[infinispan-issues] [JBoss JIRA] (ISPN-2357) New coordinator installs old CH containing crashed members

Dan Berindei (JIRA) jira-events at lists.jboss.org
Mon Oct 1 06:04:03 EDT 2012


    [ https://issues.jboss.org/browse/ISPN-2357?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12722765#comment-12722765 ] 

Dan Berindei commented on ISPN-2357:
------------------------------------

This should be handled on the coordinator - it should use its own JGroups view to prune leavers from the consistent hashes. After all, it couldn't be the coordinator if it still had the old coordinator in its JGroups view.

Sorry I missed this, it looks like our test suite always gives the coordinator enough time to install the new topology before shutting down.
                
> New coordinator installs old CH containing crashed members
> ----------------------------------------------------------
>
>                 Key: ISPN-2357
>                 URL: https://issues.jboss.org/browse/ISPN-2357
>             Project: Infinispan
>          Issue Type: Feature Request
>          Components: State transfer
>    Affects Versions: 5.2.0.Alpha4
>            Reporter: Erik Salter
>            Assignee: Dan Berindei
>
> After merging the round of pull requests today (including ISPN-2349), there still is an issue where a killed node can appear in the CH even after JGroups passes up the new view.  This appears to happen when the old coordinator was killed.
> Obviously, if the CH contains the old member, writes that map to the segments previously owned by the dead node will fail.
> Here are ISPN trace logs.  You are looking for erm-cluster.
> http://dl.dropbox.com/u/50401510/5.2.0.ALPHA4/relay2/10.30.12.83/server.log.gz (phl-dg1)
> http://dl.dropbox.com/u/50401510/5.2.0.ALPHA4/relay2/10.30.12.84/server.log.gz (phl-dg2)
> http://dl.dropbox.com/u/50401510/5.2.0.ALPHA4/relay2/10.30.12.83/server.log.gz (phl-dg3, the node that was killed)

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira


More information about the infinispan-issues mailing list