aeondave/python-async-patterns

Async Python Patterns

This skill focuses on practical asyncio patterns for I/O-bound concurrency.

When to activate

• You’re building an async service/client (HTTP, DB, queues, websockets)
• You need concurrency with limits (rate limiting, semaphores)
• You need safe cancellation and timeouts
• You suspect event loop blocking (sync call inside async path)

Rules of engagement

• Prefer async only for I/O-bound workloads.
• Never block the event loop (no time.sleep(), no sync HTTP/DB in async code).
• Make cancellation and timeouts explicit.
• Bound concurrency; unbounded gather() can turn memory into a queue.

---

Outcome expectations

• Concurrent I/O tasks are orchestrated with clear boundaries and failure semantics.
• Cancellation and timeouts are explicit and tested.
• Event loop is never blocked by sync calls; backpressure prevents unbounded growth.

---

Recommended workflow

1. Define scope and concurrency bounds before writing async code.
2. Use TaskGroup for orchestration; avoid fire-and-forget patterns.
3. Apply timeouts at I/O boundaries, not broad scopes.
4. Test cancellation paths; use pytest-asyncio with care for shared state.
5. Profile event loop blocking; offload sync work via to_thread() when necessary.

---

Quick patterns

Concurrent fan-out with bounds (TaskGroup preferred)

Prefer asyncio.TaskGroup (Python 3.11+) for structured concurrency with clear failure propagation.

async with asyncio.TaskGroup() as tg:
    tg.create_task(fetch_url(url1))
    tg.create_task(fetch_url(url2))
# All tasks joined; exceptions aggregated

For concurrency limits, add a semaphore:

sem = asyncio.Semaphore(10)
async def bounded():
    async with sem:
        return await fetch_url(url)

Timeouts

• Prefer asyncio.timeout() (3.11+) for scoped timeouts.

Cancellation

• Catch asyncio.CancelledError only to clean up, then re-raise.

Sync/async boundary

• Offload truly blocking work via asyncio.to_thread().

Resources

Load on demand:

• references/foundations.md — event loop, coroutines vs tasks, TaskGroup vs gather
• references/cancellation-timeouts.md — cancellation semantics and timeout patterns
• references/backpressure-rate-limit.md — queues, semaphores, producer/consumer, rate limiting
• references/sync-async-interop.md — to_thread, executors, avoiding hidden blocking
• references/testing.md — testing async code patterns (pytest-asyncio) and flake avoidance