a strange situation about parallel requests #14642

walkinrain2008 · 2026-01-02T14:32:12Z

walkinrain2008
Jan 2, 2026

First Check

I added a very descriptive title here.
I used the GitHub search to find a similar question and didn't find it.
I searched the FastAPI documentation, with the integrated search.
I already searched in Google "How to X in FastAPI" and didn't find any information.
I already read and followed all the tutorial in the docs and didn't find an answer.
I already checked if it is not related to FastAPI but to Pydantic.
I already checked if it is not related to FastAPI but to Swagger UI.
I already checked if it is not related to FastAPI but to ReDoc.

Commit to Help

I commit to help with one of those options 👆

Example Code

def test():
    print("test func start ",threading.get_native_id())
    time.sleep(3)
    return "test func function finish!"

@app.get("/test/{p}",tags=["sysm"])
def tests(): 
    print("test receive web request! ")
    r = test()
    return "ok!"+r

@app.get("/test2/",tags=["sysm"])
def tests2(): 
    print("test2 receive web request! ")
    r = test()
    return "ok"+r

Description

use axios to sent 2 request

["a","b"].map(item =>{axios.get(`/test/${item}`).then(v=>console.log(v.data))})
["a","b"].map(item =>{axios.get(`/test2/`).then(v=>console.log(v.data))})

server log is

test receive web request!
test receive web request!
test func start  9936
test func start  16512
INFO:     127.0.0.1:60413 - "GET /test/a HTTP/1.1" 200 OK
INFO:     127.0.0.1:60412 - "GET /test/b HTTP/1.1" 200 OK
test2 receive web request!
test func start  9936
INFO:     127.0.0.1:60434 - "GET /test2/ HTTP/1.1" 200 OK
test2 receive web request!
test func start  9936
INFO:     127.0.0.1:60436 - "GET /test2/ HTTP/1.1" 200 OK

it's means :
when app.get have path arg , requests is parallel processing。
when app.get no path arg, requests are being responded to one by one。

In reality, the test function is a database update feature and cannot be cached. This means that every time the test is executed, the results are different, even though the parameters have not changed.

My design is:
When the first request for test2 is received, add test to background tasks to run, and immediately return 'The task has started running.'
When test is already running and a test2 request is received, return 'The task is already running.'

But the current situation: when multiple test2 requests are received, they end up running one after another in a queue.

I have looked for official documentation but could not find relevant technical details. I am seeking help here.

Operating System

Windows

Operating System Details

windows 10

FastAPI Version

0.128.0

Pydantic Version

2.12.4

Python Version

Python 3.13.1

Additional Context

No response

ceb10n · 2026-01-03T16:34:40Z

ceb10n
Jan 3, 2026

I’m not sure if I fully understood what you’re trying to accomplish, but I’ll try to help.

You cannot predict the order of the responses.

Even if you send the requests in a specific order, it is never guaranteed that the server will respond in the same order. In a real-world scenario, with multiple workers, servers, or pods, this becomes even harder to predict.

In your example, you are using a sync function, which means your code is executed in a separate thread pool so it doesn’t block the event loop. This thread pool is managed by anyio, but the actual scheduling and execution time are handled by the operating system.

You could use some kind of lock to achieve what you’re trying to do, but this approach would be limited to a single worker, server, or pod.

For this use case, it might make more sense to move the processing to a worker using redis or another message broker.

9 replies

ceb10n Jan 7, 2026

I was able to test it on Windows 10, and I could reproduce it.

But unfortunately (or maybe fortunately), it’s not related to fastapi. As I mentioned before, your requests are handled by fastapi -> starlette -> anyio -> asyncio / threadpool -> os scheduler.

I asked chatgpt to convert your app into starlette, pure uvicorn, and pure python + asyncio, and the issue occurs in all cases. You can try it:

pure python

import asyncio
import threading
import time
import os
import mimetypes

HOST = "0.0.0.0"
PORT = 9000


def log_access(writer, method, path, status):
    peer = writer.get_extra_info("peername")
    if peer:
        ip, port = peer
    else:
        ip, port = "?", "?"

    reason = {
        200: "OK",
        404: "Not Found",
    }.get(status, "")

    print(
        f'INFO:     {ip}:{port} - "{method} {path} HTTP/1.1" {status} {reason}'
    )


async def test_a():
    print("test_a start in", threading.get_native_id())
    await asyncio.sleep(3)
    return "test func function finish!"


def test_s():
    print("test_s start", threading.get_native_id())
    time.sleep(3)
    return "test func function finish!"


def http_response(body, status=200, content_type="text/plain"):
    if isinstance(body, str):
        body = body.encode()

    return (
        f"HTTP/1.1 {status} OK\r\n"
        f"Content-Type: {content_type}; charset=utf-8\r\n"
        f"Content-Length: {len(body)}\r\n"
        f"Connection: close\r\n"
        f"\r\n"
    ).encode() + body


def http_404():
    body = b"Not Found"
    return (
        b"HTTP/1.1 404 Not Found\r\n"
        b"Content-Type: text/plain\r\n"
        b"Content-Length: 9\r\n"
        b"Connection: close\r\n"
        b"\r\n"
        b"Not Found"
    )


def serve_file(path):
    if not os.path.exists(path):
        return http_404()

    with open(path, "rb") as f:
        body = f.read()

    mime, _ = mimetypes.guess_type(path)
    mime = mime or "application/octet-stream"

    return (
        f"HTTP/1.1 200 OK\r\n"
        f"Content-Type: {mime}\r\n"
        f"Content-Length: {len(body)}\r\n"
        f"Connection: close\r\n"
        f"\r\n"
    ).encode() + body


async def handle_client(reader, writer):
    status = 500  # default

    try:
        data = await reader.read(1024)
        if not data:
            return

        request = data.decode(errors="ignore")
        request_line = request.splitlines()[0]
        method, path, protocol = request_line.split()

        if method == "GET" and path == "/":
            response = serve_file("test.html")
            status = 200

        elif method == "GET" and path.startswith("/ta/"):
            print("ta_p receive web request!")
            r = await test_a()
            response = http_response("ok!" + r)
            status = 200

        elif method == "GET" and path == "/ta2/":
            print("ta2 receive web request!")
            r = await test_a()
            response = http_response("ok" + r)
            status = 200

        elif method == "GET" and path.startswith("/ts/"):
            print("ts_p receive web request!")
            r = await asyncio.to_thread(test_s)
            response = http_response("ok!" + r)
            status = 200

        elif method == "GET" and path == "/ts2/":
            print("ts2 receive web request!")
            r = await asyncio.to_thread(test_s)
            response = http_response("ok" + r)
            status = 200

        else:
            response = http_404()
            status = 404

        writer.write(response)
        await writer.drain()

        # 🔥 Uvicorn-style log
        log_access(writer, method, path, status)

    except Exception as e:
        print("ERROR:", e)

    finally:
        writer.close()
        await writer.wait_closed()


async def main():
    server = await asyncio.start_server(handle_client, HOST, PORT)
    print(f"Serving on http://{HOST}:{PORT}")

    async with server:
        await server.serve_forever()


if __name__ == "__main__":
    asyncio.run(main())

pure uvicorn

import asyncio
import threading
import time
import os
import mimetypes


async def test_a():
    print("test_a start in", threading.get_native_id())
    await asyncio.sleep(3)
    return "test func function finish!"


def test_s():
    print("test_s start", threading.get_native_id())
    time.sleep(3)
    return "test func function finish!"


async def send_text(send, text, status=200):
    await send({
        "type": "http.response.start",
        "status": status,
        "headers": [
            (b"content-type", b"text/plain; charset=utf-8"),
        ],
    })
    await send({
        "type": "http.response.body",
        "body": text.encode("utf-8"),
    })


async def send_file(send, filepath):
    if not os.path.exists(filepath):
        await send_text(send, "File not found", 404)
        return

    mime, _ = mimetypes.guess_type(filepath)
    mime = mime or "application/octet-stream"

    with open(filepath, "rb") as f:
        body = f.read()

    await send({
        "type": "http.response.start",
        "status": 200,
        "headers": [
            (b"content-type", mime.encode()),
            (b"content-length", str(len(body)).encode()),
        ],
    })
    await send({
        "type": "http.response.body",
        "body": body,
    })


async def app(scope, receive, send):
    if scope["type"] != "http":
        return

    path = scope["path"]

    if path == "/" and scope["method"] == "GET":
        await send_file(send, "test.html")
        return

    if path.startswith("/ta/") and scope["method"] == "GET":
        print("ta_p receive web request!")
        r = await test_a()
        await send_text(send, "ok!" + r)
        return

    if path == "/ta2/" and scope["method"] == "GET":
        print("ta2 receive web request!")
        r = await test_a()
        await send_text(send, "ok" + r)
        return

    if path.startswith("/ts/") and scope["method"] == "GET":
        print("ts_p receive web request!")
        r = await asyncio.to_thread(test_s)
        await send_text(send, "ok!" + r)
        return

    if path == "/ts2/" and scope["method"] == "GET":
        print("ts2 receive web request!")
        r = await asyncio.to_thread(test_s)
        await send_text(send, "ok" + r)
        return

    await send_text(send, "Not Found", 404)


if __name__ == "__main__":
    import uvicorn
    uvicorn.run("test:app", host="0.0.0.0", port=9000)

starlette

from starlette.applications import Starlette
from starlette.responses import FileResponse, PlainTextResponse
from starlette.routing import Route

import threading
import asyncio
import time


async def show_root(request):
    """
    / 显示页面
    """
    return FileResponse("test.html")


async def test_a():
    print("test_a start in ", threading.get_native_id())
    await asyncio.sleep(3)
    return "test func function finish!"


async def ta(request):
    print("ta_p receive web request!")
    r = await test_a()
    return PlainTextResponse("ok!" + r)


async def ta2(request):
    print("ta2 receive web request!")
    r = await test_a()
    return PlainTextResponse("ok" + r)


def test_s():
    print("test_s start ", threading.get_native_id())
    time.sleep(3)
    return "test func function finish!"


def ts(request):
    print("ts_p receive web request!")
    r = test_s()
    return PlainTextResponse("ok!" + r)


def ts2(request):
    print("ts2 receive web request!")
    r = test_s()
    return PlainTextResponse("ok" + r)


routes = [
    Route("/", show_root, methods=["GET"]),
    Route("/ta/{p}", ta, methods=["GET"]),
    Route("/ta2/", ta2, methods=["GET"]),
    Route("/ts/{p}", ts, methods=["GET"]),
    Route("/ts2/", ts2, methods=["GET"]),
]

app = Starlette(
    debug=True,
    routes=routes
)

if __name__ == "__main__":
    import uvicorn
    uvicorn.run("test:app", host="0.0.0.0", port=9000)

In all cases, the same behavior occurs.

walkinrain2008 Jan 9, 2026
Author

Nothing else, but I was amazed at ChatGPT's transformation. Are these codes directly converted from chatGPT? Is it so practical?

A lot of AI code is written like a mentally retarded person.

It seems that this special situation is purely a problem of the win10 system.

Since Uvicorn is not very familiar, I tested it again with the more familiar asyncio. I found some interesting phenomena.

Post method, there is no such problem.
The get method, as long as it is in the path, has different characters, or different queries, there is no problem, and it can be kept concurrent.
Path and query are the same, even if the variable is put into the headers, it is still queued execution.

ps: If you access the server from the external network, the continuous '/ta2/', '/ts2/' requests will be discarded, and only one will be received!

This phenomenon, no data has been found to describe this information.
At first, I thought it was related to asyncio's EventLoop Class and loop's executor. Changing to selector and increasing the max-workers of executor has no effect.

Well, this is beyond my capabilities. System-level questions, it is estimated that only stackoverflow in its heyday can find the answer!
My English is so poor that if you're interested in posting this question to discuss on stackoverflow, please give me a link so I can learn it.

walkinrain2008 Jan 9, 2026
Author

thanks for your help!

ceb10n Jan 9, 2026

About chatgpt, I just pasted your code and asked it to convert it, no special prompts, etc.

Unfortunately, I wasn't logged in, so I don't have the history to show you 😢

walkinrain2008 Jan 10, 2026
Author

serve is correct, it's a browser issue.
I sent the request using the console in the Chromium Edge developer tools.
I went to Stack Overflow, and luckily, I guessed a few keywords for the search and found a related question from three years ago.
Among the many answers, I found a clue. The reason for this phenomenon is a browser issue. Any browser with the Chromium core may have this problem.

Conclusion:
Chromium will intercept the same GET requests and try to wait for the results to establish a cache. When connecting to 127 local, it may allow multiple requests to execute one by one because of the local client-server setup. When calling remotely, only one request remains.
At the same time, I rewrote the consecutive requests using Python and Node.js and got the expected results.

Stack Overflow really lives up to its reputation!

luzzodev · 2026-01-05T10:24:48Z

luzzodev
Jan 5, 2026
Collaborator

Hello @walkinrain2008, i have run your mre but i m unable to reproduce your issue.

test receive web request! 
test func start  1644
test receive web request! 
test func start  11364
test2 receive web request! 
test func start  11856
test2 receive web request! 
test func start  26748
INFO:     127.0.0.1:64508 - "GET /test/a HTTP/1.1" 200 OK
INFO:     127.0.0.1:64509 - "GET /test/b HTTP/1.1" 200 OK
INFO:     127.0.0.1:64511 - "GET /test2/ HTTP/1.1" 200 OK
INFO:     127.0.0.1:64510 - "GET /test2/ HTTP/1.1" 200 OK

8 replies

luzzodev Jan 6, 2026
Collaborator

Rerun your MRE multiple times with windows, linux and on mac. Used both firefox and chrome and also a python script and i'm unable to reproduce your issue.
Have your tried on another device?

walkinrain2008 Jan 9, 2026
Author

thanks for your help!

It seems that this special situation is purely a problem of the win10 system.

walkinrain2008 Jan 10, 2026
Author

I want to know how you send requests. Node.js? Chrome DevTools? What's your Chrome version number?

walkinrain2008 Jan 10, 2026
Author

The operating system has no issues!
Server is has no issues!
it's a browser issue!
I sent the request using the console in the Chromium Edge developer tools.

Have you encountered the same problem using Chrome?

walkinrain2008 Jan 10, 2026
Author

The operating system has no issues!
Server is has no issues!
it's a browser issue!
I sent the request using the console in the Chromium Edge developer tools.

Have you encountered the same problem using Chrome?

Uh oh!

a strange situation about parallel requests #14642

Uh oh!

walkinrain2008 Jan 2, 2026

First Check

Commit to Help

Example Code

Description

Operating System

Operating System Details

FastAPI Version

Pydantic Version

Python Version

Additional Context

Replies: 2 comments · 17 replies

Uh oh!

ceb10n Jan 3, 2026

Uh oh!

ceb10n Jan 7, 2026

pure python

pure uvicorn

starlette

Uh oh!

walkinrain2008 Jan 9, 2026 Author

Uh oh!

walkinrain2008 Jan 9, 2026 Author

Uh oh!

ceb10n Jan 9, 2026

Uh oh!

walkinrain2008 Jan 10, 2026 Author

Uh oh!

luzzodev Jan 5, 2026 Collaborator

Uh oh!

luzzodev Jan 6, 2026 Collaborator

Uh oh!

walkinrain2008 Jan 9, 2026 Author

Uh oh!

Uh oh!

walkinrain2008 Jan 10, 2026 Author

Uh oh!

walkinrain2008 Jan 10, 2026 Author

Uh oh!

walkinrain2008 Jan 10, 2026 Author

walkinrain2008
Jan 2, 2026

Replies: 2 comments 17 replies

ceb10n
Jan 3, 2026

walkinrain2008 Jan 9, 2026
Author

walkinrain2008 Jan 9, 2026
Author

walkinrain2008 Jan 10, 2026
Author

luzzodev
Jan 5, 2026
Collaborator

luzzodev Jan 6, 2026
Collaborator

walkinrain2008 Jan 9, 2026
Author

walkinrain2008 Jan 10, 2026
Author

walkinrain2008 Jan 10, 2026
Author

walkinrain2008 Jan 10, 2026
Author