canonical-ci-engineering team mailing list archive

[Evan Dandrea <evan.dandrea@xxxxxxxxxxxxx>]

On 3 September 2014 09:37, Vincent Ladeuil <vila+ci@xxxxxxxxxxxxx> wrote:
> * The 'smart endpoints and dump pipes' also resonates: it captures what
>   I've been trying to advocate about rabbit queues. We should keep them
>   "dumb", rock solid but dumb. For example: I'm still against the idea
>   that we should nack messages coming from queues because it means we
>   try to make our pipes smarter. I'd rather make our endpoints (workers)
>   smarter by creating a different message to express that the input
>   message couldn't be processed and leave rabbit handle workers death
>   (the only case where we want a message delivered to another worker),

Nack'ing is (I think), and should be nothing more than an
optimisation. If the message isn't nack'ed and isn't ack'ed, it should
time out and get put back on the queue by Rabbit. We seem to agree on
this for the case of worker death.

I don't know of a case where we're using the nack mechanism for
anything else. Can you provide an example, and how you'd rather it be
implemented?