vp run hangs after a failure when a cancelled task's detached descendant holds its IPC connection
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 48/100
调研方向
Start with the cancellation path in crates/vt/src/session/execute/spawn.rs and the IPC client drain in crates/vt_server/src/lib.rs, both linked in the issue. Reproduce with the supplied vp run repro setup, then trace how cancellation and open descendant connections affect shutdown. Done means fail-fast exits promptly with a nonzero status while a detached descendant holds the IPC connection; the issue names no tests.
由索引模型根据 Issue 内容生成。
描述
Written by an AI agent (Claude) on behalf of @jasonkuhrt.
When a sibling task fails, vp run kills a cached task's direct child, then keeps running for as long as a detached descendant of that task holds its runner IPC connection. In the repro below, 20 seconds after the failure the task's process was dead, its detached child was alive, and the session was still running. Killing only the detached child let the session exit 1 at once.
We hit this with Playwright: its webServer starts vp dev in a detached process group, so when fail-fast SIGKILLs Playwright, the dev server survives with an open IPC connection and the session hangs until it is killed by hand (locally) or the job times out (CI).
Repro
package.json
{
"name": "ipc-hang-repro",
"private": true,
"type": "module",
"scripts": {
"fail": "node repro.cjs fail"
},
"devDependencies": {
"vite-plus": "1.3.0"
}
}
vite.config.ts
export default {
run: {
tasks: {
repro: {
command: 'echo done',
dependsOn: ['fail', 'cached'],
},
cached: {
command: 'node repro.cjs parent',
cache: { input: ['repro.cjs'], output: [] },
},
},
},
};
repro.cjs
const { spawn } = require('node:child_process');
const fs = require('node:fs');
switch (process.argv[2]) {
// Fails once the grandchild has connected.
case 'fail':
setInterval(() => {
if (fs.existsSync('ready.json')) {
console.log('FAIL: IPC connected; exiting 1');
process.exit(1);
}
}, 100);
break;
// The cached task: starts a detached child, then idles.
case 'parent': {
const child = spawn(process.execPath, [__filename, 'grandchild'], {
detached: true,
stdio: 'ignore',
});
child.unref();
setInterval(() => {}, 1000);
break;
}
// Opens a runner IPC connection, as `vp dev` does, and stays alive.
case 'grandchild': {
globalThis.client = require(process.env.VP_RUN_NODE_CLIENT_PATH).load();
globalThis.client.getEnv('PATH');
fs.writeFileSync(
'ready.json',
JSON.stringify({ parent: process.ppid, grandchild: process.pid }),
);
setInterval(() => {}, 1000);
break;
}
}
Run
bun install # any package manager
export PATH="$PWD/node_modules/.bin:$PATH"
rm -f ready.json
vp run repro > run.log 2>&1 &
runner=$!
sleep 20
kill -0 "$runner" && echo 'still running after 20s'
kill -KILL "$(node -p 'require("./ready.json").grandchild')"
wait "$runner"; echo "session exit=$?"
Observed
$ node repro.cjs fail ⊘ cache disabled
$ node repro.cjs parent
FAIL: IPC connected; exiting 1
still running after 20s
session exit=1
vp run: 0/1 cache hit (0%), 0/2 successful.
--last-details reports fail exit 1 and cached exit 137.
Expected
Fail-fast finishes promptly with a nonzero status, and the cancelled task's descendants don't keep the session alive.
Cause
At 9ba4cc9 (the commit Vite+ 1.3.0 pins):
- On Unix, cancellation SIGKILLs only the direct child (Windows terminates the job object).
- The IPC server then drains existing clients with no deadline, and a descendant that loaded the client from
VP_RUN_NODE_CLIENT_PATHkeeps its connection open. - Playwright's
webServerpassesprocess.envthrough and starts the server in a detached process group, and SIGKILL skips its teardown.
Possible fixes
- On cancellation, end the task's IPC client handlers, or bound the drain.
- Kill the task's process group or descendants (detached groups escape a group kill).
Related
- #822: keep going instead of cancelling sibling tasks.
- #485: on Windows, completion blocked by descendant-held stdio.
- #396: an earlier browser hang from descendant-held fspy locks.
Environment
macOS 27.0 (26A428), arm64; Vite+ 1.3.0; Node 24.22.1.
- 主要语言
- Rust
- 星标
- 480
- 派生
- 50
- 平均合并
- 1 天 3 小时
- 30 天内合并 PR
- 62
环境准备
在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
voidzero-dev/vite-task 的其他 Issue
-
难度 4/5 3-5 天 新手友好度 45/100
voidzero-dev/vite-task#888 ·
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 58/100
voidzero-dev/vite-task#886 ·
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 45/100
voidzero-dev/vite-task#884 ·
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 72/100
voidzero-dev/vite-task#881 ·
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 48/100
voidzero-dev/vite-task#879 ·
维护者通常 1 天内回复
查看 voidzero-dev/vite-task 的全部 Issue
相似的 Issue
-
`mise doctor` Shows Not Installed Lazy Tools as Problems可能已有人在做 @jdx 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 67/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 66/100
维护者通常 1 天内回复
-
nn::linear: weight initialization depends on `bias` because the bias is created before the weight可能已有人在做 @monfortema 今天认领。 未关闭
难度 2/5 1-3 小时 新手友好度 72/100
LaurentMazare/tch-rs#1067 ·
-
难度 2/5 1-3 小时 新手友好度 64/100
astral-sh/astral-codec#234 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 67/100