[
https://issues.apache.org/jira/browse/AMQ-8287?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17355895#comment-17355895
]
ASF subversion and git services commented on AMQ-8287:
------------------------------------------------------
Commit 0f952f72f798c5626dcd6270c9282e0e8988bd41 in activemq's branch
refs/heads/activemq-5.16.x from Christopher L. Shannon (cshannon)
[ https://gitbox.apache.org/repos/asf?p=activemq.git;h=0f952f7 ]
AMQ-8287 - fix NIOSSLTransport deadlock with serviceRead lock
This narrows the lock that was added to serviceRead() to secureRead()
which prevents processing commands while locked which should solve the
deadlock issues
(cherry picked from commit 2fcf2fd75f7c2a04151a2284595a08bf4c03240b)
> Deadlock caused by synchronized on serviceRead() in NIOSSLTransport
> -------------------------------------------------------------------
>
> Key: AMQ-8287
> URL: https://issues.apache.org/jira/browse/AMQ-8287
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.15.15, 5.16.2
> Reporter: Christopher L. Shannon
> Assignee: Christopher L. Shannon
> Priority: Major
> Fix For: 5.17.0, 5.16.3
>
> Time Spent: 50m
> Remaining Estimate: 0h
>
> I ran into a deadlock caused by the fix for AMQ-8169 when using Stomp
> NIOSSLTransport (but it could probably happen for other nio ssl transports)
> The newly added synchronized on the serviceRead() caused a deadlock between
> the transport and the TransportConnection. One thread acquired a lock on the
> TransportConnection and was waiting on serviceRead() to acquire the
> NIOSSLTransport lock. Another thread had was inside serviceRead() so it
> acquired the NIOSSLTransport lock and was then later waiting for the
> TransportConnection lock.
> The main issue is that processCommand(plain) ends up being protected by the
> lock and since there's multiple brokers/filters running we run into a
> deadlock (my current deadlock happened processing a ConsumerInfo command)
> To fix this we simply need to narrow the lock as it's too broad. The goal
> here was to protect the the reading off the channel concurrently (so really
> the secureRead() method so we can move the lock to secureRead() and not lock
> the entire serviceRead() call and that should fix the deadlock problem while
> still solving the initial issue which was demonstrated by the StompNIOSSL
> failing before this fix. I will open a new Jira shortly and push a fix.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
