Increase the ping timeout from 5 seconds to a minute. This gives a bit more breathing room in case the workers somehow become unresponsive .

PiperOrigin-RevId: 343400013 Change-Id: I83027aa9a6a9ec2307d88e0b7c7139da0bf7a682
2020-11-19 17:09:13 -08:00 · 2020-11-19 17:09:13 -08:00 · 8a5c91886f
commit 8a5c91886f
parent 8e0d4085f6
1 changed files with 1 additions and 1 deletions
--- a/tensorflow/python/tpu/session_support.py
+++ b/tensorflow/python/tpu/session_support.py
@ -106,7 +106,7 @@ class WorkerHeartbeatManager(object):
    self._session.run(self._ops,
                      {self._request_placeholder: message.SerializeToString()})

-  def ping(self, request=None, timeout_in_ms=5000):
+  def ping(self, request=None, timeout_in_ms=60000):
    """Ping all workers, returning the parsed status results."""
    if request is None:
      request = event_pb2.WorkerHeartbeatRequest()