, so marked as todo.

There are some alternatives if the problem need to be fixed, e.g.
able to find reservation when nova compute restart, so we can double
check whether the quota is committed or not.

Partial-Bug: #1296414

Change-Id: Idf9c179b2dd439462a646568ffd5098cd5d7851f

nova/compute/manager.py

@@ -5490,11 +5490,12 @@ def _reclaim_queued_deletes(self, context):
             LOG.debug("CONF.reclaim_instance_interval <= 0, skipping...")
             return
 
-        # FIXME(comstud): Dummy quota object for now. See bug 1296414.
-        # We have potential for inconsistency if we decide here to not
-        # update quotas. _delete_instance() should determine whether or
-        # not to update quotas based on if instance is in a SOFT_DELETED
-        # state or not.
+        # TODO(comstud, jichenjc): Dummy quota object for now See bug 1296414.
+        # The only case that the quota might be inconsistent is
+        # the compute node died between set instance state to SOFT_DELETED
+        # and quota commit to DB. When compute node starts again
+        # it will have no idea the reservation is committed or not or even
+        # expired, since it's a rare case, so marked as todo.
         quotas = quotas_obj.Quotas.from_reservations(context, None)
 
         filters = {'vm_state': vm_states.SOFT_DELETED,