riverqueue/river v0.51.0 on GitHub

Added JobDeleteManyParams.Where, an escape hatch for deleting jobs by an arbitrary SQL predicate, like JobListParams.Where does for lists. For example, it can select queued jobs whose args name a given customer. A Where predicate counts as a filter, so it doesn't need UnsafeAll, and each predicate is wrapped in parentheses so that an OR in it can't widen the deletion past other filters or to running jobs. The existing batch limit (100 jobs by default) and Postgres behavior of skipping locked jobs still apply. PR #1570.

Fixed a running job not being cancelled after Client.JobCancel when the cancellation notification was lost, as happens if it's sent while the client's notifier is reconnecting. Clients with a notifier now also check their running jobs for cancellation requests once a minute. Client.JobRetry now clears an earlier attempt's cancellation request so that the retried job isn't cancelled again. PR #1564.
Fixed Client.InsertManyFast panicking after its transaction committed on a client configured with queues when the driver doesn't support a listener, such as a riverdatabasesql driver created with New instead of NewWithPgxListener. It now wakes producers for the inserted jobs' queues right away, like Client.InsertMany. PR #1558.
Fixed Client.QueueUpdate without metadata crashing the program when the client works that queue and the driver doesn't support a listener. PR #1558.
Fixed the notifier's run loop blocking forever when a topic LISTEN or UNLISTEN handed to it finished after its caller had stopped waiting (after 5 seconds, with timed out waiting for interrupt operation), as can happen on a slow or unresponsive connection. Client.Stop then never completed, and neither did a Client.Start that failed this way. PR #1559.