bombsquad

Isolate OOM-prone functions in a separate process, raise exceptions on SIGKILL or segfault.

Methods that cause out-of-memory errors (OOM) or segfaults are killed by the OS immediately, crashing the interpreter without raising an exception.

To solve this problem, the bombsquad package provides the @bombsquad decorator. This runs the decorated function in a separate, isolated process. If that process is terminated by the OS (e.g. SIGKILL), bombsquad catches the exit code and raises a NonzeroExitcode exception in your main process.

Normal Python exceptions raised within the function are propagated back to the parent process transparently.

Installation

pip install bombsquad

Quick Start

1. Trapping a Crash

Protect your main application from dangerous functions.

import os
import signal
from bombsquad import bombsquad, NonzeroExitcode

@bombsquad
def risky_business():
    print("Running risky code...")
    # Simulate a hard crash (like an OOM killer)
    os.kill(os.getpid(), signal.SIGKILL)

try:
    risky_business()
except NonzeroExitcode as e:
    print(f"Caught crash! Process exited with code {e.exitcode}")

2. Propagating Exceptions

Standard exceptions work exactly as expected.

@bombsquad
def divide(a, b):
    return a / b

try:
    divide(10, 0)
except ZeroDivisionError:
    print("Caught division by zero from child process.")

Advanced Configuration

The decorator accepts arguments to tune performance and safety.

@bombsquad(start_method="spawn", backend="file")
def massive_processing(data):
    ...

Parameters

Choosing a backend

• "file" (Default, Recommended):

  • Mechanism: Serializes results to a temporary file on disk.
  • Pros: Supports unlimited return sizes (limited only by disk space).
  • Cons: Slower due to disk I/O.
  • Use for: Returning large arrays, datasets, or huge strings.

• "queue":

  • Mechanism: Uses multiprocessing.Queue (memory pipes).
  • Pros: Faster (in-memory transfer).
  • Cons: Limited to ~4GB payloads (Python pickle limit).
  • Use for: Returning small results (status codes, metrics, small dicts).

License

MIT