On 09/08/2010 02:17 PM, Yanick Duchesne wrote:
Hello,
we have an issue when redundancy (clustering) is enabled: Qpid crashes for no
(apparent)
reason. We have attempted to diagnose the problem but could not find the cause.
Here it goes:
We have a Java node connecting to one of two Qpid brokers (0.5 or 0.6,
doesn't matter, we get the same) as a client of another Java node which acts
as the server. The client sends messages over a queue, awaiting a response
message from the server on a temporary queue. Both brokers use clustering
(and are part of the same cluster). At some point, after 1 to 3 exchanges
between the client and server nodes, both Qpid brokers simply crash, without
any error being logged. Broker logfile is attached. The typical output that
appears in the broker log files - the last line in the log is always the
same, so it seems that the place at which the failure occurs is consistent.
Can you get a core dump & stack trace from the failed qpidd process?
Do "ulimit -c unlimited" before running qpid. The core file will be dumped in
qpidd's PID directory which is ~/.qpidd by default if you're not using the
service script. Then do "gdb <path-to-qpidd> <core-file>" and at the gdb prompt
"thread apply all backtrace".
Mail the output, that should give us a clue as to whats going on.
Another experiment that might be worthwhile is to try the latest trunk. There
have been a lot of cluster-related fixes since 0.6.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
