[ https://issues.apache.org/jira/browse/HBASE-20700?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16505674#comment-16505674 ]
stack commented on HBASE-20700: ------------------------------- Agree with your reasoning even that RMP is strange. It has in it all needed to onilne the meta region including recovery (meta recovery is not like other recovery -- it has its own dedicated WALs so it can be onlined before any other region). We run RMP always, because it looks for WALs to split always... its hard to discern it a clean startup from a messy one, just so there just the one way only of onlining meta. SCP fires off an RMP when it notices the crashed server was carrying meta. bq. Oh for meta there is another problem... The RecoverMetaProcedure will hold the exclusive lock for the meta table, and since the MRP for meta will hold the shared lock on meta table so the RecoverMetaProcedure can not be executed... This is a problem though. I need to run the unit test to manufacture the condition? MRP and hbase:meta needs particular treatment? Its not like any other region. It has to be online for all other stuff to work... .so RMP should have precedence. > Move meta region when server crash can cause the procedure to be stuck > ---------------------------------------------------------------------- > > Key: HBASE-20700 > URL: https://issues.apache.org/jira/browse/HBASE-20700 > Project: HBase > Issue Type: Sub-task > Reporter: Duo Zhang > Assignee: Duo Zhang > Priority: Major > Attachments: HBASE-20700-UT.patch > > > As said in HBASE-20682. -- This message was sent by Atlassian JIRA (v7.6.3#76005)