[rabbitmq-discuss] Federation and upstream cluster

[Vladislav Pernin]

Hi,

So you finally recommend to set a policy ha-mode all on the queue created
in the upstream exchange's broker, like I did ?

The test is very simple. I send 200 000 messages on the upstream exchange
with a connection factory with both nodes and a retry timer (like in the
Spring Rabbit AmqpAppender).
I receive all messages on the downstream queue and verify that each sent
message has been received.
During the test, I shutdown and restart upstream node.

But you are right, I do not use publisher confirm.
I probably hit the "unsynchronized queue" problem.

Otherwise, do you think that my setup is correct to make sure that the
messages will arrive at the downstream exchange, not duplicated and without
loss (publisher confirm excepted) ?

Regards,
Vladislav

2012/12/28 Matthias Radestock <matthias at rabbitmq.com>

> Vladislav,
>
>
> On 28/12/12 09:32, Vladislav Pernin wrote:
>
>> {{shutdown,{server_initiated_**close,404,
>>                                     <<"NOT_FOUND - home node
>> 'rabbit at remote-server1' of durable queue 'federation: upstream-exchange
>> -> federation-local:downstream-**exchange' in vhost '/' is down or
>> inaccessible">>}},
>>
>
> That happens because the queue used by federation is durable but not
> mirrored, so when the node on which it was created goes down it is no
> longer accessible.
>
> You said you tried making that queue mirrored, but that resulted in
> message loss. How did you detect the message loss? Does the upstream
> publisher have confirms enabled? If it received a confirmation for a
> message then I certainly would expect the message to eventually make it to
> the downstream. In the absence of such confirmation all bets are off.
>
> Regards,
>
> Matthias.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20121228/affbab87/attachment.htm>