activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Eric-AWL <eric.vinc...@atosorigin.com>
Subject broker dead block. Multicast Discovery Notifier thread doesn't stop
Date Mon, 15 Jun 2009 13:09:15 GMT

Hi 

I want to simulate different channels by implementing connections and
network of brokers in an unusual way.

I have 2 processes A and B on the same server. Each have their own embedded
AMQ broker. B is a supervisor process which echanges signal messages.

A is connected on its broker (conn1), and A is optionnaly connected on the
B's broker (conn2), if a previous message on conn1 (answer to a previous
broadcast), tells him that B is on.

A's broker and B's broker are in the same network of brokers
(networkConnector is instance-nc, associated with multicast protocol). Each
has the same networkConnector "instance-nc" multicast configuration

A and B brokers have a transportConnector which is associated with
instance-nc, multicast protocol.
A connector : port 61616
B connector : port 61600

B has an other connector 61602 

If A is started without B, and A is gracefully stop. All is right.

I have a problem when
1 : B (supervisor) is started (B is connected on its broker : vm transport)
2 : A is started (A is connected on its broker : vm transport)
3 : network connections are automatically established by ActiveMQ (61616 and
61600 ports) for instance-nc
4 : conn2 (A->B:61602) is established by my program (tcp transport) 
5 : B is gracefully killed (B internal connection is gracefully closed, and
AMQ embedded broker is stopped)
6 : on A, ExceptionListener is aware that A->B:61602 is broken.
7 : Then, myprogram on A, I wish to leave 
7-1 : conn1 is gracefully stopped and close
7-2 : internal broker is gracefully stopped

In this situation, on A, the instance-nc multicast Discovery thread and
scheduler, doesn't want to stop. (see thread dump)

With trace on, I see that conn2 (A->B:61602) is closed when B is killed. I
see that "instance-nc" networkConnector is closed when A's broker is stopped
by my program


For your information, in the trace, you will see that there is a third
connection (61620) towards a distinct AMQ stand-alone broker which is active
and not killed during the operation. I think all is right with this
"passive" connection.

In the trace, you will see at the end that I tried to close the broken
connection and that "close" generates the last exception. But the thread
stay active although embedded broker is down

Then, if I interrupt the blocked process (CTRL-C), the 2 AMQ threads die
very quickly, and my program leaves.

Version : AMQ 5.3-snapshot

I joined the trace of A internal embedded Broker and the thread dump at the
end of A process.

Eric-AWL

http://www.nabble.com/file/p24034367/AMQ%2Bthread%2Bdump.txt
AMQ+thread+dump.txt 
http://www.nabble.com/file/p24034367/AMQ%2Btrace.txt AMQ+trace.txt 
-- 
View this message in context: http://www.nabble.com/broker-dead-block.-Multicast-Discovery-Notifier-thread-doesn%27t-stop-tp24034367p24034367.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.


Mime
View raw message