[ https://issues.apache.org/jira/browse/HBASE-22642?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16880011#comment-16880011 ]
Xiaolin Ha commented on HBASE-22642: ------------------------------------ Two important things of RSGroup are stored meta info and region movements. I think we can make move operations in RSGroup be idempotent, because repeatedly moving tables/servers to a group might not make regions be moved repeatedly, like the operations of balancing. Meta info contains reflects of group name and tables/servers, which should not mind to be written repeatedly. As a result, there is no benefit of refusing to move tables/servers to a target group which is the same as their origin group. Any comments is welcome. > Move operations of RSGroup should allow retry and correct region assignments > ---------------------------------------------------------------------------- > > Key: HBASE-22642 > URL: https://issues.apache.org/jira/browse/HBASE-22642 > Project: HBase > Issue Type: Bug > Components: rsgroup > Reporter: Xiaolin Ha > Assignee: Xiaolin Ha > Priority: Major > > Currently, when moving tables or servers to a group, only groupInfo is > checked. And in RSGroup implementation, groupinfo is written to disk before > regions movements are done. If there are some problems caused move regions > abort, some regions will be on wrong regionservers. What's the worse, retry > the move operation will be rejected because of the correct groupinfo. > We think when moving, not only groupInfo should be checked, but also relevant > region assignments should be checked and corrected. -- This message was sent by Atlassian JIRA (v7.6.3#76005)