Github user gzm55 commented on the pull request:
https://github.com/apache/spark/pull/154#issuecomment-49120476
@tgravescs I file a jira describing this bug:
https://issues.apache.org/jira/browse/SPARK-2508
In a yarn-cluster, the driver is actually running as 'yarn' user in some
container by a nodemanager. When
submitting jobs from other users, e.g. user1, the stagingDir is created in
submitting stage, which is actually a progress run by user1. So the full path
of stagingDir on the hdfs should look like '/user/user1/.staging/...'.
And in cleaning up stage, stagingDir could be removed by the driver. At
this time, we need give stagingDir a full path, so
the driver could remove the correct directory ('/user/user1/.staging/...'),
not '/user/yarn/.staging/...'.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
