[
https://issues.apache.org/jira/browse/HADOOP-3103?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vinod Kumar Vavilapalli updated HADOOP-3103:
--------------------------------------------
Attachment: HADOOP-3103.for.17
HADOOP-3103.for.16
Attaching two patches: HADOOP-3103.for.16 and HADOOP-3103.for.17.
* The patch for Hadoop 17 has testcases, while the one for Hadoop 16 doesn't
(and it cannot have, 'coz the testcase framework is not there in 16).
* testcases are written for verifying the presence of the parameter only on
the client side. Testcases for server side are more involving. Skipping them
for now, seems ok, 'coz, as of now, the parameter is not going to be used at
all on the server side.
* As per Hemanth's comments(outside Jira), making hadoop.tmp.dir as a single
value parameter instead of a comma separated list, which seems to be how
hadoop.tmp.dir is being used currently.
* Hadoop.tmp.dir *is* marked final on daemons side, so client side config
parameter won't override this on the server side.
> [HOD] Hadoop.tmp.dir should not be set to cluster directory
> -----------------------------------------------------------
>
> Key: HADOOP-3103
> URL: https://issues.apache.org/jira/browse/HADOOP-3103
> Project: Hadoop Core
> Issue Type: Bug
> Components: contrib/hod
> Reporter: Vinod Kumar Vavilapalli
> Assignee: Vinod Kumar Vavilapalli
> Priority: Blocker
> Fix For: 0.16.2
>
> Attachments: HADOOP-3103, HADOOP-3103.for.16, HADOOP-3103.for.17
>
>
> Currently HOD generates hadoop.tmp.dir pointing to the cluster
> directory(hod.clusterdir), which users typically create in their home
> directories. This hadoop.tmp.dir is used on the client side when a jar is
> run(org/apache/hadoop/util/RunJar.java). The user home directories might be
> NFS mounted in some environments, which might result in a hit, for example
> when running a jar.
> So, HOD *should not* generate hadoop.tmp.dir pointing to cluster directory.
> Instead, it should use hod.temp-dir, which *is* the location that
> dfs.client.buffer.dir uses to alleviate any such NFS hit problems. Further,
> though currently hadoop.tmp.dir is not used on hadoop daemons side, it would
> be good if HOD generates it on daemons side too, and thus preclude any
> related problems in future.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
