[
https://issues.apache.org/jira/browse/HBASE-9097?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13791729#comment-13791729
]
Gary Helmling commented on HBASE-9097:
--------------------------------------
Maybe we do need more flexibility here (putting HBASE_CLASSPATH first or last),
but this specific example sounds like a bug in the YCSB packaging. FAT jars
are an anti-pattern, and in my opinion, Hadoop and HBase dependencies from
applications should always use something like a "provided" scope so that they
are not bundled. Applications should be using the versions deployed on a
cluster.
> Set HBASE_CLASSPATH before rest of the classpath
> ------------------------------------------------
>
> Key: HBASE-9097
> URL: https://issues.apache.org/jira/browse/HBASE-9097
> Project: HBase
> Issue Type: Bug
> Components: scripts
> Affects Versions: 0.98.0, 0.95.2, 0.94.11
> Reporter: Jesse Yates
> Assignee: Jesse Yates
> Fix For: 0.98.0, 0.95.2, 0.94.11
>
> Attachments: hbase-9097-v0.patch
>
>
> We encountered this when one of the hadoop test jars (specifically
> hadoop-mapreduce-client-jobclient-2.0.0-cdh4.3.0-tests.jar, but that's beside
> the point) had an hdfs-site.xml. This clobbered the hdfs-site.xml that we
> included on the classpath via HBASE_CLASSPATH in hbase-env.sh, meaning the
> master didn't start in HA NN mode, because the proxy-provider wasn't found in
> the hdfs-site.xml from the test jar (even though it was in our config file)
> because that was the first resolution of that file.
> This should be a fairly simple fix in bin/hbase, but has some potentially
> wide-ranging effects on existing installs that just 'happen' to work.
> Generally, I'd expect things set on the HBASE_CLASSPATH to take precedence
> over anything else when starting the hbase daemon.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
