Python WebFrameworks Performance

WebFrameworks for testing:

• aiohttp
• fastapi
• litestar

Description

Here 3 identical web applications with the same components and routes are described:

Components:

• Database - asyncpg;
• Cache - redis;
• DI - dishka;

Routes:

• GET /ping
• GET /db/one
• POST /db/insert
• GET /cache/get
• POST /cache/set
• GET /mix

Run services

docker compose up --build -d

Test Performance

export $APP_HOST=127.0.0.1 # or your real IP

Ping

wrk -t8 -c512 -d2m http://$APP_HOST:8001/ping          # aiohttp
wrk -t8 -c512 -d2m http://$APP_HOST:8002/ping          # litestar
wrk -t8 -c512 -d2m http://$APP_HOST:8003/ping          # fastapi

Read from DB

wrk -t8 -c512 -d2m --latency -s ./wrk_scripts/db_one.lua http://$APP_HOST:8001
wrk -t8 -c512 -d2m --latency -s ./wrk_scripts/db_one.lua http://$APP_HOST:8002
wrk -t8 -c512 -d2m --latency -s ./wrk_scripts/db_one.lua http://$APP_HOST:8003

Mix

wrk -t8 -c512 -d2m --latency -s ./wrk_scripts/mix.lua http://$APP_HOST:8001
wrk -t8 -c512 -d2m --latency -s ./wrk_scripts/mix.lua http://$APP_HOST:8002
wrk -t8 -c512 -d2m --latency -s ./wrk_scripts/mix.lua http://$APP_HOST:8003

Experiments

You can try granian for litestar and fastapi servers:

granian performance.server_litestar:app --host 0.0.0.0 --port 8000 --workers 4 --workers-max-rss 1024 --workers-lifetime 6h --interface asginl --loop uvloop --backlog 4096 --backpressure 1024 --no-access-log --log-level info

granian performance.server_litestar:app --interface asginl --workers 4 --workers-lifetime 6h --workers-max-rss 1024 --loop uvloop --backlog 4096 --backpressure 1024 --no-access-log --log-level info