[ https://issues.apache.org/jira/browse/HBASE-26884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17513293#comment-17513293 ]
Zheng Wang commented on HBASE-26884: ------------------------------------ Seen this in 2.0(cdh6.0.1, at about 1 years ago) and 2.2.0 rencently, not sure it could happen without misoperation by user. [~anoop.hbase] > Find unavailable regions by the startcode checking on hmaster start up and > reassign them > ---------------------------------------------------------------------------------------- > > Key: HBASE-26884 > URL: https://issues.apache.org/jira/browse/HBASE-26884 > Project: HBase > Issue Type: Improvement > Components: master > Reporter: Zheng Wang > Assignee: Zheng Wang > Priority: Major > > Sometimes we have seen there are regions in open or opening state, but does > not deployed on any rs and without procs for them, and afting checking the > meta table, we find these startcode are expired. > It is no easy to reproduce, may be caused by corner bug or user misoperation. > My approach is add some checking on hmaster start up, if the startcode of the > regionLocation expired, and neither TRSP on region nor SCP on regionserver, > then we should reassign the region, then we can resovle it easily just by > restart hmaster. > Hbck2 maybe also useful for some of them cases, but not easily for common > user to use, especially the number of these regions not small and need to be > recovery quickly. -- This message was sent by Atlassian Jira (v8.20.1#820001)