安装wrk压测工具

git clone https://github.com/wg/wrk.git
make

计算机环境

python3.6.8
tornado==6.0.2
sanic==19.3.1
cpu Intel(R) Core(TM) i5-8250U CPU @ 1.60GHz
测试语句：

wrk -t10 -d1m -c200 http://127.0.0.1:8080

1. 测试python标准库自带的HttpServer

from http.server import BaseHTTPRequestHandler
from urllib import parse


class GetHandler(BaseHTTPRequestHandler):
    def do_GET(self):
        message = "Hello World"
        self.send_response(200)
        self.end_headers()
        self.wfile.write(message.encode('utf-8'))

if __name__ == '__main__':
    from http.server import HTTPServer
    server = HTTPServer(('localhost', 8080), GetHandler)
    server.serve_forever()

wrk -t10 -d1m -c200 http://127.0.0.1:8080
结果

Running 1m test @ http://127.0.0.1:8080
  10 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     6.25ms   61.93ms   1.67s    99.04%
    Req/Sec   556.71    549.62     1.76k    67.14%
  4261 requests in 1.01m, 424.44KB read
  Socket errors: connect 0, read 4261, write 0, timeout 0
Requests/sec:     70.28
Transfer/sec:      7.00KB

每秒请求70

2. 测试sanic

from sanic import Sanic
from sanic.response import json, text
import asyncio

app = Sanic()

@app.route("/")
async def test(request):
    asyncio.sleep(1)
    return json({"message": "hello world"})

if __name__ == "__main__":
    app.run(host="0.0.0.0", debug=False, port=8080, access_log=False, workers=1)

这里测试一个进程也就是一个cpu的，测试结果

Running 1m test @ http://127.0.0.1:8080
  10 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    10.58ms    2.46ms  48.66ms   86.05%
    Req/Sec     1.90k   343.11     4.43k    74.63%
  1136502 requests in 1.00m, 146.32MB read
Requests/sec:  18915.42
Transfer/sec:      2.44MB

sanic每秒请求大概1.9w

3. 测试tornado

import tornado.ioloop
import tornado.web

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write("Hello, world")

def make_app():
    return tornado.web.Application([
        (r"/", MainHandler),
    ])

if __name__ == "__main__":
    app = make_app()
    app.listen(8080)
    tornado.ioloop.IOLoop.current().start()

tornado并没有返回json了，直接测试性能，测试结果：

Running 1m test @ http://127.0.0.1:8080
  10 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    66.14ms    9.37ms 159.04ms   91.26%
    Req/Sec   304.24     88.16   600.00     30.50%
  181609 requests in 1.00m, 35.85MB read
Requests/sec:   3021.95
Transfer/sec:    610.88KB

每秒请求3k

4 测试go标准库net/http

package main

import (
    "encoding/json"
    "fmt"
    "log"
    "net/http"
    "runtime"
    "time"
)

type Hello struct {
}

type Msg struct {
    Message string
}

func (h Hello) ServeHTTP(w http.ResponseWriter, r *http.Request) {
    time.Sleep(1)
    msg := Msg{Message: "Hello world"}
    s, _ := json.Marshal(msg)
    fmt.Fprint(w, string(s))
}

func main() {
    h := Hello{}
    runtime.GOMAXPROCS(1)
    err := http.ListenAndServe("localhost:8080", h)
    if err != nil {
        log.Fatal(err)
    }
}

这里限制核心数为1核，和sanic平级。结果如下

Running 1m test @ http://127.0.0.1:8080
  10 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     6.14ms   17.53ms 343.37ms   98.64%
    Req/Sec     4.54k   821.84    18.65k    80.78%
  2713334 requests in 1.00m, 367.44MB read
Requests/sec:  45151.63
Transfer/sec:      6.11MB

每秒4.5w，其实如果sanic改为多进程，go也改为没有cpu限制，他们之间还是会相差2倍。

5 测试node.js

var http = require('http');

http.createServer(function (request, response) {

    // 发送 HTTP 头部 
    // HTTP 状态值: 200 : OK
    // 内容类型: text/plain
    response.writeHead(200, {'Content-Type': 'text/plain'});

    // 发送响应数据 "Hello World"
    response.end('Hello World\n');
}).listen(8080);

Running 1m test @ http://127.0.0.1:8080
  10 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     5.86ms    1.70ms  85.42ms   95.21%
    Req/Sec     3.46k   494.21     5.75k    89.76%
  2065598 requests in 1.00m, 307.31MB read
Requests/sec:  34378.55
Transfer/sec:      5.11MB

每秒3.4w左右，由于node.js使用的也是libuv,但是python本身语言还是慢了。或者说，Sanic是一个框架，逻辑也比go的http和node的http模块逻辑要复杂的多，执行时间自然足药更多的时间。

uvloop

sanic是使用的uvloop作为asyncio的loop实现，uvloop本身非常快，他是Cython写的。如果使用c语言写的http解析器，那么它实现的http的性能和node.js持平, 和go也有的拼。
https://magic.io/blog/uvloop-blazing-fast-python-networking/