[ https://issues.apache.org/jira/browse/HBASE-22658?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16892913#comment-16892913 ]
Reid Chan commented on HBASE-22658: ----------------------------------- Attached two pics, with_patch and without patch. The env is, rsgroup 'a' contains only two servers (13 and 16) and two tables which don't matter. With-patch one worked as expectation, but the without-patch one, region movers moved target regions to whole cluster which violates the table's rsgroup constraints. Didn't encounter any exception, at least 10 tests. FYI sir, [~apurtell] > region_mover.rb should choose same rsgroup servers as target servers > ---------------------------------------------------------------------- > > Key: HBASE-22658 > URL: https://issues.apache.org/jira/browse/HBASE-22658 > Project: HBase > Issue Type: Improvement > Components: rsgroup, shell > Affects Versions: 1.4.10 > Reporter: liang.feng > Priority: Major > Labels: gracefulshutdown, region_mover, rsgroup > Attachments: HBASE-22658.branch-1.002.patch, > HBASE-22658.branch-1.patch, with_patch.png, without_patch.png > > > There are many retries when i am using graceful_stop.sh to shutdown region > server after using regroup, because the target server in a different rsgroup. > This makes it slow to graceful shutdown a regionserver. So i think that > region_mover.rb should only choose same rsgroup servers as target servers. > Region mover is implemented by jruby in hbase1.x and is implemented by java > in hbase2.x . I tried to modify the RegionMover.java class to use the same > logic in hbase2.x, but mvn package failed due to hbase-server module and > hbase-rsgroup moudle needed to depend on each other, then maven throw a "The > projects in the reactor contain a cyclic reference". I couldn't solve it.So I > just uploaded patch for hbase1.x . > -- This message was sent by Atlassian JIRA (v7.6.14#76016)