如何同时向 FastAPI 端点发送大量请求？

Question

使用如下所示的“正常”协程，结果是首先打印所有请求，然后在大约 5 秒后打印所有响应：

import asyncio

async def request():
    print('request')
    await asyncio.sleep(5)
    print('response')

loop = asyncio.get_event_loop()

tasks = [
    loop.create_task(request())
    for i in range(30)
]
loop.run_until_complete(asyncio.wait(tasks))
loop.close()

我想在 FastApi 中复制相同的行为，所以我有这样的端点：

import asyncio

from fastapi import FastAPI

app = FastAPI()

@app.post("/")
async def root():
    print('request')
    await asyncio.sleep(5)
    return 'OK'

我用来自前端的多个请求轰炸它：

 const url = 'http://localhost:8000'
 const data = [1, 2, 3]
 const options = {
     method: 'POST',
     headers: new Headers({'content-type': 'application/json'}),
     body: JSON.stringify({data}),
     mode: 'no-cors',
 }

 for (let i=0; i<30; i++) {
     fetch(url, options)
 }

然而，在终端中我可以清楚地看到 FastAPI 一次只接受 6 个请求，returns 响应它们，然后接受另外 6 个：

request
request
request
request
request
request
←[32mINFO←[0m:     127.0.0.1:63491 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:58337 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:50479 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:60499 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:56990 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:56107 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
request
request
request
request
request
request
←[32mINFO←[0m:     127.0.0.1:58337 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:63491 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:60499 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:56990 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:56107 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m
←[32mINFO←[0m:     127.0.0.1:50479 - "←[1mPOST / HTTP/1.1←[0m" ←[32m200 OK←[0m

等等

这是因为某些 FastAPI/uvicorn 设置或限制吗？

能不能增加，是否合理？

Answer 1

您在浏览器中执行此操作，因此您实际上达到了浏览器中的并行请求限制。它与 API 本身无关。如果您想测试 API 的性能，请使用专门为此设计的工具 - 例如 siege、httperf、ab 或类似工具。

来自the answer documenting the current parallel request limits in browsers：

Firefox 3+: 6
...
Edge:       6
Chrome:     6

如何同时向 FastAPI 端点发送大量请求？

How to send a lot of simultaneous requests to FastAPI endpoint?

asynchronous

fastapi

uvicorn