dsh-plugin-weapp-gateway 0.8.33 → 0.8.35

package/README.md

@@ -6,11 +6,25 @@
 
 `dsh-plugin-weapp-gateway` · DeepSeek Harness 移动客户端连接插件
 
+### v0.8.35 · 修复 Windows 上的连接器测试
+
+- 插件把 `spawnProcess` 选项转发给内置中继连接器，与 Cloudflare 隧道已有的注入缝一致；默认仍是真实 `spawn`，生产行为不变。测试可用它包住真实子进程并记录启动环境，无需依赖平台专有的进程检查工具。
+- 修复 `relay-connector` 测试用 `URL.pathname` 生成连接器脚本路径：Windows 上会得到 `/D:/…`，交给 `spawn` 后变成 `D:\D:\…`，连接器无法启动。改用与生产代码一致的 `fileURLToPath`。
+- 修复 `restart-reconnect` 测试用 POSIX 的 `ps eww` 读取连接器环境变量：改为通过注入缝记录真实的 `GATEWAY_URL`，在 Windows 与 POSIX 上都能验证重启后连接器拨向新端口。32 个测试文件全部通过。
+
 ### v0.8.32 · 修复 Windows 上网关完全没有日志
 
 - 修复默认日志路径写死成 POSIX 的 `/tmp/mobile-gateway.log`：Windows 上它会解析到「当前盘:\tmp」（一般不存在），`appendFileSync` 抛 ENOENT 又被 `log()` 的 catch 静默吞掉，运营者于是**完全看不到任何日志**——排查「手机为什么掉线」时没有任何线索。现在按平台取 `os.tmpdir()`，Windows 上落到 `%TEMP%\mobile-gateway.log`；Linux/macOS 行为不变。
 - 日志内容本身不变：连接/断开、配对、鉴权拒绝、设备撤销照旧记录。`logFile` 选项仍可覆盖，测试与并行安装仍写各自的文件。
 
+### v0.8.34 · 修复 DSH 重启后手机重连不上
+
+- 内置中继连接器新增父进程守护：DSH 崩溃或强退后连接器不再继续存活占锁。此前幸存者会一直拨已死亡的 Web 端口，之后每次 DSH 重启都只会委派给它，走中继的手机永久重连不上（连接器随插件进程在 5 秒内退出并释放 pid 锁；手动运行的 systemd/nohup 连接器不注入该变量，行为不变）。
+- 监督者在委派给其他连接器后会复查锁状态：持有者消失时自动接管并拉起新连接器，无需再重启一次 DSH。
+- 「临时开启」的等待超时不再把「已关闭」写回状态文件：超时只是运行时保险，运营者保存的是临时开启；落盘 disabled 会让下次重启直接关闭并拒绝所有手机重连，直到有人手动重新开启。现在重启后按保存的模式重新计时。
+- 局域网监听在绑定失败（EADDRINUSE，重启时旧进程尚未释放 3081）后自动重试约一分钟，而不是永久放弃导致局域网手机 ECONNREFUSED。
+- 新增 `test/restart-reconnect.test.mjs`：完整模拟 DSH 重启（dispose + 更换端口重新 apply），覆盖持久模式恢复、长期 token 重连、连接器用新端口重拉、临时模式语义与局域网端口竞态；连接器生命周期回归进入 `test/relay-connector.test.mjs`。
+
 ### v0.8.33 · 可信设备连接线路徽标
 
 - 「设置 → 鲸尾移动网关 → 可信设备」为每台在线设备添加连接线路徽标：**局域网**（手机与 DSH 同网直连）、**公网**（公网服务器、反向代理或鲸尾中继）、**Cloudflare 外网**（命名 / Quick Tunnel）。徽标跟随手机当前实际连接实时变化，面板 3 秒轮询自动刷新，手机从 Wi-Fi 切到蜂窝即换线路。

package/lib/index.mjs

@@ -2757,6 +2757,9 @@ const plugin = {
       gatewayName: gatewayIdentity.gatewayName,
       mgmtUrl: `http://127.0.0.1:${webServer.port}`,
       stateDir: resolveStateDir(),
+      // 注入缝：与 cloudflared 一致。测试用它包住真正的 spawn，
+      // 既拿到连接器的环境变量（GATEWAY_URL 等）又不妨碍它真的跑起来。
+      spawnProcess: options.spawnProcess,
       log,
     })
     let relayEnsureTimer = null
@@ -2819,6 +2822,7 @@ const plugin = {
     let lanListening = false
     let lanListenError = null
     let lanBoundPort = options.lanPort
+    let lanDisposeLanRetry = null
     let tunnel = null
     const configuredPublicUrl = () => {
       if (typeof options.publicUrl === 'string' && options.publicUrl.trim()) return options.publicUrl.trim()
@@ -2864,9 +2868,12 @@ const plugin = {
           gatewayEnabled = false
           gatewayMode = 'disabled'
           waitExpiresAt = null
-          try { gatewayState.setMode('disabled') } catch (error) {
-            log(`failed to persist automatic gateway disable: ${error.message}`)
-          }
+          // Deliberately NOT persisted: the operator's saved choice is
+          // 'temporary', and the panel documents that a restart re-counts
+          // the window. Writing 'disabled' here would silently turn their
+          // choice into a permanent off switch, so every phone reconnect
+          // after the next restart would be rejected with 503 until
+          // somebody noticed and re-enabled the gateway by hand.
           log('mobile gateway automatically disabled: no device connected before timeout')
           void tunnel?.reconcile()
         }, options.gatewayWaitTimeoutMs)
@@ -3967,19 +3974,40 @@ const plugin = {
         }
         handleMobileUpgrade(req, socket, head, { lan: true, port: lanBoundPort })
       })
+      // A restart can overlap the previous process still holding the LAN
+      // port (app relaunch handoff, lingering socket). Giving up on the
+      // first EADDRINUSE leaves every LAN phone with ECONNREFUSED until
+      // somebody restarts again, so retry with backoff for about a minute.
+      let lanRetryTimer = null
+      let lanRetries = 0
+      const clearLanRetry = () => {
+        if (lanRetryTimer === null) return
+        clearTimeout(lanRetryTimer)
+        lanRetryTimer = null
+      }
+      const listenLan = () => {
+        clearLanRetry()
+        lanServer.listen(options.lanPort, options.lanHost, () => {
+          lanRetries = 0
+          lanListening = true
+          lanListenError = null
+          const address = lanServer.address()
+          if (address && typeof address === 'object') lanBoundPort = address.port
+          const urls = lanWebSocketUrls(options, wsPath, lanBoundPort)
+          log(`LAN listener ready: ${options.lanHost}:${lanBoundPort}${wsPath}${urls.length ? ` (${urls.join(', ')})` : ''}; authentication forced`)
+        })
+      }
       lanServer.on('error', (error) => {
         lanListening = false
         lanListenError = error && error.message ? error.message : String(error)
         log(`LAN listener failed: ${lanListenError}`)
+        lanRetries += 1
+        if (lanRetries > 30 || lanRetryTimer !== null) return
+        lanRetryTimer = setTimeout(listenLan, 2_000)
+        lanRetryTimer.unref?.()
       })
-      lanServer.listen(options.lanPort, options.lanHost, () => {
-        lanListening = true
-        lanListenError = null
-        const address = lanServer.address()
-        if (address && typeof address === 'object') lanBoundPort = address.port
-        const urls = lanWebSocketUrls(options, wsPath, lanBoundPort)
-        log(`LAN listener ready: ${options.lanHost}:${lanBoundPort}${wsPath}${urls.length ? ` (${urls.join(', ')})` : ''}; authentication forced`)
-      })
+      lanDisposeLanRetry = clearLanRetry
+      listenLan()
     }
 
     tunnel = createCloudflareTunnel({
@@ -4334,6 +4362,7 @@ const plugin = {
       disposeQuestions()
       disposeApprovals()
       if (lanServer) lanServer.close()
+      if (lanDisposeLanRetry) lanDisposeLanRetry()
       clearRelayEnsure()
       relayConnector.dispose()
       tunnel.dispose()

package/lib/relay-connector.mjs

@@ -46,13 +46,14 @@ export function resolveStateDir(env = process.env, home = os.homedir(), legacyHo
 
 export function createRelayConnector({ relay, gatewayUrl, gatewayName, mgmtUrl,
   stateDir = resolveStateDir(), log = () => {}, spawnProcess = spawn, now = () => Date.now(),
-  restartDelayMs = 5_000 } = {}) {
+  restartDelayMs = 5_000, delegationRecheckMs = 15_000 } = {}) {
   // Whale public service is configured internally, not a user-facing domain.
   const configuredRelay = String(relay || '').replace(/[/]$/, '')
   relay = ['', 'wss://relay.ahwe.top', 'wss://relay.ahwe.top/relay'].includes(configuredRelay)
     ? 'wss://dsh.ahwe.top/relay' : configuredRelay
   let child = null
   let restartTimer = null
+  let delegationTimer = null
   let stopped = false
   let restarts = 0
   let windowStart = now()
@@ -74,9 +75,31 @@ export function createRelayConnector({ relay, gatewayUrl, gatewayName, mgmtUrl,
     restartTimer = null
   }
 
+  const clearDelegationTimer = () => {
+    if (delegationTimer === null) return
+    clearTimeout(delegationTimer)
+    delegationTimer = null
+  }
+
+  // A holder can disappear after we deferred to it (a zombie from an
+  // unclean DSH exit that only now noticed, or an operator connector that
+  // died). Without a re-check the delegation would be permanent: relay-
+  // routed phones would keep failing against a bridge that dials a dead
+  // port, and no restart would help because the next one defers again.
+  const scheduleDelegationRecheck = () => {
+    if (delegationTimer !== null || stopped) return
+    delegationTimer = setTimeout(() => {
+      delegationTimer = null
+      if (stopped || child !== null) return
+      start()
+    }, delegationRecheckMs)
+    delegationTimer.unref?.()
+  }
+
   const stop = () => {
     stopped = true
     clearTimer()
+    clearDelegationTimer()
     const previous = child
     child = null
     if (previous === null || previous.exitCode !== null || previous.signalCode !== null) return
@@ -108,6 +131,7 @@ export function createRelayConnector({ relay, gatewayUrl, gatewayName, mgmtUrl,
       if (holder !== null) {
         delegatedTo = holder
         lastError = null
+        scheduleDelegationRecheck()
         return snapshot()
       }
     }
@@ -128,6 +152,13 @@ export function createRelayConnector({ relay, gatewayUrl, gatewayName, mgmtUrl,
           ...process.env,
           // Desktop runs under Electron; child scripts must execute as Node.
           ...(process.versions.electron ? { ELECTRON_RUN_AS_NODE: '1' } : {}),
+          // The bundled connector must not outlive the DSH process that
+          // spawned it: after a crash or force-quit a survivor would keep
+          // the pid lock and keep dialling the dead web port, so every
+          // later restart would delegate to it and relay-routed phones
+          // could never reconnect. Manually run connectors
+          // (systemd/nohup) never receive this variable.
+          PARENT_PID: String(process.pid),
           RELAY_URL: relay,
           GATEWAY_URL: gatewayUrl,
           GATEWAY_NAME: gatewayName || '',
@@ -143,6 +174,7 @@ export function createRelayConnector({ relay, gatewayUrl, gatewayName, mgmtUrl,
     }
     child = spawned
     delegatedTo = null
+    clearDelegationTimer()
     log(`bundled connector starting: pid=${spawned.pid} relay=${relay} state=${stateDir}`)
     spawned.on('error', (error) => {
       if (child !== spawned) return

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-plugin-weapp-gateway",
-  "version": "0.8.33",
+  "version": "0.8.35",
   "description": "鲸尾移动网关：DeepSeek Harness 移动客户端连接与会话同步，支持安全配对、远程访问、上下文、编队、额度和语音输入",
   "main": "lib/index.mjs",
   "files": [
@@ -29,7 +29,7 @@
   },
   "bin": "bin/setup-ip.mjs",
   "scripts": {
-    "test": "node test/wire-json.test.mjs && node --test test/history-cache.test.mjs && node test/session-target.test.mjs && node test/setup-ip.test.mjs && node test/host-adapter.test.mjs && node test/file-upload-adapter.test.mjs && node test/file-upload.test.mjs && node test/agent-team.test.mjs && node test/account-bridge.test.mjs && node test/context-bridge.test.mjs && node test/cost-meter.test.mjs && node --test test/cost-meter-dashboard.test.mjs && node test/clawbot-service.test.mjs && node test/speech-input.test.mjs && node test/speech-official.test.mjs && node test/plugin-management.test.mjs && node test/plugin-updates.test.mjs && node test/session-preset.test.mjs && node test/session-follower.test.mjs && node test/completion-notifications.test.mjs && node --test test/workspace-changes.test.mjs && node test/gateway.test.mjs && node test/auth.test.mjs && node test/lan.test.mjs && node test/cloudflared-binary.test.mjs && node test/cloudflare-tunnel.test.mjs && node --test test/public-endpoint-check.test.mjs && node test/gateway-tools.test.mjs && node test/relay-connector.test.mjs && node test/cloudflare-linux.test.mjs && node test/multi-gateway.test.mjs"
+    "test": "node test/wire-json.test.mjs && node --test test/history-cache.test.mjs && node test/session-target.test.mjs && node test/setup-ip.test.mjs && node test/host-adapter.test.mjs && node test/file-upload-adapter.test.mjs && node test/file-upload.test.mjs && node test/agent-team.test.mjs && node test/account-bridge.test.mjs && node test/context-bridge.test.mjs && node test/cost-meter.test.mjs && node --test test/cost-meter-dashboard.test.mjs && node test/clawbot-service.test.mjs && node test/speech-input.test.mjs && node test/speech-official.test.mjs && node test/plugin-management.test.mjs && node test/plugin-updates.test.mjs && node test/session-preset.test.mjs && node test/session-follower.test.mjs && node test/completion-notifications.test.mjs && node --test test/workspace-changes.test.mjs && node test/gateway.test.mjs && node test/auth.test.mjs && node test/lan.test.mjs && node test/cloudflared-binary.test.mjs && node test/cloudflare-tunnel.test.mjs && node --test test/public-endpoint-check.test.mjs && node test/gateway-tools.test.mjs && node test/relay-connector.test.mjs && node test/restart-reconnect.test.mjs && node test/cloudflare-linux.test.mjs && node test/multi-gateway.test.mjs"
   },
   "exports": {
     ".": "./lib/index.mjs",

package/relay/agent-connector.js

@@ -55,6 +55,25 @@ process.on('exit', () => {
   try { if (JSON.parse(fs.readFileSync(lockFile, 'utf8')).pid === process.pid) fs.unlinkSync(lockFile); } catch {}
 });
 
+// 父进程守护（仅插件内置连接器）：DSH 崩溃或强退时，子进程会被过继给 init 继续
+// 存活——它占着 pid 锁、拨已死亡的端口，之后每次 DSH 重启都只会委派给这个僵尸，
+// 手机的中继线路永久重连不上。插件启动时注入 PARENT_PID；手动运行
+// （systemd / nohup）不设置该变量，行为完全不变。POSIX 上父进程死亡后 ppid 会被
+// 改写，同时兜住 pid 复用。
+const PARENT_PID = Number(process.env.PARENT_PID) || 0;
+if (PARENT_PID > 0) {
+  const startupPpid = process.ppid;
+  const parentGone = () => {
+    if (process.ppid !== startupPpid) return true;
+    try { process.kill(PARENT_PID, 0); return false; } catch (e) { return e.code === 'ESRCH'; }
+  };
+  setInterval(() => {
+    if (!parentGone()) return;
+    console.log('[' + new Date().toISOString() + '] parent process ' + PARENT_PID + ' exited; connector shutting down');
+    process.exit(0);
+  }, 5000).unref?.();
+}
+
 // ---------- 身份 ----------
 
 const PROTOCOL = 'dsh-weapp-e2ee-v1';