[jboss-jira] [JBoss JIRA] (JGRP-1931) Don't drop UNICAST3 message when the window is not created

[Radim Vansa (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-1931?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13079695#comment-13079695 ] 

Radim Vansa commented on JGRP-1931:
-----------------------------------

The idea is that once the message is really received, we shouldn't just throw that away.
{quote}
This is needed to know the lowest seqno of the sender.
{quote}
Can't be the table in a state when it would store the received messages but the lowest seqno is not required yet? (I was not checking the code how complicated this would be).

It's not a critical issue - the tests expected the RPC to be propagated in 200 ms, so I have rewritten them to the assertEventually-style (with polling for long timeout).

> Don't drop UNICAST3 message when the window is not created 
> -----------------------------------------------------------
>
>                 Key: JGRP-1931
>                 URL: https://issues.jboss.org/browse/JGRP-1931
>             Project: JGroups
>          Issue Type: Bug
>    Affects Versions: 3.6.3
>            Reporter: Radim Vansa
>            Assignee: Bela Ban
>             Fix For: 3.6.4
>
>
> When an application sends first two (or more) messages in parallel and the the second one is marked as DONT_BUNDLE, usually this one arrives as the first one and is dropped (see https://github.com/belaban/JGroups/blob/master/src/org/jgroups/protocols/UNICAST3.java#L943). After first message arrives, the connection is properly set up, but we need to wait until retransmission for the message (and other non-OOB) messages to be delivered.
> In practice this behaviour should be acceptable, but it causes failures in some time-sensitive tests (particularly in Hibernate ORM 2nd level cache).



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)