[rabbitmq-discuss] Getting Rid of a Faulty message

[Emile Joubert]

Hi Oren,

On 15/01/12 09:16, Oren Shochat wrote:
> I have a RabbitMQ on Amazon load balancing between N processors that do
> heavy image processing algorithms (running 1-8 min on an 8 core
> computer) . I'm using RabbitMQ fair dispatch and message acknowledgement
> to make sure the Algorithm requests are distributed fairly amongst the
> processors.
> 
> However once in a while I get a message that causes the processor to
> crash. RabbitMQ of course (since no ack received and connection
> terminated) makes sure the message goes to the next processor which also
> crashes because processors share the same code.  Soon all the processors
> are down. They go up again, but the faulty message stays in the system
> until I get the alarms and purges the RabbitMQ queue.

Full support for toxic/poison message handling is a planned feature (See
the "basic / deliver / 01" reference in the 0-9-1 rules here:
http://www.rabbitmq.com/specification.html#rules)

In the meantime you can make use of the "redelivered" flag as an
indicator that a processor previously failed to process a message and
treat it with greater circumspection.



-Emile