[jboss-jira] [JBoss JIRA] Commented: (JGRP-1027) Race conditions can have NAKACK try to send to hosts don't have a receiver window yet

[Bela Ban (JIRA)]

    [ https://jira.jboss.org/jira/browse/JGRP-1027?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12482639#action_12482639 ] 

Bela Ban commented on JGRP-1027:
--------------------------------

I argue that this cannot happen: in ClientGmsImpl, a JoinRsp with a digest excluding the joiner wil throw an exception. This means, a digest always contains an entry for local_addr. And a send() won't be permitted until after the JOIN returned.

Are you running JOIN and SEND in parallel, on separate threads ?

So if A sends a message, xmit_table.get(A) must always return a non-null window !

Have you seen this not being the case ? If so, can this be reproduced ?

Setting the seqno to INITIAL_SEQNO might have big implications and could lead to subtle errors. I poured many hours into getting the digests and seqnos right and would like to avoid touching this part of the code.

Again, if there's a null window for a sender, then that's a bug and we need to look into what's causing this and fix the bug, rather than create a new window in ad-hoc fashion !

> Race conditions can have NAKACK try to send to hosts don't have a receiver window yet 
> --------------------------------------------------------------------------------------
>
>                 Key: JGRP-1027
>                 URL: https://jira.jboss.org/jira/browse/JGRP-1027
>             Project: JGroups
>          Issue Type: Bug
>    Affects Versions: 2.6.12
>            Reporter: Gray Watson
>            Assignee: Bela Ban
>            Priority: Minor
>             Fix For: 2.9
>
>         Attachments: NAKACK.java.patch
>
>
> We have seen NPEs in NAKACK trying to send to hosts who do not have a registered receiver window.  I'd like to apply the attached patch which does a better job of creating a receiver window on the fly in these circumstances.  One issue what the sequence number should be initialize to -- the patch uses 0 which has worked for us.
> Thanks,

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira