[
https://issues.apache.org/jira/browse/SPARK-20776?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Josh Rosen updated SPARK-20776:
-------------------------------
Summary: Fix JobProgressListener perf. problems caused by empty TaskMetrics
initialization (was: Fix performance problems in TaskMetrics.nameToAccums map
initialization)
> Fix JobProgressListener perf. problems caused by empty TaskMetrics
> initialization
> ---------------------------------------------------------------------------------
>
> Key: SPARK-20776
> URL: https://issues.apache.org/jira/browse/SPARK-20776
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core
> Affects Versions: 2.2.0
> Reporter: Josh Rosen
> Assignee: Josh Rosen
> Attachments: screenshot-1.png
>
>
> In
> {code}
> ./bin/spark-shell --master=local[64]
> {code}
> I ran
> {code}
> sc.parallelize(1 to 100000, 100000).count()
> {code}
> and profiled the time spend in the LiveListenerBus event processing thread. I
> discovered that the majority of the time was being spent initializing the
> {{TaskMetrics.nameToAccums}} map (see attached screenshot). By replacing the
> use of Scala's LinkedHashMap with a pre-sized Java hashmap I was able to
> remove this bottleneck and prevent dropped listener events.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
