lock performance about cashews

[Krukov]

Huge Thanks for sharing it.

I found the problem, is_locked function reserved a connection. So 100 concurrent is_locked coroutines blocking other operations including release lock (unlock)

Also locking without timeout (ttl=None) was unexpected - in this case if lock did not acquire it will raise an exception without waiting for lock release.

Fixed in 5.3.1

I saw your libraries for cache too, rust based memory storage is impressive, do you mind if I decided to use it too as an alternative backend ?

And also could you please share your results of benchmarks or share the repo with code?

[Yiling-J]

Great, then I can continue my benchmarks. Feel free to use Theine as your backend, my Cacheme is doing the same.

I'm still refactoring my Cacheme, this is the results currently:

https://github.com/Yiling-J/cacheme/blob/refactor/reduce_overhead/benchmarks/trace.py

Redis GET means how many redis get command encountered and Load Hits is how many times the slow load function called.
totaly 200k requests and use a zipf generator to generate numbers. Aiocache-2 is aiocache_stampede decorator and cashews-2 is lock=True

1k concurrency

	Time	Redis GET	Load Hits
Cacheme	30 s	166454	55579
Aiocache	46 s	200000	56367
Aiocache-2	63 s	256492	55417
Cashews	51 s	200000	56920
cashews-2	x	x	x

100k concurrency

	Time	Redis GET	Load Hits
Cacheme	30 s	60990	56782
Aiocache	80 s	200000	125085
Aiocache-2	178 s	326417	65598
Cashews	88 s	200000	87894
cashews-2	x	x	x

[Yiling-J]

fix confirmed, this is the result for locked cashews:
1k: redis count 200000, load count 55425, spent 137.26466012001038s
100k: redis count 200000, load count 55348, spent 236.61019372940063s

[Krukov]

Thanks for sharing