claw-subagent-service 0.0.4 → 0.0.5

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claw-subagent-service",
-  "version": "0.0.4",
+  "version": "0.0.5",
   "description": "虾说静态服务",
   "main": "cli.js",
   "bin": {

package/service/daemon.js

@@ -1,20 +1,70 @@
-const { fork } = require('child_process');
+const { fork, execSync } = require('child_process');
 const path = require('path');
 const { createLogger } = require('./logger');
 const { Updater } = require('./updater');
+const { checkPortListening } = require('./modules/port-checker');
 
 const log = createLogger('daemon');
 const WORKER_PATH = path.join(__dirname, 'worker.js');
+const PORT = process.env.SILENT_SERVICE_PORT ? parseInt(process.env.SILENT_SERVICE_PORT, 10) : 38765;
 
 let worker = null;
 let stopping = false;
 let isRollingBack = false;
 let healthTimer = null;
 let currentBackupDir = null;
+let crashCount = 0;
+let lastCrashTime = 0;
 const updater = new Updater();
 
 process.chdir(__dirname);
 
+/**
+ * 尝试释放占用的端口（杀死占用进程）
+ */
+function freePortIfNeeded(port) {
+  try {
+    if (process.platform === 'win32') {
+      const out = execSync(`netstat -ano | findstr ":${port}"`, {
+        encoding: 'utf8', timeout: 5000, windowsHide: true,
+      });
+      for (const line of out.split('\n')) {
+        if (line.includes('LISTENING')) {
+          const pid = parseInt(line.trim().split(/\s+/).pop(), 10);
+          if (pid && pid > 0) {
+            log.warn(`[DAEMON] 端口 ${port} 被进程 ${pid} 占用，正在释放...`);
+            execSync(`taskkill /F /PID ${pid}`, { timeout: 5000, windowsHide: true });
+          }
+        }
+      }
+    } else {
+      const out = execSync(`lsof -i :${port} -t 2>/dev/null`, { encoding: 'utf8', timeout: 5000 });
+      const pid = parseInt(out.trim(), 10);
+      if (pid && pid > 0) {
+        log.warn(`[DAEMON] 端口 ${port} 被进程 ${pid} 占用，正在释放...`);
+        process.kill(pid, 'SIGKILL');
+      }
+    }
+  } catch { /* 端口已被释放或无法查询 */ }
+}
+
+/**
+ * 获取重启延迟（指数退避，最大 60 秒）
+ */
+function getRestartDelay() {
+  const now = Date.now();
+  // 如果距离上次崩溃超过 30 秒，重置计数器
+  if (now - lastCrashTime > 30000) {
+    crashCount = 0;
+  }
+  lastCrashTime = now;
+  crashCount++;
+  // 指数退避: 1s, 2s, 4s, 8s, 16s, 32s, 60s(封顶)
+  const delay = Math.min(1000 * Math.pow(2, crashCount - 1), 60000);
+  log.info(`[DAEMON] Worker 连续崩溃 ${crashCount} 次，等待 ${delay/1000}s 后重启`);
+  return delay;
+}
+
 function startWorker(isAfterUpdate = false, backupDirForRollback = null) {
   if (stopping || isRollingBack) return;
 
@@ -68,7 +118,10 @@ function startWorker(isAfterUpdate = false, backupDirForRollback = null) {
     }
 
     if (!stopping && !isRollingBack) {
-      setTimeout(() => startWorker(false, null), 3000);
+      // 释放端口后使用指数退避重启
+      freePortIfNeeded(PORT);
+      const delay = getRestartDelay();
+      setTimeout(() => startWorker(false, null), delay);
     }
   });
 

package/service/worker.js

@@ -2,6 +2,7 @@ const http = require('http');
 const fs = require('fs');
 const path = require('path');
 const os = require('os');
+const { execSync } = require('child_process');
 const { createLogger } = require('./logger');
 const { RongCloudClient, MessageHandler, ensurePluginsAllow } = require('./rongcloud');
 const { RongyunMessageHandler } = require('./modules/rongyun-message-handler');
@@ -14,6 +15,65 @@ const { startOpencodeService, stopOpencodeService } = require('./modules/opencod
 const log = createLogger('worker');
 const PORT = process.env.SILENT_SERVICE_PORT ? parseInt(process.env.SILENT_SERVICE_PORT, 10) : 38765;
 
+/**
+ * 查找占用指定端口的进程 PID
+ */
+function findPidOnPort(port) {
+  try {
+    if (process.platform === 'win32') {
+      const out = execSync(`netstat -ano | findstr ":${port}"`, {
+        encoding: 'utf8', timeout: 5000, windowsHide: true,
+      });
+      for (const line of out.split('\n')) {
+        if (line.includes('LISTENING')) {
+          const pid = parseInt(line.trim().split(/\s+/).pop(), 10);
+          if (!isNaN(pid)) return pid;
+        }
+      }
+    } else {
+      const out = execSync(`lsof -i :${port} -t 2>/dev/null`, { encoding: 'utf8', timeout: 5000 });
+      const pid = parseInt(out.trim(), 10);
+      if (!isNaN(pid)) return pid;
+    }
+  } catch { /* port is free */ }
+  return null;
+}
+
+/**
+ * 强制终止进程
+ */
+function forceKill(pid) {
+  try {
+    if (process.platform === 'win32') {
+      execSync(`taskkill /F /PID ${pid}`, { timeout: 5000, windowsHide: true });
+    } else {
+      process.kill(pid, 'SIGKILL');
+    }
+    return true;
+  } catch { return false; }
+}
+
+/**
+ * 确保端口未被占用：如果端口被占用，尝试杀死占用进程
+ * 最多重试 3 轮，每轮间隔 1 秒
+ */
+function ensurePortFree(port) {
+  for (let i = 0; i < 3; i++) {
+    const pid = findPidOnPort(port);
+    if (!pid) return true;
+    log.warn(`[WORKER] 端口 ${port} 被进程 ${pid} 占用，正在释放...`);
+    forceKill(pid);
+    // 同步等待端口释放（最多 1.5s）
+    Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, 1500);
+  }
+  const finalPid = findPidOnPort(port);
+  if (finalPid) {
+    log.error(`[WORKER] 端口 ${port} 被进程 ${finalPid} 占用，无法释放`);
+    return false;
+  }
+  return true;
+}
+
 // Timestamp 校验专用日志
 const timestampLogPath = path.join(__dirname, '..', 'logs', 'timestamp-validation.log');
 const logTimestampValidation = (message) => {
@@ -281,10 +341,23 @@ const server = http.createServer((req, res) => {
   res.end('not found');
 });
 
+// 启动前确保端口未被占用（防止 EADDRINUSE 导致崩溃循环）
+if (!ensurePortFree(PORT)) {
+  log.error(`[WORKER] 端口 ${PORT} 无法使用，进程退出`);
+  process.exit(1);
+}
+
 server.listen(PORT, '127.0.0.1', () => {
   log.info(`[WORKER] HTTP 服务已启动: http://127.0.0.1:${PORT}/health`);
 });
 
+server.on('error', (err) => {
+  if (err.code === 'EADDRINUSE') {
+    log.error(`[WORKER] 端口 ${PORT} 被占用且无法释放，进程退出`);
+    process.exit(1);
+  }
+});
+
 process.on('message', (msg) => {
   if (msg?.type === 'prepare-shutdown') {
     log.info(`[WORKER] 收到${msg.reason || 'unknown'}通知，准备优雅退出...`);