[ https://issues.apache.org/jira/browse/HBASE-2975?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Todd Lipcon updated HBASE-2975: ------------------------------- Status: Resolved (was: Patch Available) Hadoop Flags: [Reviewed] Fix Version/s: 0.90.0 Resolution: Fixed Committed to 0.89.20100830 branch, plus trunk. Thanks for reviewing stack. > DFSClient names in master and RS should be unique > ------------------------------------------------- > > Key: HBASE-2975 > URL: https://issues.apache.org/jira/browse/HBASE-2975 > Project: HBase > Issue Type: Bug > Components: master, regionserver > Affects Versions: 0.90.0 > Reporter: Todd Lipcon > Assignee: Todd Lipcon > Priority: Critical > Fix For: 0.90.0 > > Attachments: hbase-2975.txt > > > In the post-newmaster trunk, there's some code which gives the regionserver > and masters fancy names based on hostname and port. This breaks log recovery, > though, if the master starts recovering a log (ie has an append lease on a > log file), then crashes and comes back on the same port. The NN doesn't see > this as a new client, since the client name is the same, so it thinks it > still holds a lease on the file. The new master, though, can't call append() > because the NN thinks it's appending, so it loops forever. -- This message is automatically generated by JIRA. - You can reply to this email to add a comment to the issue online.