Python signal handlers: install from the main interpreter thread

Operation contract

A worker attempts to replace the SIGINT handler and receives ValueError. The main thread installs an owned handler and immediately restores the previous handler. The program never sends SIGINT and does not change process state after completion.

Failure boundary

Python runs signal handlers in the main thread, but signal delivery timing is not a transaction boundary. Do not acquire an ordinary lock or perform complex cleanup inside a handler. Convert a signal into a simple shutdown request and let normal control flow close resources. Signal availability differs by operating system; the fixture checks registration only.

Working program

import signal
from threading import Thread

observed = []

def request_shutdown(signal_number, frame):
    observed.append(signal_number)

def worker_registration():
    try:
        signal.signal(signal.SIGINT, request_shutdown)
    except ValueError:
        observed.append("worker_rejected")

worker = Thread(target=worker_registration)
worker.start()
worker.join()
previous = signal.signal(signal.SIGINT, request_shutdown)
signal.signal(signal.SIGINT, previous)
print("worker_rejected", "worker_rejected" in observed)
print("main_restored", signal.getsignal(signal.SIGINT) == previous)

Output

worker_rejected True
main_restored True

Costs and limits

Registration is constant-time; handler design matters more than installation cost. A process shutdown still needs bounded worker joins and explicit resource owners.

Common Mistakes

• Do not register Python handlers from a worker thread.
• Do not run lock-heavy application work inside a handler.
• Restoring a previous handler is necessary in a reusable process.

Connected lessons

• Python locks: protect the complete inventory transition
• Python Queue.shutdown: drain accepted work before workers exit
• Python subprocess timeout: kill and reap the timed-out child

Test this contract.