[
https://issues.apache.org/jira/browse/SPARK-18881?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16030025#comment-16030025
]
Mathieu D commented on SPARK-18881:
---
Just to mention a workaround for those experiencing the problem : try increase
{{spark.scheduler.listenerbus.eventqueue.size}} (default 1).
It may only postpone the problem, if the queue filling is faster than listeners
for a long time. In our case, we have bursts of activity and raising this
limits helps.
> Spark never finishes jobs and stages, JobProgressListener fails
> ---
>
> Key: SPARK-18881
> URL: https://issues.apache.org/jira/browse/SPARK-18881
> Project: Spark
> Issue Type: Bug
>Affects Versions: 2.0.2
> Environment: yarn, deploy-mode = client
>Reporter: Mathieu D
>
> We have a Spark application that process continuously a lot of incoming jobs.
> Several jobs are processed in parallel, on multiple threads.
> During intensive workloads, at some point, we start to have hundreds of
> warnings like this :
> {code}
> 16/12/14 21:04:03 WARN JobProgressListener: Task end for unknown stage 147379
> 16/12/14 21:04:03 WARN JobProgressListener: Job completed for unknown job
> 64610
> 16/12/14 21:04:04 WARN JobProgressListener: Task start for unknown stage
> 147405
> 16/12/14 21:04:04 WARN JobProgressListener: Task end for unknown stage 147406
> 16/12/14 21:04:04 WARN JobProgressListener: Job completed for unknown job
> 64622
> {code}
> Starting from that, the performance of the app plummet, most of Stages and
> Jobs never finish. On SparkUI, I can see figures like 13000 pending jobs.
> I can't see clearly another related exception happening before. Maybe this
> one, but it concerns another listener :
> {code}
> 16/12/14 21:03:54 ERROR LiveListenerBus: Dropping SparkListenerEvent because
> no remaining room in event queue. This likely means one of the SparkListeners
> is too slow and cannot keep up with the rate at which tasks are being started
> by the scheduler.
> 16/12/14 21:03:54 WARN LiveListenerBus: Dropped 1 SparkListenerEvents since
> Thu Jan 01 01:00:00 CET 1970
> {code}
> This is very problematic for us, since it's hard to detect, and requires an
> app restart.
> *EDIT :*
> I confirm the sequence :
> 1- ERROR LiveListenerBus: Dropping SparkListenerEvent because no remaining
> room in event queue
> then
> 2- JobProgressListener losing track of job and stages.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org