[rabbitmq-discuss] publisher confirm ack nack ordering

[Tobias Gutzmann]

Hi,

(this message is intended as a reply to https://groups.google.com/forum/#!searchin/rabbitmq-discuss/publisher$20ack$20order/rabbitmq-discuss/0O8Dick9xGA/7mGy8ebV1gUJ )

Assuming that a few messages succeed in order, I sometimes see the ACKs like this (RabbitMQ 3.1.1):
ACK(5, true)
ACK(1, false) -- surely sent first by the broker, but lagged a bit longer on the wire

No problem here, I can handle that, though message 1 appears to be ACKed twice, and documentation says "After a channel is put into confirm mode, all subsequently published messages will be confirmed or nack'd once." which should probably read as "at least once".

Now the problem, which I haven't observed as it should be extremely rare but which still is possible:
ACK (5, true)
NACK (1, false) -- sent first by broker but received last.

Since the client may have already acted upon the ACK, there's a problem here. Or did I miss something?

Regards,
Tobias

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20130704/f778369b/attachment.htm>