HF RL Explorer

EphemeralRunners get stuck in "Running" or "Failed" state after job failure/cancellation, blocking new jobs

EphemeralRunners get stuck in "Running" or "Failed" state after job failure/cancellation, blocking new jobs: a task in LegoFlow-SWE (Harbor dataset). Secondary scenario – quick cancellation: 5. Queue a job and cancel it immediately after an EphemeralRunner starts being provisioned. 6. The…

The task

*Secondary scenario – quick cancellation:* 5. Queue a job and cancel it immediately after an EphemeralRunner starts being provisioned. 6. The provisioning runner gets assigned to the cancelled job and exits with an error.

Part of Lego-X/LegoFlow-SWE.