Cancel Async Tasks

Async Task Cancellation with Cleanup

This skill provides guidance for implementing proper cancellation handling in Python asyncio code, with particular focus on keyboard interrupt (Ctrl+C) handling and graceful cleanup.

Core Concepts

KeyboardInterrupt vs CancelledError

Understanding the difference is critical:

• KeyboardInterrupt: A BaseException raised in the main thread when Ctrl+C is pressed. It does NOT propagate into async coroutines directly.
• CancelledError: An Exception raised inside coroutines when task.cancel() is called. This IS what coroutines receive during cancellation.

Critical Insight: Placing except KeyboardInterrupt inside an async function will NOT catch Ctrl+C. The interrupt is raised at the event loop level, not within coroutines.

Signal Handling in asyncio

To properly handle Ctrl+C in asyncio:

1. Use loop.add_signal_handler() to register SIGINT/SIGTERM handlers
2. Or wrap asyncio.run() in a try/except at the synchronous entry point
3. The signal handler should trigger task cancellation, which then raises CancelledError in coroutines

Potential Approaches

Approach 1: Signal Handler Pattern

Register signal handlers at the event loop level to trigger graceful shutdown:

1. Get the event loop
2. Add signal handlers for SIGINT and SIGTERM
3. In the signal handler, cancel all running tasks
4. In coroutines, catch CancelledError to perform cleanup

Approach 2: Synchronous Wrapper Pattern

Wrap the async entry point to catch KeyboardInterrupt synchronously:

1. Define the async main logic
2. In a synchronous wrapper, call asyncio.run() inside try/except
3. Catch KeyboardInterrupt in the synchronous code
4. Handle cleanup or re-raise as needed

Approach 3: TaskGroup with Exception Handling (Python 3.11+)

Use asyncio.TaskGroup for structured concurrency:

1. Use async with asyncio.TaskGroup() for task management
2. TaskGroup handles cancellation propagation automatically
3. Catch ExceptionGroup for handling multiple failures

Verification Strategies

Test with Actual Signals

Do NOT rely solely on timeout-based cancellation testing. Verify with actual signals:

import os
import signal

# Simulate Ctrl+C in tests
os.kill(os.getpid(), signal.SIGINT)

Verify Cleanup Execution

Create observable side effects to confirm cleanup runs:

• Write to a file during cleanup
• Set a flag variable
• Log cleanup actions

Test Multiple Cancellation Scenarios

1. Normal completion (no cancellation)
2. Timeout-based cancellation (asyncio.timeout)
3. Manual task.cancel() calls
4. Actual SIGINT signal
5. SIGTERM signal

Verify Exception Propagation

Test that exceptions from individual tasks are properly collected and reported, not silently swallowed.

Common Pitfalls

Pitfall 1: Catching KeyboardInterrupt in Async Functions

Wrong:

async def my_coroutine():
    try:
        await some_operation()
    except KeyboardInterrupt:  # Will NOT catch Ctrl+C!
        await cleanup()

Correct:

async def my_coroutine():
    try:
        await some_operation()
    except asyncio.CancelledError:  # This WILL be raised on cancellation
        await cleanup()
        raise  # Re-raise to propagate cancellation

Pitfall 2: Not Re-raising CancelledError

Swallowing CancelledError prevents proper cancellation propagation. Always re-raise after cleanup unless there's a specific reason not to.

Pitfall 3: Testing Only Happy Path Cancellation

Testing with asyncio.timeout() or asyncio.wait_for() does NOT verify keyboard interrupt handling. These trigger TimeoutError, not the same path as SIGINT.

Pitfall 4: Ignoring Cleanup Cancellation

Cleanup code itself can be cancelled if it awaits. Use asyncio.shield() for critical cleanup:

async def cleanup():
    await asyncio.shield(critical_cleanup_operation())

Pitfall 5: Empty or Invalid Input Handling

Always validate inputs:

• Empty task lists
• Invalid concurrency limits (max_concurrent <= 0)
• None values

Pitfall 6: Assuming gather Handles All Exceptions

asyncio.gather(return_exceptions=True) collects exceptions but doesn't handle them. Results must be inspected for exception instances.

Edge Cases to Consider

1. Empty task list: Return early with empty results
2. max_concurrent <= 0: Raise ValueError or use sensible default
3. All tasks fail: Ensure all exceptions are reported
4. Partial completion: Track which tasks completed vs cancelled
5. Nested cancellation: Cleanup code getting cancelled
6. Multiple rapid signals: Debounce or ignore duplicate signals
7. Windows compatibility: add_signal_handler doesn't work on Windows for SIGINT; use alternative approaches

Platform Considerations

Unix/Linux/macOS

• loop.add_signal_handler() works for SIGINT, SIGTERM
• Signal handlers run in the main thread

Windows

• add_signal_handler() is limited; only works for SIGINT in some cases
• Consider using signal.signal() before starting the event loop
• Or rely on the synchronous wrapper pattern