[rabbitmq-discuss] RabbitMQ subscriber loses connection across firewalls after long idle

Emile Joubert emile at rabbitmq.com
Mon Jul 2 13:51:03 BST 2012


On 02/07/12 12:42, Deelo55 wrote:
> I can confirm I see the timeout has been set to 10 for all subscriber
> connections.
> The evidence for the connection not dropping is the fact that  "send_oct"
> and "recv_oct" incrementing (and also the list-connections still shows the
> connection).

Previously you showed a stack trace with that was caused by
java.io.EOFException (normally indicates a dropped connection). Is that
still happening? If I understand this correctly then you are saying that
this command

rabbitmqctl list_connections name timeout

shows an active connection, with a heartbeat set to 10 seconds, and that
successive invocations of this command:

rabbitmqctl list_connections name send_oct recv_oct

shows incrementing values for "send_oct" and  "recv_oct", yet expected
messages were not received when a message was published after 600 seconds.

I strongly suspect that there is a network error behind this problem. A
possible way to solve this is to trace all network traffic on port 5672
at the server and at the client and compare the two streams and
correlate with the broker logfile and timed diagnostic output from the
client. It should be pretty easy to determine from this evidence where
the problem lies.


More information about the rabbitmq-discuss mailing list