[infinispan-issues] [JBoss JIRA] (ISPN-2808) Make Infinispan use its own thread pool for sending messages in order to avoid thread deadlocks

[Radim Vansa (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-2808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12760222#comment-12760222 ] 

Radim Vansa commented on ISPN-2808:
-----------------------------------

I've found out that the issue is not limited to non-transactional caches -  When the tx starts before topology change and comes to nodes which are changing the topology, it resends the prepare command to all other nodes which have something to do with the transaction. With big txs (30 operations) almost all nodes might be affected, that means that we have up to cluster size * cluster size * stressor threads operations ongoing in the system, which can easily deplete the threadpool (and given ISPN-2905 these may keep blocking it).
                
> Make Infinispan use its own thread pool for sending messages in order to avoid thread deadlocks
> -----------------------------------------------------------------------------------------------
>
>                 Key: ISPN-2808
>                 URL: https://issues.jboss.org/browse/ISPN-2808
>             Project: Infinispan
>          Issue Type: Feature Request
>            Reporter: Mircea Markus
>            Assignee: Pedro Ruivo
>             Fix For: 5.3.0.Beta1, 5.3.0.Final
>
>
> - when an OOB thread sends a sync request it blocks waiting on a sync in jgroups RequestCorrelator 
> - it gets released by an another OOB thread when the remote node responds
> Now if all the OOB threads are blocked in sending, then there's no available OOB thread to unblock them even if responses from remote nodes have arrived - deadlock. In order to avoid this deadlock we can use a different thread pool for sending OOB messages.
> For a discussion around this please refer to: http://infinispan.markmail.org/search/#query:%20list%3Aorg.jboss.lists.infinispan-dev+page:1+mid:7buuwljp7i7fvnpe+state:results 

--
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