[rabbitmq-discuss] Exactly Once Delivery

Matthias Radestock matthias at rabbitmq.com
Sat Aug 7 21:33:06 BST 2010


Tony Garnock-Jones wrote:
> Matthias Radestock wrote:
>> That's end-to-end dedup you are thinking of. Nothing wrong with that,
>> and it doesn't require the broker to do/know anything. The context of
>> the discussion here was a "broker dedups publishes" feature.
> Well couldn't the broker take responsibility for the delivery itself by acking?
> And use the same protocol (including perhaps a fresh message ID) to relay a
> message out to a receiver on the outbound leg?
> Perhaps it's a distraction: I guess I was really wondering what the New York vs
> Australia part had to do with it.

1) publisher connects to cloud; ends up connecting to Australia node
2) publisher sends message
3) connection drops, publisher didn't get ack so must resend ...
4) publisher connects to cloud; ends up connecting to NY node
5) publisher re-sends message

At that point the cloud messaging service has two copies of the same 
message in different locations. For the duplication to be detected, some 
information needs to flow between the two locations. Which is expensive.



More information about the rabbitmq-discuss mailing list