add a retry on DBDeadlock to _set_allocations()

A job failed with an uncaught DBDeadlock when attempting to run the
claim process in https://review.openstack.org/#/c/483566/. This is
likely due to the update_available_resource() method of the resource
tracker running and updating a resource provider's inventory (and thus
the resource provider record's generation field) at the exact same time
that the scheduler has already started a transaction that is attempting
to claim resources (set allocations) against that same resource
provider. It's a rare situation that is highlighted by the test runs
but we should handle it regardless.

Change-Id: I2c276dc0125b5b9f7a54a1cd431b1b2f5239e93a

nova/objects/resource_provider.py

@@ -18,6 +18,7 @@
 
 import os_traits
 from oslo_concurrency import lockutils
+from oslo_db import api as oslo_db_api
 from oslo_db import exception as db_exc
 from oslo_log import log as logging
 from oslo_utils import versionutils
@@ -1791,6 +1792,7 @@ def _ensure_consumer_project_user(self, conn, consumer_id):
                     # if the consumer already exists.
                     pass
 
+    @oslo_db_api.wrap_db_retry(max_retries=5, retry_on_deadlock=True)
     @db_api.api_context_manager.writer
     def _set_allocations(self, context, allocs):
         """Write a set of allocations.