[jboss-jira] [JBoss JIRA] (JGRP-1732) UNICAST/NAKACK: threads from the internal thread pool should not do work stealing

[Bela Ban (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-1732?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12934019#comment-12934019 ] 

Bela Ban commented on JGRP-1732:
--------------------------------

This can be reproduced with UPerf and 1 sender thread, sending messages to itself: every now and then the sender thread times out on the RPC because the request is not delivered and so no response is received.
When using multiple sender threads, the problem is less likely to occur as different threads will pick up 'leftover' messages and deliver them.
                
> UNICAST/NAKACK: threads from the internal thread pool should not do work stealing
> ---------------------------------------------------------------------------------
>
>                 Key: JGRP-1732
>                 URL: https://issues.jboss.org/browse/JGRP-1732
>             Project: JGroups
>          Issue Type: Enhancement
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>             Fix For: 3.5
>
>
> In NAKACK\{2\} and UNICAST\{2,3\}, threads from all thread pools (regular, OOB and internal) add messages to the table, then grab as many (ordered) messages as possible from the table and pass them up.
> This could lead to the case where an internal thread passes up regular or OOB messages, which might block. There's a (small) chance that this exhausts the internal thread pool.
> Internal threads should therefore never block, and never steal work from other thread pools.
> SOLUTION:
> * An internal thread only adds the message to the table and passes it up if in order
> * If the internal message is OOB, it is passed up and then the thread returns

--
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