reduce check fault timeout and check straggler timeout to 1 sec (#1439)

Co-authored-by: Ma Jie Yue <[email protected]>
intelligent-machine-learning · Jan 14, 2025 · b09b7f7 · b09b7f7
1 parent f762e17
commit b09b7f7
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/dlrover/python/common/constants.py b/dlrover/python/common/constants.py
@@ -363,10 +363,10 @@ class JobConstant(object):
     MASTER_CLIENT_GRPC_DEFAULT_TIMEOUT = 60
 
     # sleep 3s on NetworkFailureReason.WAITING_NODE
-    MASTER_CLIENT_CHECK_FAULT_TIMEOUT = 3
+    MASTER_CLIENT_CHECK_FAULT_TIMEOUT = 1
 
     # sleep 3s on NetworkFailureReason.WAITING_NODE
-    MASTER_CLIENT_CHECK_STRAGGLER_TIMEOUT = 3
+    MASTER_CLIENT_CHECK_STRAGGLER_TIMEOUT = 1
 
     # sleep 5s before next node check round
     NODE_CHECK_NEXT_ROUND_TIMEOUT = 5