[rabbitmq-discuss] Destroying connection after ShutdownSignalException

[Michael Klishin]

On 1 Nov 2013, at 12:46, Abid Varshani <abid.varshani at gmail.com> wrote:

> 8. Android client receives every other message. That gives me a feeling that RabbitMQ is still assuming the earlier client is still connected and sending the messages to it.

If you reconnect, RabbitMQ treats every new connection as a separate one. It cannot know if it’s coming
from a client that was connected earlier or not. In fact, as soon as it detects a closed connection,
all unacknowledged messages will be requeued.

Please investigate what changes between reconnections. It can be queue names (for server-named queues),
queues getting deleted (for exclusive or auto-delete queues), bindings not being recovered, etc.

Both the client and RabbitMQ do not notice TCP connection loss immediately because of how TCP works:
it tries retransmitting segments first, then gives up after a certain period of time. AMQP 0-9-1 has heartbeats
exactly for this reason: to make it easier for detect unresponsive peers.

So try setting a heartbeat interval to a low value (say, 4-6 seconds) and see if it helps.

As far as connection loss for publishers goes, unless the peer gets unresponsive in the middle of a write
(to the socket), you should get a visible IOException the next time you publish a message.
--
MK

Software Engineer, Pivotal/RabbitMQ