使用gunicorn部署fastapi开发的web服务

前言

Gunicorn是一种流行的WSGI HTTP服务器，常用于部署Django和Flask等Python Web框架程序。Gunicorn具有轻量级、高稳定性和高性能等特性，可以轻易提高Python WSGI App运行时的性能。

基本原理

Gunicorn采用了pre-fork模型，也就是一个工作进程和多个worker进程的工作模式。在这个模型中，master进程负责接收并处理外部的连接请求，并将这些请求分配给多个worker进程来处理。

当一个gunicorn服务启动时，master进程会首先创建多个worker工作进程，并将它们初始化为一个无限循环的状态，确保worker进程可以不断地等待接收新请求。当一个请求进来，master会将该请求分发给其中一个worker进程。不同的请求分发到不同的worker进程，可以有效提高请求的响应速度和并发处理能力。当某个worker进程崩溃或异常终止时，master进程会自动重新启动一个新的worker工作进程。

安装

<code>pip install gunicorn

示例

首先编写一个简单的服务，请求响应一个字符串"ok"，代码文件名为demo1.py

from fastapi import FastAPI

import uvicorn

from fastapi.responses import PlainTextResponse

app = FastAPI()

@app.get("/")

async def root():

return PlainTextResponse("ok")

if __name__ == "__main__":

uvicorn.run(app=app, host="0.0.0.0", port=8000, access_log=False)code>

正常启动，然后使用wrk测试。使用uvicorn的qps为11733.60

$ ps -ef | grep demo1

atlas 2449 1162 44 22:06 pts/0 00:01:11 python demo1.py

atlas 2478 2462 0 22:08 pts/6 00:00:00 grep demo1

$ wrk -t2 -c 100 -d 60s http://127.0.0.1:8000

Running 1m test @ http://127.0.0.1:8000

2 threads and 100 connections

Thread Stats Avg Stdev Max +/- Stdev

Latency 8.59ms 4.36ms 276.86ms 91.44%

Req/Sec 5.90k 520.55 7.84k 68.54%

704720 requests in 1.00m, 90.73MB read

Requests/sec: 11733.60

Transfer/sec: 1.51MB

使用gunicorn启动。wrk的测试结果：QPS为25859.47，性能直接提升120%

$ gunicorn demo1:app -b 127.0.0.1:8000 -w 4 -k uvicorn.workers.UvicornWorker

$ ps -ef | grep demo1

atlas 2640 1162 0 22:12 pts/0 00:00:00 /home/atlas/workspace/fastapi-study/bin/python3 /home/atlas/workspace/fastapi-study/bin/gunicorn demo1:app -b 127.0.0.1:8000 -w 4 -k uvicorn.workers.UvicornWorker

atlas 2641 2640 8 22:12 pts/0 00:00:40 /home/atlas/workspace/fastapi-study/bin/python3 /home/atlas/workspace/fastapi-study/bin/gunicorn demo1:app -b 127.0.0.1:8000 -w 4 -k uvicorn.workers.UvicornWorker

atlas 2642 2640 8 22:12 pts/0 00:00:40 /home/atlas/workspace/fastapi-study/bin/python3 /home/atlas/workspace/fastapi-study/bin/gunicorn demo1:app -b 127.0.0.1:8000 -w 4 -k uvicorn.workers.UvicornWorker

atlas 2645 2640 10 22:12 pts/0 00:00:47 /home/atlas/workspace/fastapi-study/bin/python3 /home/atlas/workspace/fastapi-study/bin/gunicorn demo1:app -b 127.0.0.1:8000 -w 4 -k uvicorn.workers.UvicornWorker

atlas 2646 2640 11 22:12 pts/0 00:00:52 /home/atlas/workspace/fastapi-study/bin/python3 /home/atlas/workspace/fastapi-study/bin/gunicorn demo1:app -b 127.0.0.1:8000 -w 4 -k uvicorn.workers.UvicornWorker

atlas 2665 2462 0 22:20 pts/6 00:00:00 grep demo1

$ wrk -t2 -c 100 -d 60s http://127.0.0.1:8000

Running 1m test @ http://127.0.0.1:8000

2 threads and 100 connections

Thread Stats Avg Stdev Max +/- Stdev

Latency 4.13ms 3.38ms 112.09ms 88.85%

Req/Sec 13.00k 1.44k 19.86k 71.58%

1554274 requests in 1.00m, 200.11MB read

Requests/sec: 25859.47

Transfer/sec: 3.33MB