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

Silent self-deadlock with SetMaxOpenConns(1) when insert hooks use the pool

オープン
  • 何か問題がありますか？
#1,411 コメント 1 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
4/5
見積もり時間
3〜5日
初心者へのやさしさ
52/100
issue の種類
バグ
明瞭さ
おおむね明確
活発さ
活発
技術スタック
go, sqlite
領域
backend, database

調査の方向性

Start by running the provided Go reproduction with MaxOpenConns(1), then compare the client.Insert and InsertTx paths with hook and middleware execution. Review the SetMaxOpenConns guidance and relevant hook documentation before choosing a scoped direction. Done should include coverage for the deadlock scenario and a clear user-facing explanation or failure mode.

索引モデルが issue の本文から書いたものです。

説明

Note: this issue was written by Claude (an AI assistant) from an investigation it ran, and @bgentry reviewed it and okayed filing it. @brandur, could you take a look?

The riversqlite package docs recommend dbPool.SetMaxOpenConns(1) to avoid SQLITE_BUSY errors from River's parallel internal operations. With that setting, two fairly natural patterns hang silently until the context deadline, or forever if the context has none. database/sql has no way to detect the self-deadlock, so the user sees no error, only a stuck insert.

1. A hook or middleware that uses the pool during a non-transactional insert. client.Insert opens its own transaction, which takes the pool's only connection, and runs insert hooks and middleware inside it. Hooks and middleware don't receive that transaction, so a hook that needs the database has to go through the pool. It then waits for the connection its own insert is holding.

2. client.Insert (instead of InsertTx) while the app holds an open transaction on the same pool. River's BEGIN waits for the connection that the app's transaction holds.

Reproduction against master (9573203e) with riversqlite and modernc.org/sqlite, a file database in WAL mode:

type poolHook struct {
	river.HookDefaults
	db *sql.DB
}

func (h *poolHook) InsertBegin(ctx context.Context, params *rivertype.JobInsertParams) error {
	_, err := h.db.ExecContext(ctx, "SELECT 1")
	return err
}

// db.SetMaxOpenConns(n); client configured with Hooks: []rivertype.Hook{&poolHook{db: db}}
ctx, cancel := context.WithTimeout(ctx, 2*time.Second)
_, err := client.Insert(ctx, someArgs{}, nil)

MaxOpenConns(1) hook uses pool during client.Insert: err=context deadline exceeded after 2.001s
MaxOpenConns(4) hook uses pool during client.Insert: err=<nil> after 1ms
MaxOpenConns(1) client.Insert while app holds a tx: err=error beginning transaction: context deadline exceeded after 2s

This isn't strictly SQLite-specific. Any database/sql or pgx pool capped at one connection behaves the same way. It mostly bites SQLite users in practice, because that's where the docs recommend a single connection.

Some possible directions, not mutually exclusive:

  1. Document it. Next to the SetMaxOpenConns(1) advice, and in the hooks/middleware docs, explain that hook and middleware code runs while River holds a connection, and that app code inside its own transaction should use InsertTx.
  2. Fail fast on re-entry. River could mark the context while it holds its own insert transaction, so a re-entrant River call on the same client returns a clear error instead of hanging. This wouldn't catch a hook calling db.Exec directly on the raw pool, but it would cover River-level re-entry.
  3. Let hooks and middleware use River's transaction. Expose the insert's transaction to them, for example through the context, so they can do database work without a second connection. This is a bigger API change.

My suggestion is (1) plus (2), with (3) as a possible follow-up.

主要言語
Go
スター
5.7k
フォーク
187
平均マージ
1日 11時間
マージ済み PR（30日）
63

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

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。

riverqueue/river のほかの issue

riverqueue/river の issue をすべて見る

似ている issue

Go 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。