[
https://issues.apache.org/jira/browse/SPARK-4921?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14256425#comment-14256425
]
Rui Li commented on SPARK-4921:
-------------------------------
I'm not sure if this is intended, but returning process_local for no_pref tasks
may reset {{currentLocalityIndex}} to 0 which may cause more delay later. Seems
there's a check to avoid this but I doubt it's sufficient:
{code}
// Update our locality level for delay scheduling
// NO_PREF will not affect the variables related to delay scheduling
if (maxLocality != TaskLocality.NO_PREF) {
currentLocalityIndex = getLocalityIndex(taskLocality)
lastLaunchTime = curTime
}
{code}
> Performance issue caused by TaskSetManager returning PROCESS_LOCAL for
> NO_PREF tasks
> -------------------------------------------------------------------------------------
>
> Key: SPARK-4921
> URL: https://issues.apache.org/jira/browse/SPARK-4921
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 1.2.0
> Reporter: Xuefu Zhang
> Attachments: NO_PREF.patch
>
>
> During research for HIVE-9153, we found that TaskSetManager returns
> PROCESS_LOCAL for NO_PREF tasks, which may caused performance degradation.
> Changing the return value to NO_PREF, as demonstrated in the attached patch,
> seemingly improves the performance.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
