issue with 0.2.0 - publisher killed randomly about amqp HOT 6 CLOSED

@matteosister With the change in 0f0c825, if the channel process is dead, :amqp_channel.call will fail, whereas with :amqp_channel.cast it would just return :ok even if no channel process is there to transmit the message to the broker.

My first thought is that the change would just make the problem visible, not actually cause it. Could it be that you were previously just losing messages under high load?

from amqp.

Another guess is that there might be code closing a channel or connection somewhere in your tests. Try this code and see if your tests pass?

def handle_call({:publish, exchange, routing_key, payload}, _from, chan) do
    temp_chan = get_channel()
    :ok = Basic.publish temp_chan, exchange, routing_key, payload
    Channel.close temp_chan
    Connection.close temp_chan.connection
    {:reply, :ok, chan}
 end

Obviously it is not efficient to run this on production so just to confirm the suspicion.
If you close a channel in middle of tests, I suggest you to open it again.

from amqp.

thanks @pma and @ono

I confirm that by manually opening a new channel and closing it after publishing the message the test suite work! So there must be something that closes the channel...currently investigating...

from amqp.

it turns out it was just a problem with amqp. I was not declaring the exchange I was publishing to. So, as @pma said, the change from cast to call in the publishing call surfaced the problem. It was happening only in test, and not in prod (due to the fact that exchange in prod is already declared). So, I'm happy to finally be on 0.2.0 👍

many thanks!

from amqp.

@matteosister Great. Glad to hear it's sorted. Yeah, we expect this kind of implications but in a good way. You want to be alerted when the message was not delivered to the queue/exchange.

Thanks for the follow-up 👍

from amqp.

@ono totally agree with the design choice! 👍

from amqp.