Support for asyncpg

[davidolrik]

Would it be feasible to add support for asyncpg? (Maybe even replace aiopg)

procrastinate is written specifically for running on postgres, and so is asyncpg, which is why it is extremely performant. asyncpg is way faster than aiopg, and it is even faster than golangs pgx.

Furthermore, it is supported by encode/databases which plays very well with FastAPI, SQLAlchemy Core and ormar.

[elemoine]

Yes, we've considered it. We could start by adding support for asyncpg as opposed to replacing aiopg. PRs are very welcome!

[ewjoachim]

One thing I see that might make things a bit more complicated is that the format for query parameters is different in aiopg (%(name)s) and asyncpg ($n).

That being said, I believe we should be able to automate the translation between the 2 formats:

In [3]: class Converter:
   ...:     def __init__(self):
   ...:         self.order = []
   ...:
   ...:     def __getitem__(self, key):
   ...:         self.order.append(key)
   ...:         return f"${len(self.order)}"
   ...:
   ...:     def convert_args(self, args):
   ...:         return tuple(args[e] for e in self.order)
   ...:

In [4]: c = Converter()

In [5]: s = "%(abc)s yay %(def)s" % c

In [6]: s
Out[6]: '$1 yay $2'

In [8]: c.convert_args({"def": "bar", "abc": "foo"})
Out[8]: ('foo', 'bar')

(This is just a POC code, we should be able to make something cleaner, but at least we can be reasonably sure that it's feasible)

[ewjoachim]

I don't think I'd like to support asyncpg. psychopg3 is much better. Support for asyncpg can be done as part of an external package.