[
https://issues.apache.org/jira/browse/HADOOP-1864?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12537750
]
Milind Bhandarkar commented on HADOOP-1864:
-------------------------------------------
Yiping,
This is clearly out of scope for Hadoop because Java uses its own zlib that
does not handle >2GB jar files. There is another Jira issue pending that will
allow .tgz files as cache archives (HADOOP-2019). Would that meet your needs ?
> Support for big jar file (>2G)
> ------------------------------
>
> Key: HADOOP-1864
> URL: https://issues.apache.org/jira/browse/HADOOP-1864
> Project: Hadoop
> Issue Type: Bug
> Components: contrib/streaming
> Affects Versions: 0.14.1
> Reporter: Yiping Han
> Priority: Critical
>
> We have huge size binary that need to be distributed onto tasktracker nodes
> in Hadoop streaming mode. We've tried both -file option and -cacheArchive
> option. It seems the tasktracker node cannot unjar jar files bigger than 2G.
> We are considering split our binaries into multiple jars, but with -file, it
> seems we cannot do it. Also, we would prefer -cacheArchive option for
> performance issue, but it seems -cacheArchive does not allow more than
> appearance in the streaming options. Even if -cacheArchive support multiple
> jars, we still need a way to put the jars into a single directory tree,
> instead of using multiple symbolic links.
> So, in general, we need a feasible and efficient way to update large size
> (>2G) binaries for Hadoop streaming. Don't know if there is an existing
> solution that we either didn't find or took it wrong. Or there should be some
> extra work to provide a solution?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
