[rabbitmq-discuss] Active-Active implementation issues

[Matthias Radestock]

On 01/07/13 11:57, Gomathi Nayagam wrote:
> [Reply] Let's take the below example, the queue has 20 messages and
> after 13 messages are consumed (11 messages are acked, 2 messages ack
> are not reached to broker) Node1 is down. So in the Node2 we could see 9
> messages.

So far so good.

> But the consumer consumes and acked all 9 messages (i can get
> the print statements inside handleCancel()) during the fail over

I have no idea what you observing here. How can the consumer possibly 
consume and ack messages "during the fail over"?

Well, the application code can, i.e. it can process messages that were 
still buffered, and it can issue acks for them. But these acks will be 
lost during the failover.

> It leads to duplication of messages.

If by that you mean that a consumer can see the same message again after 
an abrupt disconnect and reconnect (e.g. due to node failure) then yes, 
that is unavoidable.

Matthias.