If the managed Postgres pod is not ready to accept connections by the time the init Job which is responsible for creating the pg_tgrm extension is ready, then the rollout of a QuayRegistry becomes stuck (even if the extension eventually is created manually or by the Job). The only way to get the Operator to retry the upgrade is by using the Kubernetes API to remove the MigrationsInProgress condition (via cURL since kubectl doesn't support modifying status block).