[
https://issues.apache.org/jira/browse/HIVE-16104?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15904273#comment-15904273
]
Siddharth Seth commented on HIVE-16104:
---------------------------------------
bq. Wrt special values, monotonic/nanotime can take any value so special values
should not be used.
How about "-1" as a value, don't think we'll see that as a Time.monotonicNow
bq. #3 is completely unrelated, strange that you would ask for it here I'll
make the change.
Think it is related, since the change increases the chances of hitting this
(seen while walking through the patch)
bq. The class has so many TODO comments that adding another relevant one should
not be a problem
That's a style choice. There's nothing wrong with using an executor with a
single thread. If anything, this should have been
Executors.newSingleThreadExecutor
Think we're better off using the mega lock in the scheduler for the wait - that
way it gets interrupted if some other task completes, another task asks to be
scheduled, etc. (Doesn't need to wait for the specific fragment to end)
Using isComplete as the wait mechanism has a race when the setIsCompleted()
invocation happens in TaskRunnerCallable.
{code}
lastVictim = handleScheduleAttemptedRejection(task);
// We killed something.
lastKillTimeMs = clock.getTime();
{code}
Should this check for lastVictim being null?
> LLAP: preemption may be too aggressive if the pre-empted task doesn't die
> immediately
> -------------------------------------------------------------------------------------
>
> Key: HIVE-16104
> URL: https://issues.apache.org/jira/browse/HIVE-16104
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-16104.01.patch, HIVE-16104.02.patch,
> HIVE-16104.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
