cancellable-http-client 1.0

A tiny, dependency-free HTTP client for Python with cancellable in-flight requests and hard wall-clock timeout.

cancellable_http_client

A tiny, dependency-free HTTP client for Python with cancellable in-flight requests and hard wall-clock timeout.

• Standard library only — no requests, no httpx, no urllib3.
• Single file, ~300 lines.
• Synchronous API that plays well with threading-based workers.
• Safe close() from any thread, at any time, including mid-transfer.
• Hard wall-clock timeout that bounds the entire request.

Why it exists

Python has no clean way to interrupt a thread that is blocked on a socket read. concurrent.futures.Future.cancel() does nothing once the task has started, and requests / urllib.urlopen() give you no handle to abort an in-flight request.

The one primitive that does work is closing the underlying socket: any pending recv() immediately unblocks with an error. This module wraps that trick behind a tiny, boring API so you don't have to reinvent it — or worry about the lifecycle edge cases — every time you need it.

If your codebase is built around asyncio, you don't need this; use httpx and task.cancel() instead. This module targets the very real case where you have existing threaded code and you want one HTTP call in the middle of it to be cancellable, without rewriting everything to be async.

Usage

import time
import cancellable_http_client as client

req = client.Request("https://example.com/")
req.start()           # the actual TCP connection happens here
start = time.monotonic()
while not req.done:
    if time.monotonic() - start > 5:
        print("taking too long, aborting...")
        req.close()   # interrupts the request if it's still in-flight
    req.wait(0.1)     # wait a bit before checking again
if req.error:
    print(f"failed: {req.error}")
elif req.response and req.response.status == 200:
    print(req.response.body)
req.close()           # safe to call any time, even mid-flight

You can also use it as a context manager:

with client.Request("https://example.com/") as req:
    req.start()
    req.wait(timeout=5)
    ...
# close() is called automatically on exit

Sharing a thread pool

By default each Request spawns its own daemon thread. To reuse a pool instead, assign an Executor to the module-level attribute:

from concurrent.futures import ThreadPoolExecutor
client.executor = ThreadPoolExecutor(max_workers=8)

API

Request(url, method="GET", headers=None, body=b"", socket_timeout=30, timeout=None)

Construct a request. No network I/O happens here — connection failures are reported via error after start().

• socket_timeout — per-socket-operation timeout in seconds, passed to http.client.HTTPConnection.
• timeout — wall-clock limit in seconds for the entire request. Triggers close() automatically if the request is not done in time. None disables.
• start() — kick off the request. Non-blocking.
• wait(timeout=None) -> bool — block until the request finishes. Returns True on completion, False on timeout.
• close() — abort the request and release resources. Safe to call any time, from any thread, any number of times.
• done (property) — True once the request has finished (success, failure, or close).
• response — a Response object on success, otherwise None.
• error — the exception raised during the request, or None.

Response

A read-only, socket-free container exposing the same attributes as http.client.HTTPResponse:

• status, reason, version
• headers (an http.client.HTTPMessage)
• body (bytes, eagerly read)
• getheader(name, default=None), getheaders()

Robust timeout

Most Python HTTP clients set a per-socket-operation timeout (socket.settimeout). This leaves several gaps:

• Slow drip — a server that sends one byte every 29 seconds never triggers a 30-second socket timeout, yet the total transfer can take arbitrarily long.
• DNS resolution — socket.getaddrinfo() is a blocking C library call with no timeout parameter. Python cannot interrupt it.
• Total elapsed time — there is no built-in way to cap the wall-clock time of an entire request across connection, TLS handshake, sending, and receiving.

cancellable_http_client addresses this with two separate knobs:

Parameter	Scope	Default
socket_timeout	Per socket operation (connect, send, recv)	30 s
timeout	Wall-clock limit on the entire request	None (no limit)

When timeout fires it calls close(), which immediately unblocks any pending socket operation by closing the underlying connection. This gives you a hard upper bound on how long wait() will block — something that socket_timeout alone cannot guarantee.

Comparison with existing libraries

	cancellable_http_client	httpx	requests
Cancel an in-flight request from another thread	✅	⚠️ async, unreliable	⚠️ hacky
Hard wall-clock timeout on entire request	✅	⚠️ per-operation	⚠️ per-operation
Synchronous API	✅	✅ (also async)	✅
No third-party dependencies	✅	❌	❌
Line count	~300	thousands	thousands
Fits a threading-based worker	✅	❌	⚠️
Redirects, cookies, User-Agent	⚠️ manual	✅	✅

httpx is a good choice if you are already in an asyncio world, though task.cancel() on in-flight requests can leave the connection pool in a broken state. requests does not offer a reliable way to interrupt an in-flight call; Session.close() does not forcibly close active sockets (psf/requests#5633).

Limitations

This library is a thin wrapper around http.client and does not provide the high-level conveniences found in requests or httpx:

• No automatic redirect following
• No cookie management
• No default User-Agent header
• No HTTP/2 support

These are all http.client limitations, not restrictions added by this library. You can still handle them manually via the headers parameter.

Tests

python -m unittest discover -s tests -v

No third-party test dependencies. Tests use local throwaway servers (normal, slow, blackhole, mid-body disconnect) to exercise cancellation, timeout, and error paths without touching the network.

License

Copyright 2026 Sakilabo Corporation Ltd. Licensed under the Universal Permissive License v 1.0 (UPL-1.0).

---

References (for the eventual public release)

Background reading and related work collected during the design of this module. Useful as citations in the README or a launch blog post.

Prior art / closest existing work

• "TIL: Stopping Requests Mid Flight" — haykot.dev A blog post describing the "close the socket from another thread" trick as a personal discovery. The same underlying idea as this module, but kept as a snippet rather than packaged as a library.

• httpcore on PyPI the low-level HTTP engine behind httpx. Supports cancellation via async task cancellation; relies on asyncio or trio.

• HTTPX modern high-level HTTP client; cancellation is done via task.cancel() in an async context.

• asyncio-cancel-token a cancellation-token utility for asyncio-based code.

The underlying Python pain points

• Graceful exit from ThreadPoolExecutor when blocked on IO — discuss.python.org Ongoing discussion acknowledging that Python has no clean way to cancel a worker that is blocked on I/O. This module is effectively a targeted workaround for the HTTP-specific case.

• threading — Python docs Thread has no cancel() / interrupt(); cooperation via Event is the only sanctioned approach.

• Session.close() does not close underlying sockets — psf/requests#5633 Illustrates why "just use requests and close the session" is not a reliable answer.

• Unclosed socket in urllib when ftp request times out after connect — cpython#140691 A related stdlib lifecycle bug — background for why we deliberately take full control of the connection object.

Related stdlib primitives this module builds on

• http.client — Python docs the low-level HTTP protocol implementation we wrap.
• threading.Event — Python docs used internally to signal completion.