[
https://issues.apache.org/jira/browse/SPARK-34063?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Calvin Pietersen updated SPARK-34063:
-------------------------------------
Description:
Spark streaming application runs at 60s batch intervals.
The application runs fine processing batches around 40s. After ~8600 batches
(around 6 days), the application all of a sudden hits a wall and processing
time jumps to 2-2.4 minutes, and eventually dies with exit code 137. This
happens consistently every 6 days, regardless of data.
Looking at the application logs, it seems like when the issue begins, tasks are
being completed by executors, however the driver is taking a while to
acknowledge. I have taken numerous memory dumps of the driver (before it hits
the 6 day wall) using *jcmd* and can see the
*org.apache.spark.scheduler.AsyncEventQueue* is growing in size despite the
fact that the application is able to keep up with batches. I have yet to take a
snapshot of the application in the broken stage.
was:
Spark streaming application runs at 60s batch intervals.
The application runs fine processing batches around 40s. After ~8600 batches
(around 6 days), the application all of a sudden hits a wall and processing
time jumps to 2-2.4 minutes, and eventually dies with exit code 137. This
happens consistently every 6 days, regardless of data.
Looking at the application logs, it seems like when the issue begins, tasks are
being completed by executors, however the driver is taking a while to
acknowledge. I have taken numerous memory dumps of the driver (before it hits
the 6 day wall) using *jcmd* and can see the
*org.apache.spark.scheduler.AsyncEventQueue* is growing in size despite the
fact that the application is able to keep up with batches. I have yet to take a
snapshot of the application in the broken stage.
> Major slowdown in spark streaming after 6 days
> ----------------------------------------------
>
> Key: SPARK-34063
> URL: https://issues.apache.org/jira/browse/SPARK-34063
> Project: Spark
> Issue Type: Bug
> Components: Scheduler, Spark Core
> Affects Versions: 3.0.0
> Reporter: Calvin Pietersen
> Priority: Major
> Attachments: normal-job, slow-job
>
>
> Spark streaming application runs at 60s batch intervals.
> The application runs fine processing batches around 40s. After ~8600 batches
> (around 6 days), the application all of a sudden hits a wall and processing
> time jumps to 2-2.4 minutes, and eventually dies with exit code 137. This
> happens consistently every 6 days, regardless of data.
> Looking at the application logs, it seems like when the issue begins, tasks
> are being completed by executors, however the driver is taking a while to
> acknowledge. I have taken numerous memory dumps of the driver (before it hits
> the 6 day wall) using *jcmd* and can see the
> *org.apache.spark.scheduler.AsyncEventQueue* is growing in size despite the
> fact that the application is able to keep up with batches. I have yet to take
> a snapshot of the application in the broken stage.
>
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
