Optimize endpoint blacklist

[jpcaruana]

Hi @MickaelBergem,

as I was reading through the code, I realized that checking for blacklisted endpoint could be tedious as the list grows.

I tested it

if __name__ == "__main__":
    b = ['/jobs/*/results', '/jobs/*/*/progress', ]
    times = timeit.Timer(functools.partial(is_in_blacklist, '/test', b)).repeat(repeat=3, number=1000)
    print(times)

    b = b + b + b + b + b + b
    b = b + b + b + b + b + b
    b = b + b + b + b + b + b
    b = b + b + b + b + b + b
    b = b + b + b + b + b + b

    times = timeit.Timer(functools.partial(is_in_blacklist, '/test', b)).repeat(repeat=3, number=1000)
    print(times)

the result was obvious, the more endpoint you add, the longer it takes:

short list: [0.004838313005166128, 0.004496372013818473, 0.004295461985748261]
long list:  [30.687488282012055, 33.16138723300537, 27.81307909899624]

with the code from this PR, the result is better even for a short list:

short list: [0.001244339015102014, 0.0012035770050715655, 0.0014499509998131543]
long list:  [4.755746807000833, 4.928609041991876, 4.693075067014433]

With this PR, you pay the re.compiile prize once at the init of HowFastFlaskMiddleware - I don't know if it is worth it.

[MickaelBergem]

That performance improvement looks amazing, thanks a lot for the contribution!

Two ideas I had reading this PR (nothing immediately important):

• we could probably support passing regexps in the blacklist if some users need it, in the future
• once the library grows, I would be curious to explore what are the best practices around running performance benchmarks in CI

Also, regarding your note on the price of the initial compile, I assume this would only have an impact in serverless environments where the entire Flask application would be initialized often (and still), so it looks like a real win for me given the gains you mentioned.

Feel free to merge as soon as you want :)