Skip heartbeat timeout check for workers with no incomplete tasks

Prevents false "heartbeat timed out" warnings when workers finish
their work before the master completes local tile processing.
This commit is contained in:
Robert Wojciechowski
2026-03-08 03:11:50 +00:00
parent ab4e1536a5
commit d387769453
+4
View File
@@ -99,6 +99,10 @@ async def _check_and_requeue_timed_out_workers(multi_job_id, total_tasks):
f"Probe diagnostics: online={probe_queue is not None} queue_remaining={probe_queue}"
)
if incomplete_assigned == 0:
debug_log(f"Worker {worker} heartbeat stale but all tasks complete; skipping")
continue
if busy:
workers_graced.append(worker)
debug_log(f"Heartbeat grace: worker {worker} busy via probe; skipping requeue")