gnet VS typedload

Exciting question. Many goroutines consume resources accordingly and could also generate latency due to this. If you work under Linux: Have you looked at epoll? Since the system has to do basic connection handling anyway, you can shift or save at least part of the task. gnet is a very good library here to handle epoll & co. Maybe it makes sense to have a look at it or directly use epoll under go.

Since this heavily involves networking, take a look into using gnet [0]. You might find some interesting performance improvements by using that over just net.Conn.

[0] https://github.com/panjf2000/gnet

Node is a joke. It's not good for this.

Check out https://github.com/panjf2000/gnet, it also has some links at the end.

Since we're in the useless benchmark, this Go native library completely wreck any C/C++ lib wrapped by Python: https://github.com/panjf2000/gnet

Go does not forces you to do any of that: https://github.com/panjf2000/gnet

Author of typedload here!

FastAPI relies on (not so fast) pydantic, which is one of the slowest libraries in that category.

Don't expect to find such benchmarks on the pydantic documentation itself, but the competing libraries will have them.

[0] https://ltworf.github.io/typedload/

I wrote typedload, which is significantly faster than pydantic. Just uses normal dataclasses/attrs/NamedTuple, has a better API and is pure Python!

I read all the perftests in the repo. I think they nearly all parse a structure that contains a repetition of the same or similar thing a couple hundred thousand times times and the timing function returns the min and max of 5 attempts. I just picked one example for posting.

Not a Python expert, but could the Pydantic tests be possibly not realistic and/or misleading because they are using kwargs in __init__ [1] to parse the object instead of calling the parse_obj class method [2]? According to some PEPs [3], isn't Python creating a new dictionary for that parameter which would be included in the timing? That would be unfortunate if that accounted for the difference.

Something else I think about is if a performance test doesn't produce a side effect that is checked, a smart compiler or runtime could optimize the whole benchmark away. Or too easy for the CPU to do branch prediction, etc. I think I recall that happening to me in Java in the past, but probably not happened here in Python.

[1] https://github.com/ltworf/typedload/blob/37c72837e0a8fd5f350...

[2] https://docs.pydantic.dev/usage/models/#helper-functions

[3] https://peps.python.org/pep-0692/