[jboss-jira] [JBoss JIRA] Commented: (JGRP-1336) GMS: if a member leaving takes a long time, it can get merged back into the cluster

[Bela Ban (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-1336?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12631005#comment-12631005 ] 

Bela Ban commented on JGRP-1336:
--------------------------------

Solution: add an is_leaving flag to Discovery. Set it to true when leaving and false when connecting. When true, discard any discovery requests (and responses).

I have no test case for this as it is an edge case and usually won't occur

> GMS: if a member leaving takes a long time, it can get merged back into the cluster
> -----------------------------------------------------------------------------------
>
>                 Key: JGRP-1336
>                 URL: https://issues.jboss.org/browse/JGRP-1336
>             Project: JGroups
>          Issue Type: Bug
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>            Priority: Minor
>             Fix For: 3.0
>
>
> When JChannel.disconnect() is called, and the leave processing takes a long time (e.g. due to excessive CPU use and GC), then it can happen that the leaving node is merged back into the cluster.
> This occurred in the Infinispan testsuite which is run in parallel (high CPU contention).
> SOLUTION: after sending the LEAVE request, disable discovery, so MERGE2 won't discover us.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira