[Bug]: SSH reconnect fails for as long as the old remote server is stuck shutting down

オープン
#39,797 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
4/5
見積もり時間
3〜5日
初心者へのやさしさ
48/100
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
typescript

調査の方向性

Start with the reconnect launch script in packages/ssh/src/tunnel.ts, especially the managed_server_running check and wait_ready path. Trace how the remote server's runtime record and shutdown lifecycle are handled, including the mentioned releaseManagedTunnelOnShutdown path. Run the SSH reconnect tests and add coverage for a live server whose listener has closed; done means reconnect recovers or reports the previous process is still shutting down.

索引モデルが issue の本文から書いたものです。

説明

Version: desktop and remote runtime 0.1.11-nightly.20261118.4890. Remote is an arm64 Linux VM reached over SSH.

What happened

The remote t3 serve process received SIGTERM and began shutting down. It closed its HTTP listener, but it could not exit, because one of its worker threads was blocked in an uninterruptible filesystem wait (a slow fsync on the VM). For the next ~22 minutes, every reconnect from the desktop failed with:

Remote T3 server is still running but did not become ready. Retry the connection; the server was left running.

Each retry spent about 70 s inside launchOrReuseRemoteServer before failing. The connection came back only after the old process finally exited.

Cause

In packages/ssh/src/tunnel.ts, the launch script checks managed_server_running "$REMOTE_PID", which tests only kill -0 plus the process's command line. A server that is mid-shutdown passes that check. The script then runs wait_ready, which can never succeed because the listener is closed, and the script exits 1 with "the server was left running". Nothing in the reconnect path can tell "alive but slow" apart from "alive but shutting down", so the desktop retries the same failing path indefinitely.

Observed on the remote during the outage:

  • The pid was alive in state S. Its main thread was in epoll_pwait and one WorkerThread was in D (wait_for_commit).
  • curl http://127.0.0.1:3773/ was refused immediately (no listener).
  • The trace showed environment.cloud.releaseManagedTunnelOnShutdown and the root server.runServerCommand fibers being interrupted, so teardown had started but not finished.

Expected

The reconnect either recovers or reports what is actually wrong. Options:

  • The server marks its runtime record as stopping when teardown begins, and the launcher treats a stopping server as gone: it waits a bounded time, then starts a fresh server or fails with "the previous server (pid N) is still shutting down".
  • At minimum, when the port is refused (as opposed to timing out), the launcher should not say the server "did not become ready".

Created with Claude Opus 5.5 in Claude Code.

主要言語
TypeScript
スター
24.8k
フォーク
6.4k
平均マージ
4時間 29分
マージ済み PR(30日)
488

環境構築

Codespaces で開く

このプロジェクトの開発コンテナを、あなたの GitHub アカウントでブラウザ上に起動します。

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。

pingdotgg/t3code のほかの issue

pingdotgg/t3code の issue をすべて見る

似ている issue

TypeScript 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。