Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

Redesign AsyncioWaker around caller-context-independent state transitions

オープン
  • 何か問題がありますか？
#217 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
25/100
issue の種類
リファクタリング
明瞭さ
説明が足りない
活発さ
活発
技術スタック
rust
領域
backend, testing-qa

調査の方向性

Start with src/future/asyncio/waker.rs and the runtime attachment check in src/utils.rs, then run the subprocess regressions in python/tests/test_deadlocks.py. Agree on the state-machine design before implementation; done means the listed wake, callback, shutdown, finalization, and subprocess tests define correct behavior without public Python API changes.

索引モデルが issue の本文から書いたものです。

説明

enhancement

Problem and proposed change

AsyncioWaker connects Rust future wakeups to an asyncio.Future. A mutex-protected slot records whether the coroutine is idle, was woken before parking, or is parked on a Python future.

A Rust future can wake:

  • synchronously while Python is polling it; or
  • later from a detached Tokio worker.

The standard Wake trait does not report which context called it. Current code therefore detects attachment at runtime. Parking also invokes Python event-loop methods while holding the slot mutex. These assumptions make lock ordering, shutdown, and interpreter finalization difficult to reason about.

Redesign the state machine so the mutex protects only short state transitions and ownership moves. Release the mutex before calling Python or destroying Python-owned values. Make wake_by_ref work without detecting caller attachment.

Agree on the design first, then implement it separately.

Context

Current slot states and parking path:

https://github.com/scylladb/python-rs-driver/blob/ef53b6c01f174add44c87c184740a02eadf51d7f/src/future/asyncio/waker.rs#L39-L110

Current explicit wake methods and mixed-context Wake implementation:

https://github.com/scylladb/python-rs-driver/blob/ef53b6c01f174add44c87c184740a02eadf51d7f/src/future/asyncio/waker.rs#L113-L172

Current runtime attachment check:

https://github.com/scylladb/python-rs-driver/blob/ef53b6c01f174add44c87c184740a02eadf51d7f/src/utils.rs#L308-L323

Existing subprocess regressions:

https://github.com/scylladb/python-rs-driver/blob/ef53b6c01f174add44c87c184740a02eadf51d7f/python/tests/test_deadlocks.py#L27-L132

The same-thread re-entry report #212 was closed as not planned. This proposal is broader structural hardening, not an automatic reopening:

https://github.com/scylladb/python-rs-driver/issues/212

An unconditional Python attachment on every worker wake could hurt performance. Related investigation:

https://github.com/scylladb/python-rs-driver/issues/179

Acceptance criteria

  • The slot guard protects only state transitions and ownership moves.
  • Python calls and Python-owned-value destruction occur after unlocking.
  • wake_by_ref does not need to detect caller attachment.
  • Wake before, during, and after parking remains correct.
  • Tests cover synchronous self-wake and detached worker wake.
  • Tests cover callback completion, close, cancel, and exception delivery.
  • Task destruction and runtime shutdown behavior are defined.
  • Interpreter-finalization behavior is defined if any fallback attaches.
  • Potential hangs run in subprocesses with external watchdogs.
  • Tests detect lost and duplicate wakeups.
  • The worker fast path does not gain unconditional Python attachment without an explicit performance decision.
  • No public Python API changes.
主要言語
Rust
スター
6
フォーク
8
平均マージ
5日 9時間
マージ済み PR（30日）
14

環境構築

  • Dockerfile または Docker Compose ファイルあり
  • プルリクエストのテンプレートあり
  • コントリビューションガイドなし

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

scylladb/python-rs-driver のほかの issue

scylladb/python-rs-driver の issue をすべて見る

似ている issue

Rust の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。