I'm trying to work out how to approach building a "machine" to send and receive messages to WebSphere MQ, via Twisted. I want it to be as generic as possible, so I can reuse it for many different situations that interface with MQ.
I've used Twisted before, but many years ago now and I'm trying to resurrect the knowledge I once had...
The specific problem I'm having is how to implement the MQ IO using Twisted. There's a pymqi Python library that interfaces with MQ, and it provides all the interfaces I need. The MQ calls I need to implement are:
All of these involve blocking calls to MQ.
As I'm intending to reuse the Twisted/MQ interface many times across a range of projects, should I be looking to implement the MQ IO as a Twisted protocol, as a Twisted transport, or just call the pymqi methods via deferToThread() calls? I realise this is a very broad question with possibly no definitive answer; I'm really after advice from those who may have encountered similar challenges before (i.e. working with queueing interfaces that will always block) and found a way that works well.
If you're going to use this functionality a lot, then having a native Twisted implementation is probably worth the effort. A wrapper based on deferToThread will be less work, but it will also be harder to test and debug, perform less well, and have problems on certain platforms where Python threads don't work extremely well (eg FreeBSD).
The approach to take for a native Twisted implementation is probably to implement a protocol that can speak to MQ servers and give it a rich API for interacting with channels, queues, queue managers, etc, and then build a layer on top of that which abstracts the actual network connection away from the application (as I believe mqi/pymqi largely do).