This incident has been resolved.

During a roll out of a new early error detection system, some jobs have been identified as stalled, and thus have been interrupted and restarted.

We implemented a fix and are currently monitoring the result.