[
https://issues.apache.org/jira/browse/HDFS-3128?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13966706#comment-13966706
]
Colin Patrick McCabe commented on HDFS-3128:
--------------------------------------------
Let me summarize this to help anyone reading it later. Some parts of this
change fixed issues where tests were creating entries in the /tmp directory on
the local filesystem. This is undesirable because it creates problems when
tests run concurrently on the same machine.
However, other parts of this change altered where tests created files in HDFS
MiniDFSClusters. These parts were not needed, and caused problems on Windows.
This is because {{test.build.data}} contains a C: drive on Windows, which
causes problems on HDFS. Since the Jenkins pre-commit jobs don't include
Windows, Andrew and I were not aware of this.
In HDFS-6189, [~cnauroth] fixed the Windows-specific issues by going back to
the old way of manually specifying HDFS test roots for a few tests. Other
tests like TestResolveHdfsSymlink haven't been altered from how they were
changed by this patch.
Overall, I think this problem is bound to keep occurring until we get Windows
build slaves. Basically just as we discussed in the branch-1-win merge vote.
> Unit tests should not use a test root in /tmp
> ---------------------------------------------
>
> Key: HDFS-3128
> URL: https://issues.apache.org/jira/browse/HDFS-3128
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: test
> Affects Versions: 2.3.0
> Reporter: Eli Collins
> Assignee: Andrew Wang
> Priority: Minor
> Fix For: 2.4.0
>
> Attachments: hdfs-3128-1.patch
>
>
> Saw this on jenkins, TestResolveHdfsSymlink#testFcResolveAfs creates
> /tmp/alpha which interferes with other executors on the same machine.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
