[
https://issues.apache.org/jira/browse/SPARK-7941?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14564459#comment-14564459
]
Cory Nguyen edited comment on SPARK-7941 at 5/29/15 9:28 AM:
-
I'm not entirely sure what you meant by app data from the hadoop user, rather
than you or yarn or spark user. hadoop is the standard user when Spark is
deployed on AWS EMR. The hadoop user submits the spark jobs to yarn - I think
that is why you may be confused by what you saw. However that appcache folder
is spark related because only spark is ran on this cluster, I monitored the
individual node as the job was running and was able to see the growing of the
appcache due to the job being ran.
Yes, this is spark related. No, the containers are not still running. I know
for certain the cache data is related to the spark job running. I thought YARN
would clean this up too, but that was not the case, the data was still there
hours later after the job was killed by spark/yarn.
was (Author: cqnguyen):
I'm not entirely sure what you meant by app data from the hadoop user, rather
than you or yarn or spark user. hadoop is the standard user when Spark is
deployed on AWS EMR. The hadoop user submits the spark jobs to yarn - I think
that is why you may be confused by what you saw. However that appcache folder
is spark related because only spark is ran on this cluster, I monitored the
individual node as the job was running and was about to see the growing of the
appcache due to the job being ran.
Yes, this is spark related. No, the containers are not still running. I know
for certain the cache data is related to the spark job running. I thought YARN
would clean this up too, but that was not the case, the data was still there
hours later after the job was killed by spark/yarn.
Cache Cleanup Failure when job is killed by Spark
--
Key: SPARK-7941
URL: https://issues.apache.org/jira/browse/SPARK-7941
Project: Spark
Issue Type: Bug
Components: PySpark, YARN
Affects Versions: 1.3.1
Reporter: Cory Nguyen
Attachments: screenshot-1.png
Problem/Bug:
If a job is running and Spark kills the job intentionally, the cache files
remains on the local/worker nodes and are not cleaned up properly. Over time
the old cache builds up and causes No Space Left on Device error.
The cache is cleaned up properly when the job succeeds. I have not verified
if the cached remains when the user intentionally kills the job.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
