pwtail / readme Goto Github PK

FastAPI is kind of mainstream solution for making async APIs, and also you surely are aware of the greenlet hack that sqlalchemy uses to enable async support. So I greatly would like to get your feedback on this.

This repo (balrog project) proposes an alternative approach to the async support.

@tiangolo @Kludex

I've made an RFC for async django orm

I would be immensely grateful if you could have a look and give some feedback. This issue is good place for that!

@zzzeek

Since RFC text is changing rapidly, here is the original text on greenlets:

Is it a solution?

In my opinion, not. As I said, it is a hack. Here is the list of downsides (obviously not full):

First and the biggest: you write synchronous functions using async I/O at the same time - something considered as a hack
by the Python documentation itself. By doing so you actually acknowledge the fact that your codebase is otherwise unportable and cannot
take big changes. Obviously it gets more complex (and django for example is pretty complex by itself).

Then, the stacks of frames is also broken. It gets split between the greenlets that your code is executed in.
Thus, the debugging experience suffers. Moreover, greenlet is written partly in C and asyncio AFAIK did not
promise to always work with greenlets.

Also, you need to strictly split your API into the public and the private part. The public API should be wrapped in greenlets,
the private one should not.

The performance also can suffer, because every async coroutine hitting database is wrapped. So, you see, there are a
number of downsides.