Do not set the address of failed PS into PS hosts (#399)

intelligent-machine-learning · May 9, 2023 · edc076b · edc076b
1 parent 6ac2e11
commit edc076b
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/dlrover/python/master/node/ps.py b/dlrover/python/master/node/ps.py
@@ -286,6 +286,7 @@ def get_ps_addrs(self):
  and not ps.is_released
  and ps.status
  in [NodeStatus.INITIAL, NodeStatus.PENDING, NodeStatus.RUNNING]
+ and not ps.timeout(_dlrover_ctx.seconds_to_wait_failed_ps)
  ):
  ps_addrs[ps.rank_index] = ps.service_addr
  ps_addrs = collections.OrderedDict(sorted(ps_addrs.items()))