mridulm commented on PR #40663:
URL: https://github.com/apache/spark/pull/40663#issuecomment-1498310261
This makes sense.
The TaskRunner is visible to heartbeater since it gets added to
`runningTasks` before the task binary is deserialized.
During `TaskRunner.run`, we are registering the accumulators as part of the
deserialization - which can result in race with the heartbearter reading the
metrics.
So this is essentially a bug waiting to happen for two reasons:
* `externalAccums` is getting used in an MT-unsafe way by the task thread
and the reporter thread.
* This is broken in 2.12 as well - we were just not aware of it.
* There always was potential visibility issues in heartbeat for external
accumulators.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
