[rabbitmq-discuss] "local queue" ...."remote queue"

[Matthias Radestock]

Oren,

On 25/03/11 18:46, Oren Shomron wrote:
> RabbitMQ offers persistent messaging over durable exchanges - * if you
> can get your message published in the first place *
>
> The issue is what a publisher should do with a message when he is unable
> for whatever reason to reach the broker.
>
> I feel that this is not a problem that should be dealt with on the
> application layer, but rather through the messaging infrastructure.

You could get publishers to connect to a local broker residing in their 
network or even machine, and configure that broker with the 
rabbitmq-shovel plug-in to shovel messages to the main broker.

That kind of setup masks failures of the network and main broker. But it 
does of course introduce a different failure point, i.e. the local 
broker. What if that one fails?

The messaging infrastructure libraries the app uses could do their own 
buffering, but memory is bounded and disks can fail...

In the end, no matter how far you push the boundaries of the messaging 
infrastructure, there is always the possibility of some fault occurring 
which prevents the application from publishing messages, at least 
temporarily. The app either needs to accept that risk or handle the failure.

Regards,

Matthias.