Hive could get stuck when a singleton Oban job was left in the executing state by a worker pod that had gone away. Because the job row remained “incomplete”, Oban’s singleton uniqueness prevented any future scheduled run from being inserted, so the worker stopped firing and downstream monitors (for example the domain evolution Sentry check-in) went silent.
The release enables Oban.Plugins.Lifeline with a 30 minute rescue window. Jobs that stay in executing longer than that on a missing node are now automatically recovered, so a single crashed pod can no longer block singleton workers indefinitely. Normal singleton behavior for live workers is unchanged.