`Exchange.publish()` hanging forever after `Delivery tag %r confirmed %r was ignored` warning appears in logs

[Hellzed]

After seeing Delivery tag %r confirmed %r was ignored (a warning log from aiormq, https://github.com/mosquito/aiormq/blob/master/aiormq/channel.py#L360), attempts to use Exchange.publish() will hang forever.

I suspect the confirmation futures (created here: https://github.com/mosquito/aiormq/blob/master/aiormq/channel.py#L661) for the hanging calls to publish() never get resolved despite the client having received a confirmation frame. The warning log seem to indicate the client attempted to resolve a (different?) future, which was already in a done state.

This usually happens under heavy load, I suspect a subtle concurrency bug in the way aio-pika uses aiormq, or a confirmation future handling issue within aiormq.

@mosquito Should I open a bug on the aiormq repo instead?

[leameow]

I might have an hypothesis.

when we publish (in aiormq):

                confirmation = self.confirmations[self.delivery_tag]
                self.message_id_delivery_tag[message_id] = self.delivery_tag

and the return handler (the one that resolve confirmation):

        delivery_tag = self.message_id_delivery_tag.get(message_id)
...
        confirmation = self.confirmations.pop(delivery_tag, None)

If we publish twice the same message_id, seems like it would end up forgetting about the previous delivery_tag/confirmation future and only resolve the latest message published. I didn't test this, but looking at the code it seems to make sense? I will try to create a repro.

[leameow]

Although looking further seems like message_id should be unique (unless explicitely overriden, which we don't do, so never mind)