Open jerith opened 8 years ago
As part of this issue, I'm replacing txamqp (which is effectively unmaintained) with pika, which has support for a variety of I/O mechanisms (including Twisted plain Python stuff) and is actively maintained.
Wouldn't it be easier to first focus on splitting out the txamqp dependencies instead of also adding pika? I'm worried it'll be more difficult for people to follow along otherwise. And where's the code for this living?
The current iteration is in https://github.com/praekelt/vumi/tree/experiment/issue-1000-pika-based-service
The changes to the AMQP layer are pretty hefty even without swapping txamqp out for pika, and txamqp has enough undocumented oddities that I don't think it's really much extra effort to make the switch. I'm pretty sure the pika version will be cleaner and easier to understand, given the amount of time I've spent struggling with txamqp connection setup this week.
That said, this is still fairly experimental and I might end up dropping pika if it doesn't go as smoothly as I expect.
This is necessary for #999.