Christopher L. Shannon created AMQ-8183:
-------------------------------------------
Summary: Setting a maxFrameSize can lead to infinite loop and high
cpu usage
Key: AMQ-8183
URL: https://issues.apache.org/jira/browse/AMQ-8183
Project: ActiveMQ
Issue Type: Bug
Components: Broker
Affects Versions: 5.16.1
Reporter: Christopher L. Shannon
Assignee: Christopher L. Shannon
Fix For: 5.17.0, 5.16.2
There is a regression caused by AMQ-7106 that can lead to an infinite loop and
high cpu usage when using the nio+ssl transport if maxFrameSize is configured.
The issue shows up when several messages come into the broker that exceed the
max frame size. The broker properly detects the error and kills the connection
in another thread but the transport thread can get stuck in an infinite loop
inside serviceRead() method of NIOSSLTransport.
It's a bit hard to tell what is going on but the issue appears to be due to the
fact that the transport has already been stopped but there's no condition being
hit to break the loop. The transport logic appears to get into a weird state
trying to process a partial frame in a local buffer in the serviceRead() method
but the logic just keeps spinning and never tries to re-read from the socket
because the currentBuffer in the transport is wasn't allocated due to the async
shutdown.
The fix here is pretty simple and that is to check if the transport has been
stopped and exit the loop. I have only seen this with OpenWire and nio+ssl but
the fix I'm adding will work for all nio+ssl and nio transports.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
