[ https://issues.apache.org/jira/browse/HBASE-17077?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Guanghao Zhang updated HBASE-17077: ----------------------------------- Attachment: HBASE-17077-v1.patch The zk can't delete the old queue znode directory directly. So attach v1 patch which delete the old wal znode one by one. > Don't copy the replication queue which belong to the peer has been deleted > -------------------------------------------------------------------------- > > Key: HBASE-17077 > URL: https://issues.apache.org/jira/browse/HBASE-17077 > Project: HBase > Issue Type: Improvement > Reporter: Guanghao Zhang > Assignee: Guanghao Zhang > Priority: Minor > Attachments: HBASE-17077-v1.patch, HBASE-17077.patch > > > When a region server is dead, then other live region servers will transfer > the dead rs's replication queue to their own queue. Now the live rs first > copy the wals queue to its own znode, then create a new replication source to > replicate the wals. But if the queue belong to a peer has been deleted, it > copy the queue, too. The current steps is: > 1. copy the queue to its own znode > 2. found the queue belong to a peer has been deleted > 3. remove the queue and don't create a new replication source for it > There is a small improvement. The live region server doesn't need to copy the > queue to its own znode. The new steps is: > 1. found the queue belong to a peer has been deleted > 2. remove the queue directly instead of copy it -- This message was sent by Atlassian JIRA (v6.3.4#6332)