[ https://issues.apache.org/jira/browse
/HADOOP-1648?page=com.atlassian.jira.plugin.system.issuetabp
anels:comment-tabpanel#action_12540676 ]
stack commented on HADOOP-1648:
-------------------------------
Thinking about this issue more (brought on by how to run MR
jobs from hbase shell -- HADOOP-1608), my comment above on
23/Jul/07 09:25 PM is confusing what HBASE_CONF_DIR is for;
its content is used starting and stopping hbase clusters,
not config. for a MR job w/ hbase dependencies.
Config. for a MR job -- an hbase-site.xml w/ MR config. --
can either be copied across the cluster into
HADOOP_HOME/conf or bundled up into the job jar
(HADOOP-1622).
So, having contrib jars added to the hadoop CLASSPATH will
make it so things like the hbase TableOutput/InputFormats
are generally available and hbase doesn't have to be added
to every job jar that has a hbase dependency.
But what about contrib libs? In hbase there is at least a
lucene jar that conceivably that a hbase MR might use
(indexing content).
> Add contrib jars to general hadoop CLASSPATH
> --------------------------------------------
>
> Key: HADOOP-1648
> URL: htt
ps://issues.apache.org/jira/browse/HADOOP-1648
> Project: Hadoop
> Issue Type: Improvement
> Reporter: stack
> Priority: Minor
> Attachments: hadoop1648.patch
>
>
> A mapreduce job that depends on any of the hadoop
contrib jars must bundle the contrib jar into its job jar
or copy the contrib jar to the lib dir across the cluster
because hadoop contribs are not on the general hadoop
CLASSPATH. It would be an improvement if such as the
included hbase mapreduce tasks did not require the running
of this extra step.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue
online.
|
