Worker fails if exit code = 1 (#400)

intelligent-machine-learning · May 10, 2023 · 7a166b3 · 7a166b3
1 parent edc076b
commit 7a166b3
Showing 1 changed file with 1 addition and 4 deletions.
diff --git a/dlrover/python/master/node/worker.py b/dlrover/python/master/node/worker.py
@@ -264,10 +264,7 @@ def remove_not_participated_workers(self, workers):
  def has_failed_worker(self):
  """Check whether there is failed worker except evicted workers."""
  for worker in self._nodes.values():
- if worker.exit_reason in [
- NodeExitReason.FATAL_ERROR,
- NodeExitReason.UNKNOWN_ERROR,
- ]:
+ if worker.exit_reason == NodeExitReason.FATAL_ERROR:
  return True
  return False