[
https://issues.apache.org/jira/browse/TAJO-1111?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14167118#comment-14167118
]
Hyunsik Choi commented on TAJO-1111:
------------------------------------
As my investigation on the state machines, I found the two missed cases:
KILLED state cannot be transited to T_KILL, T_ATTEMPT_KILLED, or
T_ATTEMPT_SUCCEEDED. The states may receive multiple times form workers because
workers has some state reporter will periodically send the current task state.
They are possible cases in concurrent and distributed cases.
I think that they should also handle T_ATTEMPT_FAILED in the same way. In sum,
SUCCEEDED, FAILED, KILLED should handle T_KILL, T_ATTEMPT_KILLED,
T_ATTEMPT_SUCCEEDED, or T_ATTEMPT_FAILED as ignorable state transitions.
In addition, It's possible if QUERY_ERROR state in Query can take
SUBQUERY_COMPLETED because KILLED and FAILED also send SUBQUERY_COMPLETED event
to Query. In this case, we have to ignore because Query state is already in
QUERY_ERROR.
I'll submit the patch soon.
> TestKillQuery.testKillQueryFromInitState occasionally fails
> -----------------------------------------------------------
>
> Key: TAJO-1111
> URL: https://issues.apache.org/jira/browse/TAJO-1111
> Project: Tajo
> Issue Type: Bug
> Components: query master
> Reporter: Hyunsik Choi
> Priority: Blocker
> Fix For: 0.9.0
>
>
> See the title. this test occasionally fails.
> {noformat}
> [INFO] Results :
> [INFO]
> [INFO] Failed tests:
> [INFO] TestKillQuery.testKillQueryFromInitState:104 expected:<QUERY_KILLED>
> but was:<QUERY_ERROR>
> [INFO]
> [INFO] Tests run: 1234, Failures: 1, Errors: 0, Skipped: 0
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
