[jboss-jira] [JBoss JIRA] (JGRP-2131) UNICAST3 drops all messages until it receives the first one

[Bela Ban (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-2131?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13323516#comment-13323516 ] 

Bela Ban commented on JGRP-2131:
--------------------------------

Great investigation, Dennis!

I'm thinking of using an out-of-channel mechanism (e.g. separate SSL socket) to obtain the secret key from the keyserver.
Although, if {{ASYM_ENCRYPT}} learns about the membership via {{JOIN_RSP}} or {{INSTALL_MERGE_VIEW}} messages, it could go ahead and send a request for the secret key to the keyserver...
If {{JOIN_RSP}} is the first message, then things will work, so the above mechanism would be a second-line of-defense.
I'll write a unit test that uses Byteman to inject receiving a message that's not tagged as first before the {{JON_RSP}} message. This should lead to a loop of {{JOIN_REQ}} msgs sent from joiner to coordinator, eventually leading to a singleton cluster for the joiner.

> UNICAST3 drops all messages until it receives the first one
> -----------------------------------------------------------
>
>                 Key: JGRP-2131
>                 URL: https://issues.jboss.org/browse/JGRP-2131
>             Project: JGroups
>          Issue Type: Bug
>    Affects Versions: 3.6.10
>            Reporter: Dennis Reed
>            Assignee: Bela Ban
>             Fix For: 3.6.12, 4.0
>
>
> UNICAST3.getReceiverEntry returns null if it hasn't seen the first message yet.
> This causes UNICAST3.handleDataReceived to drop the message.
> When you add *ENCRYPT, this causes a major deadlock.  *ENCRYPT will queue most messages until it gets the encryption key, which can often include the first message (so UNICAST3 won't see it yet).  Then when an important message such as JOIN_RSP comes through, UNICAST3 drops it.  Since UNICAST3 never lets any messages through in this case, the encryption key will never get set so that *ENCRYPT can pass the first message up and free the deadlock.



--
This message was sent by Atlassian JIRA
(v7.2.3#72005)