Py Async Patterns

Python Async Review

Use these rules only for async or concurrent I/O code. Prefer concrete event-loop or cancellation risks over general async style suggestions.

Review Rules

Rule: async-blocking-call

Impact: CRITICAL Applies when: Code inside async def performs sleeping, HTTP, database, filesystem, subprocess, or CPU-heavy work. Skip when: The operation is explicitly offloaded to a thread/process or is known non-blocking. Python: any Tools: ruff | project-configured Review signal: time.sleep, requests, sync database clients, or CPU loops appear inside async def.

Incorrect:

async def fetch_data(url: str) -> dict:
    time.sleep(1)
    return requests.get(url).json()

Correct:

async def fetch_data(url: str) -> dict:
    await asyncio.sleep(1)
    async with httpx.AsyncClient() as client:
        response = await client.get(url)
    return response.json()

Reason: Blocking calls stop the event loop and degrade every concurrent request or task.

Rule: async-missing-await

Impact: HIGH Applies when: Changed code calls an async function. Skip when: The coroutine is intentionally scheduled with asyncio.create_task and its lifecycle is handled. Python: any Tools: pyright | mypy | ruff | project-configured Review signal: A coroutine-returning function is called without await, create_task, gather, or equivalent scheduling.

Incorrect:

async def handler() -> dict:
    result = fetch_user()
    return {"user": result}

Correct:

async def handler() -> dict:
    result = await fetch_user()
    return {"user": result}

Reason: Calling an async function without awaiting or scheduling it returns a coroutine object and does not run the operation.

Rule: async-gather-failure-semantics

Impact: MEDIUM-HIGH Applies when: Code runs independent async operations concurrently. Skip when: Failing fast on the first exception is required and documented. Python: any Tools: none Review signal: asyncio.gather is used for independent batch work without deliberate exception handling.

Incorrect:

async def fetch_all(ids: list[str]) -> list[User]:
    return await asyncio.gather(*(fetch_user(user_id) for user_id in ids))

Correct:

async def fetch_all(ids: list[str]) -> list[User | Exception]:
    return await asyncio.gather(
        *(fetch_user(user_id) for user_id in ids),
        return_exceptions=True,
    )

Reason: Concurrent batch code needs explicit failure semantics: fail fast, collect partial failures, or cancel siblings deliberately.

Rule: async-cancellation-propagates

Impact: CRITICAL Applies when: Code catches broad exceptions inside async tasks or request handlers. Skip when: The code catches asyncio.CancelledError only to clean up and then re-raises. Python: any Tools: none Review signal: Broad except Exception/except BaseException around awaited work may swallow cancellation or hide task shutdown.

Incorrect:

async def worker() -> None:
    try:
        await run_forever()
    except BaseException:
        logger.exception("worker failed")

Correct:

async def worker() -> None:
    try:
        await run_forever()
    except asyncio.CancelledError:
        await cleanup()
        raise
    except Exception:
        logger.exception("worker failed")
        raise

Reason: Cancellation is control flow for async shutdown. Swallowing it can hang deployments and leak resources. References: https://docs.python.org/3/library/asyncio-task.html Checked: 2026-07-07 Expires: 2026-10-01

Rule: async-timeout-boundary

Impact: HIGH Applies when: Code awaits network, database, queue, subprocess, or external service calls. Skip when: The called client has a documented timeout configured at construction or service level. Python: any Tools: none Review signal: Awaited external calls have no timeout or cancellation boundary.

Incorrect:

async def load_profile(user_id: str) -> Profile:
    return await profile_client.fetch(user_id)

Correct:

async def load_profile(user_id: str) -> Profile:
    return await asyncio.wait_for(profile_client.fetch(user_id), timeout=2.0)

Reason: Unbounded awaits can pin request handlers and background workers indefinitely.

Sensitive Evidence Safety

If changed code or tool output reveals a suspected credential, token, private key, secret-bearing URL, or other sensitive value, do not quote or reproduce the value. Report only its existence and location. Treat filename and pattern checks as heuristic evidence, not proof that a repository is secret-free.

If the exposure appears credible, make it the first finding, stop lower-priority review, and recommend revocation or rotation. Never place sensitive values in reports, generated examples, or commit subjects or bodies.