maheshrajus opened a new pull request, #535:
URL: https://github.com/apache/tez/pull/535
Root cause:
The Controlled*VertexManager classes kill the AM with System.exit(-1) inside
onSourceTaskCompleted().
System.exit skips serviceStop(), so the async recovery-event thread can be
killed before it flushes
VertexFinishedEvent / VertexConfigurationDoneEvent to HDFS. On attempt 2,
v2 can't reach CONFIGURED and the
DAG ends in ERROR.
Fix:
Replace the System.exit with an external, deterministic kill (same pattern
as **TEZ-4725**):
- runDAGAndVerify polls until v1 reaches the expected state, then calls
YarnClient.failApplicationAttempt(attempt1) —
YARN shuts the AM down cleanly and the recovery queue drains.
- The three Controlled*VertexManager classes become simple passthroughs
(no more System.exit).
--
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]
