4.x Benchmarking and Performance Tuning

[kgriffs]

• Run benchmarks on 4.0 vs. 2.0 & 3.0 and take care of any performance regressions we find.
• Also, we will need to create some baseline ASGI performance tests.
• Update falconframework.org with new benchmarks

[KhDenys]

@kgriffs Will you add django 3.x with wsgi and asgi modes to the benchmarks?

[vytas7]

Thanks for the reminder @KhDenys ! Yes, we'll try to bring Django 3 into the equation.

[kgriffs]

Created: #1880

[kgriffs]

Created: #1881

[kgriffs]

Run benchmarks on 3.0 vs. 2.0 and take care of any performance regressions we find.

A quick test on my MBP shows that we lost 4-7 μs/req with non-cythonized wheels. With the wheels, we are a few μs faster.

[vytas7]

Nothing I could reproduce on CPython 3.8.5 on Ubuntu 20.04 😰

The cythonized variant is shining in extended benchmarks due to certain parts now implemented in Cython/C.
The pure Python version is very marginally slower than 2.0.0, roughly by about one function call overhead or less.

[kgriffs]

OK, so I figured out that I was accidentally cythonizing the 2.0.0 wheel that I built from source. 🤦‍♂️

No significant regression after all, so I think we are good.

[kgriffs]

I just created new images:

https://hub.docker.com/repository/registry-1.docker.io/falconry/falcon-bench/tags?page=1&ordering=last_updated&name=3.0.1

E.g.,:

docker run -it falconry/falcon-bench:py

Python Version:
==================
Python 3.8.10

Installed Packages:
==================
Package      Version
------------ -------
asgiref      3.3.4
bottle       0.12.19
click        8.0.0
Django       3.2.3
falcon       3.0.1
Flask        2.0.0
itsdangerous 2.0.1
Jinja2       3.0.0
MarkupSafe   2.0.0
pip          21.1.1
pytz         2021.1
setuptools   56.2.0
sqlparse     0.4.1
Werkzeug     2.0.1
wheel        0.36.2