worker_liveness [SQL0096]

A stream’s worker rows have no live instance, so nothing runs its upkeep.

A worker row with no live instance does no work: expired partitions are not dropped, exceptions are not retried, and the group’s cursor stops advancing. The hourly check reads the collector’s stream-level unclaimed-worker history and produces an active alert after the pending duration (two minutes by default). The worker names come from the same stored sample as the count. The alert’s data carries unclaimed_count and a workers list with each row’s worker, owner, owner_kind, and target_instances.

Fresh healthy evidence resolves the alert. Missing or stale evidence leaves it unchanged; see pending rules.

Only a process running the manager claims these rows: sqlstreams manager run in a process that stays up, or any Consume on the stream. A produce-only deployment sees this alert on every stream it produces to, because nothing in that process runs upkeep.

sqlstreams alert list