[
https://issues.apache.org/jira/browse/HBASE-3194?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12928074#action_12928074
]
Gary Helmling commented on HBASE-3194:
--------------------------------------
Using reflection for isolation should work fine and should allow running
against both versions without rebuilding. I'm working it out now.
The easy part is getting the current UGI. The harder part is "setting" the
current UGI (only needed by MiniHBaseCluster and test code at the moment),
since secure Hadoop changed this to UGI.doAs() with a PrivilegedAction instance
wrapping the actual execution. I'll sort out an initial attempt at isolating
that and we can discuss the general approach.
> HBase should run on both secure and vanilla versions of Hadoop 0.20
> -------------------------------------------------------------------
>
> Key: HBASE-3194
> URL: https://issues.apache.org/jira/browse/HBASE-3194
> Project: HBase
> Issue Type: Bug
> Reporter: Gary Helmling
>
> There have been a couple cases recently of folks trying to run HBase trunk
> (or 0.89 DRs) on CDH3b3 or secure Hadoop. While HBase security is in the
> works, it currently only runs on secure Hadoop versions. Meanwhile HBase
> trunk won't compile on secure Hadoop due to backward incompatible changes in
> org.apache.hadoop.security.UserGroupInformation.
> This issue is to work out the minimal set of changes necessary to allow HBase
> to build and run on both secure and non-secure versions of Hadoop. Though,
> with secure Hadoop, I don't even think it's important to target running with
> HDFS security enabled (and krb authentication). Just allow HBase to build
> and run in both versions.
> I think mainly this amounts to abstracting usage of UserGroupInformation and
> UnixUserGroupInformation.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
