@suwenguang/pi-feishu-bridge 0.6.2 → 0.6.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.pi/feishu/debug.pi.log +1151 -81
- package/.pi/feishu/state.pi.json +1 -1
- package/README.md +44 -0
- package/package.json +3 -2
- package/scripts/restart-worker.mjs +233 -0
package/.pi/feishu/state.pi.json
CHANGED
package/README.md
CHANGED
|
@@ -209,3 +209,47 @@ cd ~/work/b && pi # 工作区 B:直接用(各自机器人凭证写在各
|
|
|
209
209
|
**两个工作区的机器人互相抢连接 / 消息落错工作区?** 看是不是两个实例用了同一个 `appId` —— 必须一个工作区一个自建应用。只有在显式设了 `FEISHU_DATA_DIR`、把两个工作区指到同一实例目录时才会撞车,此时「裸开保护」会拒绝自动启动并标注「连接属于其他工作区」。
|
|
210
210
|
|
|
211
211
|
**能开机自启动吗?** 暂未实现。目前需要电脑开机后手动启动一次 Pi agent,之后即可关闭前台,在飞书/Lark 里继续对话。
|
|
212
|
+
|
|
213
|
+
## 发布(维护者)
|
|
214
|
+
|
|
215
|
+
npm 包名 `@suwenguang/pi-feishu-bridge`,发布由 GitHub Actions 走 **OIDC 信任发布**,仓库里不放任何 token。
|
|
216
|
+
tag 一推就自动发,不需要任何人扫码。
|
|
217
|
+
|
|
218
|
+
### 发一个版本
|
|
219
|
+
|
|
220
|
+
```bash
|
|
221
|
+
cd .pi/git/gitee.com/suveng/pi-feishu-bridge
|
|
222
|
+
npm version patch --no-git-tag-version # 或 minor / 手改 package.json
|
|
223
|
+
git commit -am "chore(发布): 0.6.3"
|
|
224
|
+
git tag v0.6.3
|
|
225
|
+
git push origin main v0.6.3 # gitee
|
|
226
|
+
git push github main v0.6.3 # GitHub(触发 Actions)
|
|
227
|
+
```
|
|
228
|
+
|
|
229
|
+
成功征兆(三步都要看到,缺一条就是没发成):
|
|
230
|
+
|
|
231
|
+
1. https://github.com/suveng/pi-feishu-bridge/actions 里 `publish` 运行变绿,日志末尾有 `+ @suwenguang/pi-feishu-bridge@0.6.3`
|
|
232
|
+
2. 日志里出现 `Provenance statement published to transparency log`(OIDC 签名成功)
|
|
233
|
+
3. 等 1-2 分钟后 `npm view @suwenguang/pi-feishu-bridge version --registry=https://registry.npmjs.org` 返回新版本号
|
|
234
|
+
|
|
235
|
+
> 发布成功后 registry 还要 1-2 分钟才可见,期间 `npm view` 报 404 属正常(新包风控),不要重发。
|
|
236
|
+
|
|
237
|
+
### 也可以手动触发
|
|
238
|
+
|
|
239
|
+
Actions 页面 → `publish` → Run workflow(`workflow_dispatch`),条件同样是 `package.json` 的版本号还没发过。
|
|
240
|
+
|
|
241
|
+
### 前置配置(只在首次或改包名/改仓库时做一次)
|
|
242
|
+
|
|
243
|
+
1. **首次发布必须手动**:npm 不允许给「还不存在的包」预配 OIDC,第一个版本只能用 `npm login` + 2FA 手动 `npm publish --access public`。
|
|
244
|
+
非交互环境(脚本 / Agent)下 npm 会把 2FA 链接打码成 `***`,要用 pty 才能拿到真链接:`script -q /dev/null npm publish --access public`。
|
|
245
|
+
2. **Trusted Publisher**:https://www.npmjs.com/package/@suwenguang/pi-feishu-bridge/access
|
|
246
|
+
—— Publisher `GitHub Actions` / Organization `suveng` / Repository `pi-feishu-bridge` / Workflow filename `publish.yml` / Environment 留空。
|
|
247
|
+
四项逐字匹配,改工作流文件名就必须同步改这里,否则发布报 `E404 ... could not be found or you do not have permission`。
|
|
248
|
+
3. **仓库内 `.npmrc` 必须固定官方源**(`registry=https://registry.npmjs.org/`):本机全局 `.npmrc` 用的是 npmmirror,
|
|
249
|
+
一旦 `package-lock.json` 里出现 `registry.npmmirror.com` 的 `resolved` 地址,CI 的 `npm ci` 会直接失败:
|
|
250
|
+
`npm error code EALLOWREMOTE` / `Fetching packages of type "remote" have been disabled`。
|
|
251
|
+
遇到时的修法:删掉 `package-lock.json` → `npm install --package-lock-only --registry=https://registry.npmjs.org` 重新生成。
|
|
252
|
+
|
|
253
|
+
### npm 作用域说明
|
|
254
|
+
|
|
255
|
+
本账号 npm 用户名是 `suwenguang`,`@suveng` 作用域属于另一个账号(用户 `suveng`),本项目无法使用。
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@suwenguang/pi-feishu-bridge",
|
|
3
|
-
"version": "0.6.
|
|
3
|
+
"version": "0.6.3",
|
|
4
4
|
"description": "Feishu/Lark bridge for the Pi coding agent — chat with Pi from Feishu or Lark",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"license": "MIT",
|
|
@@ -45,7 +45,8 @@
|
|
|
45
45
|
},
|
|
46
46
|
"files": [
|
|
47
47
|
"src",
|
|
48
|
-
".pi"
|
|
48
|
+
".pi",
|
|
49
|
+
"scripts"
|
|
49
50
|
],
|
|
50
51
|
"pi": {
|
|
51
52
|
"extensions": [
|
|
@@ -0,0 +1,233 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
/**
|
|
3
|
+
* 重启 worker:`node scripts/restart-worker.mjs <ticketPath>`
|
|
4
|
+
*
|
|
5
|
+
* 独立于桥接进程存在的原因——老进程必须死,新进程必须由别人拉起,
|
|
6
|
+
* 而且「新进程是否真的拿到网关锁」只能由第三方观察。
|
|
7
|
+
*
|
|
8
|
+
* 约定(restart.ts 写票据,这里读票据):
|
|
9
|
+
* ticket.oldPid 老 daemon 的 pid(先等它自己退,超时才 SIGTERM/SIGKILL)
|
|
10
|
+
* ticket.piBin/args 新 daemon 的启动形状(与 startDaemon 完全一致)
|
|
11
|
+
* ticket.lockPath 网关锁文件,出现在里面的 pid 才算真正接管
|
|
12
|
+
* ticket.resultPath 结果落点,新 daemon 启动时读它来回改卡片
|
|
13
|
+
*
|
|
14
|
+
* 这个文件是纯 .mjs:不能 import TS(daemon 由 `--experimental-strip-types` 跑的 TS,
|
|
15
|
+
* worker 只想依赖 node 本体,避免启动失败把重启卡死)。
|
|
16
|
+
*/
|
|
17
|
+
import { closeSync, openSync, readFileSync, renameSync, rmSync, writeFileSync } from "node:fs";
|
|
18
|
+
import { spawn, spawnSync } from "node:child_process";
|
|
19
|
+
|
|
20
|
+
const ticketPath = process.argv[2];
|
|
21
|
+
if (!ticketPath) {
|
|
22
|
+
console.error("[restart-worker] 缺少票据路径");
|
|
23
|
+
process.exit(2);
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
|
|
27
|
+
|
|
28
|
+
function readJson(path) {
|
|
29
|
+
try {
|
|
30
|
+
return JSON.parse(readFileSync(path, "utf8"));
|
|
31
|
+
} catch {
|
|
32
|
+
return undefined;
|
|
33
|
+
}
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
function writeResult(path, result) {
|
|
37
|
+
const payload = JSON.stringify(result, null, 2);
|
|
38
|
+
try {
|
|
39
|
+
const tmp = `${path}.tmp`;
|
|
40
|
+
writeFileSync(tmp, payload, "utf8");
|
|
41
|
+
renameSync(tmp, path);
|
|
42
|
+
return true;
|
|
43
|
+
} catch (error) {
|
|
44
|
+
console.error(`[restart-worker] 写结果失败:${error instanceof Error ? error.message : String(error)}`);
|
|
45
|
+
try {
|
|
46
|
+
writeFileSync(path, payload, "utf8");
|
|
47
|
+
return true;
|
|
48
|
+
} catch {
|
|
49
|
+
return false;
|
|
50
|
+
}
|
|
51
|
+
}
|
|
52
|
+
}
|
|
53
|
+
|
|
54
|
+
function isAlive(pid) {
|
|
55
|
+
if (!pid || pid <= 0) return false;
|
|
56
|
+
try {
|
|
57
|
+
process.kill(pid, 0);
|
|
58
|
+
return true;
|
|
59
|
+
} catch (error) {
|
|
60
|
+
return error?.code === "EPERM";
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
function signal(pid, name) {
|
|
65
|
+
try {
|
|
66
|
+
process.kill(pid, name);
|
|
67
|
+
return true;
|
|
68
|
+
} catch {
|
|
69
|
+
return false;
|
|
70
|
+
}
|
|
71
|
+
}
|
|
72
|
+
|
|
73
|
+
function quoteShell(value) {
|
|
74
|
+
return `'${String(value).replace(/'/g, `'\\''`)}'`;
|
|
75
|
+
}
|
|
76
|
+
|
|
77
|
+
/**
|
|
78
|
+
* 锁文件里落到「我们拉起的那个进程」才算接管。
|
|
79
|
+
*
|
|
80
|
+
* 注意 POSIX 的启动形状是 `bash -lc 'tail -f /dev/null | exec pi …'`:
|
|
81
|
+
* `spawn()` 拿到的是**外层 bash** 的 pid,真正的 daemon 是被 exec 掉的管道子进程,
|
|
82
|
+
* 所以除了 pid 相等,还要认「父进程是我们」这一种(实测 bash pid 与 daemon pid 差几号)。
|
|
83
|
+
*/
|
|
84
|
+
function ownsLock(lockPath, launcherPid) {
|
|
85
|
+
const locks = readJson(lockPath);
|
|
86
|
+
if (!locks || typeof locks !== "object") return undefined;
|
|
87
|
+
for (const entry of Object.values(locks)) {
|
|
88
|
+
if (!entry || typeof entry.pid !== "number") continue;
|
|
89
|
+
if (entry.pid === launcherPid) return entry.pid;
|
|
90
|
+
if (parentPidOf(entry.pid) === launcherPid && isAlive(entry.pid)) return entry.pid;
|
|
91
|
+
}
|
|
92
|
+
return undefined;
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
function psField(pid, field) {
|
|
96
|
+
if (process.platform === "win32") return undefined;
|
|
97
|
+
const res = spawnSync("ps", ["-o", `${field}=`, "-p", String(pid)], { encoding: "utf8" });
|
|
98
|
+
const value = res.stdout?.trim();
|
|
99
|
+
return value || undefined;
|
|
100
|
+
}
|
|
101
|
+
|
|
102
|
+
/**
|
|
103
|
+
* 收掉老 daemon 的启动壳 `bash -lc 'tail -f /dev/null | exec pi …'`。
|
|
104
|
+
*
|
|
105
|
+
* 老 daemon 自己退出时只 SIGTERM 了父 bash(旧版本行为),里面的 `tail -f /dev/null`
|
|
106
|
+
* 会被 reparent 到 init 永久挂着,每重启一次漏一个。worker 是新代码、每次现读,
|
|
107
|
+
* 所以在这里兜住:老进程死后按**进程组**整组收(detached spawn 的组里只有 bash+tail)。
|
|
108
|
+
*/
|
|
109
|
+
function cleanupLauncher(launcherPid) {
|
|
110
|
+
if (!launcherPid || launcherPid <= 1) return;
|
|
111
|
+
try {
|
|
112
|
+
const command = psField(launcherPid, "command");
|
|
113
|
+
if (!command || !command.includes("tail -f /dev/null")) return;
|
|
114
|
+
const pgid = Number(psField(launcherPid, "pgid"));
|
|
115
|
+
if (Number.isFinite(pgid) && pgid === launcherPid) {
|
|
116
|
+
process.kill(-pgid, "SIGTERM");
|
|
117
|
+
} else {
|
|
118
|
+
process.kill(launcherPid, "SIGTERM");
|
|
119
|
+
}
|
|
120
|
+
console.log(`[restart-worker] 已收掉老启动壳 ${launcherPid}(含 tail)`);
|
|
121
|
+
} catch (error) {
|
|
122
|
+
console.error(`[restart-worker] 收老启动壳失败:${error instanceof Error ? error.message : String(error)}`);
|
|
123
|
+
}
|
|
124
|
+
}
|
|
125
|
+
|
|
126
|
+
function parentPidOf(pid) {
|
|
127
|
+
if (process.platform === "win32") return undefined;
|
|
128
|
+
const res = spawnSync("ps", ["-o", "ppid=", "-p", String(pid)], { encoding: "utf8" });
|
|
129
|
+
const ppid = Number(res.stdout?.trim());
|
|
130
|
+
return Number.isFinite(ppid) && ppid > 0 ? ppid : undefined;
|
|
131
|
+
}
|
|
132
|
+
|
|
133
|
+
function spawnDaemon(ticket) {
|
|
134
|
+
const logFd = openSync(ticket.logPath, "a");
|
|
135
|
+
if (process.platform === "win32") {
|
|
136
|
+
const npmPrefix = process.env.APPDATA ? `${process.env.APPDATA}\\npm` : `${process.env.USERPROFILE}\\AppData\\Roaming\\npm`;
|
|
137
|
+
const piCliEntry = process.env.PI_CLI_ENTRY || `${npmPrefix}\\node_modules\\@earendil-works\\pi-coding-agent\\dist\\cli.js`;
|
|
138
|
+
const child = spawn("node", [piCliEntry, ...ticket.args], {
|
|
139
|
+
detached: true,
|
|
140
|
+
cwd: ticket.cwd,
|
|
141
|
+
env: { ...process.env, PI_FEISHU_DAEMON: "1" },
|
|
142
|
+
stdio: ["pipe", logFd, logFd],
|
|
143
|
+
});
|
|
144
|
+
if (child.stdin) child.stdin.resume();
|
|
145
|
+
closeSync(logFd);
|
|
146
|
+
return child;
|
|
147
|
+
}
|
|
148
|
+
const command = `tail -f /dev/null | exec ${quoteShell(ticket.piBin)} ${ticket.args.map(quoteShell).join(" ")}`;
|
|
149
|
+
const child = spawn("bash", ["-lc", command], {
|
|
150
|
+
detached: true,
|
|
151
|
+
cwd: ticket.cwd,
|
|
152
|
+
env: { ...process.env, PI_FEISHU_DAEMON: "1" },
|
|
153
|
+
stdio: ["ignore", logFd, logFd],
|
|
154
|
+
});
|
|
155
|
+
closeSync(logFd);
|
|
156
|
+
return child;
|
|
157
|
+
}
|
|
158
|
+
|
|
159
|
+
async function main() {
|
|
160
|
+
const ticket = readJson(ticketPath);
|
|
161
|
+
if (!ticket || ticket.version !== 1) {
|
|
162
|
+
console.error(`[restart-worker] 票据不可读:${ticketPath}`);
|
|
163
|
+
return;
|
|
164
|
+
}
|
|
165
|
+
const startedAt = Number(ticket.startedAt) || Date.now();
|
|
166
|
+
const resultPath = ticket.resultPath;
|
|
167
|
+
const started = Date.now();
|
|
168
|
+
const finish = (result) => writeResult(resultPath, { version: 1, at: Date.now(), elapsedMs: Date.now() - started, ...result });
|
|
169
|
+
|
|
170
|
+
console.log(`[restart-worker] 票据 ${ticketPath} · 老 pid ${ticket.oldPid} · cwd ${ticket.cwd}`);
|
|
171
|
+
|
|
172
|
+
// 1) 等老进程自己退(它发完「重启中」卡片、拉完我们就 return,正常 1s 内就没了)
|
|
173
|
+
// 票据里的 launcherPid 是老 daemon 退出前记下的(最可靠);老版本票据没有这个字段,
|
|
174
|
+
// 退回现查 ppid(只有在老 daemon 还没退时才有意义)。
|
|
175
|
+
const oldLauncherPid = ticket.launcherPid || parentPidOf(ticket.oldPid);
|
|
176
|
+
const exitDeadline = Date.now() + 15_000;
|
|
177
|
+
while (Date.now() < exitDeadline && isAlive(ticket.oldPid)) {
|
|
178
|
+
if (Date.now() - started > 5_000) signal(ticket.oldPid, "SIGTERM");
|
|
179
|
+
if (Date.now() - started > 10_000) signal(ticket.oldPid, "SIGKILL");
|
|
180
|
+
await sleep(100);
|
|
181
|
+
}
|
|
182
|
+
if (isAlive(ticket.oldPid)) {
|
|
183
|
+
console.error(`[restart-worker] 老进程 ${ticket.oldPid} 杀不掉,放弃重启`);
|
|
184
|
+
finish({ ok: false, error: `老桥接进程(pid ${ticket.oldPid})无法结束,请在终端用 /feishu restart 重试` });
|
|
185
|
+
return;
|
|
186
|
+
}
|
|
187
|
+
console.log(`[restart-worker] 老进程已退出,耗时 ${Date.now() - started}ms`);
|
|
188
|
+
cleanupLauncher(oldLauncherPid);
|
|
189
|
+
|
|
190
|
+
// 2) 按同一 spec 拉新 daemon
|
|
191
|
+
rmSync(ticket.resultPath, { force: true });
|
|
192
|
+
let child;
|
|
193
|
+
try {
|
|
194
|
+
child = spawnDaemon(ticket);
|
|
195
|
+
} catch (error) {
|
|
196
|
+
finish({ ok: false, error: `新桥接进程启动失败:${error instanceof Error ? error.message : String(error)}` });
|
|
197
|
+
return;
|
|
198
|
+
}
|
|
199
|
+
const newPid = child.pid;
|
|
200
|
+
let exited = false;
|
|
201
|
+
child.on("exit", (code, sig) => {
|
|
202
|
+
exited = true;
|
|
203
|
+
console.error(`[restart-worker] 新进程 ${newPid} 退出 code=${code} sig=${sig}`);
|
|
204
|
+
});
|
|
205
|
+
child.unref();
|
|
206
|
+
console.log(`[restart-worker] 新进程 pid ${newPid},等待网关锁`);
|
|
207
|
+
|
|
208
|
+
// 3) 等网关锁落到新 pid(启动重试退避最长约 15s,这里留 45s)
|
|
209
|
+
const lockDeadline = Date.now() + 45_000;
|
|
210
|
+
while (Date.now() < lockDeadline) {
|
|
211
|
+
if (exited) {
|
|
212
|
+
finish({ ok: false, error: `新桥接进程启动后立即退出(pid ${newPid}),详见 ${ticket.logPath}` });
|
|
213
|
+
return;
|
|
214
|
+
}
|
|
215
|
+
const lockPid = ownsLock(ticket.lockPath, newPid);
|
|
216
|
+
if (lockPid) {
|
|
217
|
+
console.log(`[restart-worker] 已接管网关锁(pid ${lockPid},launcher ${newPid}),耗时 ${Date.now() - started}ms`);
|
|
218
|
+
finish({ ok: true, newPid: lockPid, launcherPid: newPid });
|
|
219
|
+
return;
|
|
220
|
+
}
|
|
221
|
+
await sleep(250);
|
|
222
|
+
}
|
|
223
|
+
finish({ ok: false, error: `新进程 ${newPid} 45s 内未拿到网关锁,详见 ${ticket.logPath}` });
|
|
224
|
+
}
|
|
225
|
+
|
|
226
|
+
main()
|
|
227
|
+
.catch((error) => {
|
|
228
|
+
console.error(`[restart-worker] 未捕获异常:${error instanceof Error ? error.stack || error.message : String(error)}`);
|
|
229
|
+
})
|
|
230
|
+
.finally(() => {
|
|
231
|
+
// 不 await:detached 子进程不受本进程退出影响
|
|
232
|
+
process.exit(0);
|
|
233
|
+
});
|