@yeaft/webchat-agent 1.0.64 → 1.0.65

package/connection/buffer.js

@@ -16,51 +16,84 @@ export const BUFFERABLE_TYPES = new Set([
   'crew_role_compact', 'crew_context_usage'
 ]);
 
-// Send message to server (with encryption if available)
-// 断连时对关键消息类型进行缓冲，重连后自动 flush
-export async function sendToServer(msg) {
+function bufferMessage(msg, reason) {
+  if (!BUFFERABLE_TYPES.has(msg.type)) {
+    console.warn(`[WS] Cannot send message, WebSocket not open: ${msg.type}`);
+    return 'dropped';
+  }
+  if (ctx.messageBuffer.length < ctx.messageBufferMaxSize) {
+    ctx.messageBuffer.push(msg);
+    console.log(`[WS] ${reason}, buffered: ${msg.type} (queue: ${ctx.messageBuffer.length})`);
+    return 'buffered';
+  }
+  // Buffer full: drop oldest non-status messages to make room
+  const dropIdx = ctx.messageBuffer.findIndex(m => m.type !== 'crew_status' && m.type !== 'turn_completed');
+  if (dropIdx >= 0) {
+    ctx.messageBuffer.splice(dropIdx, 1);
+    ctx.messageBuffer.push(msg);
+    console.warn(`[WS] Buffer full, dropped oldest to make room for: ${msg.type}`);
+    return 'buffered';
+  }
+  console.warn(`[WS] Buffer full (${ctx.messageBufferMaxSize}), dropping: ${msg.type}`);
+  return 'dropped';
+}
+
+async function sendNow(msg) {
   if (!ctx.ws || ctx.ws.readyState !== WebSocket.OPEN) {
-    // 缓冲关键消息
-    if (BUFFERABLE_TYPES.has(msg.type)) {
-      if (ctx.messageBuffer.length < ctx.messageBufferMaxSize) {
-        ctx.messageBuffer.push(msg);
-        console.log(`[WS] Buffered message: ${msg.type} (queue: ${ctx.messageBuffer.length})`);
-      } else {
-        // Buffer full: drop oldest non-status messages to make room
-        const dropIdx = ctx.messageBuffer.findIndex(m => m.type !== 'crew_status' && m.type !== 'turn_completed');
-        if (dropIdx >= 0) {
-          ctx.messageBuffer.splice(dropIdx, 1);
-          ctx.messageBuffer.push(msg);
-          console.warn(`[WS] Buffer full, dropped oldest to make room for: ${msg.type}`);
-        } else {
-          console.warn(`[WS] Buffer full (${ctx.messageBufferMaxSize}), dropping: ${msg.type}`);
+    return bufferMessage(msg, 'Disconnected');
+  }
+
+  // feat-ws-plaintext-negotiation: encrypt only when the server has
+  // NOT advertised plaintext acceptance. Defaults to encrypted for
+  // back-compat with old servers; flipped to plaintext when the
+  // `registered` frame includes `acceptPlaintext: true`.
+  if (ctx.serverEncryptionRequired && ctx.sessionKey) {
+    const encrypted = await encrypt(msg, ctx.sessionKey);
+    ctx.ws.send(JSON.stringify(encrypted));
+  } else {
+    ctx.ws.send(JSON.stringify(msg));
+  }
+  return 'sent';
+}
+
+function scheduleOutboundDrain() {
+  if (ctx.outboundSendQueueActive) return;
+  ctx.outboundSendQueueActive = true;
+  setImmediate(async () => {
+    try {
+      while (ctx.outboundSendQueue.length > 0) {
+        const item = ctx.outboundSendQueue.shift();
+        const msg = item?.msg ?? item;
+        try {
+          const outcome = await sendNow(msg);
+          item?.resolve?.(outcome);
+        } catch (e) {
+          console.error(`[WS] Error sending message ${msg?.type}:`, e.message);
+          const outcome = msg ? bufferMessage(msg, 'Send failed') : 'dropped';
+          item?.resolve?.(outcome);
         }
+        // Yield between frames so ping/pong, inbound control messages and UI
+        // events cannot be starved by a reconnect flush or a burst of tool output.
+        await new Promise(resolve => setImmediate(resolve));
       }
-    } else {
-      console.warn(`[WS] Cannot send message, WebSocket not open: ${msg.type}`);
+    } finally {
+      ctx.outboundSendQueueActive = false;
+      if (ctx.outboundSendQueue.length > 0) scheduleOutboundDrain();
     }
-    return;
-  }
+  });
+}
 
-  try {
-    // feat-ws-plaintext-negotiation: encrypt only when the server has
-    // NOT advertised plaintext acceptance. Defaults to encrypted for
-    // back-compat with old servers; flipped to plaintext when the
-    // `registered` frame includes `acceptPlaintext: true`.
-    if (ctx.serverEncryptionRequired && ctx.sessionKey) {
-      const encrypted = await encrypt(msg, ctx.sessionKey);
-      ctx.ws.send(JSON.stringify(encrypted));
-    } else {
-      ctx.ws.send(JSON.stringify(msg));
-    }
-  } catch (e) {
-    console.error(`[WS] Error sending message ${msg.type}:`, e.message);
-    // 发送失败也缓冲
-    if (BUFFERABLE_TYPES.has(msg.type) && ctx.messageBuffer.length < ctx.messageBufferMaxSize) {
-      ctx.messageBuffer.push(msg);
-      console.log(`[WS] Send failed, buffered: ${msg.type}`);
-    }
+// Send message to server (with encryption if available)
+// 断连时对关键消息类型进行缓冲，重连后自动 flush
+export async function sendToServer(msg) {
+  if (!ctx.ws || ctx.ws.readyState !== WebSocket.OPEN) {
+    return bufferMessage(msg, 'Disconnected');
   }
+  const promise = new Promise((resolve, reject) => {
+    ctx.outboundSendQueue.push({ msg, resolve, reject });
+  });
+  scheduleOutboundDrain();
+  return promise;
 }
 
 // Flush 断连期间缓冲的消息
@@ -74,7 +107,7 @@ export async function flushMessageBuffer() {
     await sendToServer(msg);
   }
 
-  console.log(`[WS] Flush complete`);
+  console.log(`[WS] Flush queued`);
 }
 
 // Parse incoming message (decrypt if encrypted)

package/connection/heartbeat-policy.js

@@ -1,4 +1,5 @@
 const DEFAULT_AGENT_HEARTBEAT_TIMEOUT_MS = 180000;
+const DEFAULT_AGENT_HEARTBEAT_STALL_GRACE_MS = 5000;
 
 function parsePositiveInt(value, fallback) {
   const n = Number(value);
@@ -10,6 +11,17 @@ export const AGENT_HEARTBEAT_TIMEOUT_MS = parsePositiveInt(
   DEFAULT_AGENT_HEARTBEAT_TIMEOUT_MS,
 );
 
-export function shouldReconnectForHeartbeat(lastPongAt, now = Date.now(), timeoutMs = AGENT_HEARTBEAT_TIMEOUT_MS) {
+export const AGENT_HEARTBEAT_STALL_GRACE_MS = parsePositiveInt(
+  process.env.YEAFT_AGENT_HEARTBEAT_STALL_GRACE_MS || process.env.AGENT_HEARTBEAT_STALL_GRACE_MS,
+  DEFAULT_AGENT_HEARTBEAT_STALL_GRACE_MS,
+);
+
+export function shouldReconnectForHeartbeat(
+  lastPongAt,
+  now = Date.now(),
+  timeoutMs = AGENT_HEARTBEAT_TIMEOUT_MS,
+  { timerDriftMs = 0, stallGraceMs = AGENT_HEARTBEAT_STALL_GRACE_MS } = {},
+) {
+  if (Number.isFinite(timerDriftMs) && timerDriftMs > stallGraceMs) return false;
   return Number.isFinite(lastPongAt) && lastPongAt > 0 && (now - lastPongAt) > timeoutMs;
 }

package/connection/heartbeat.js

@@ -2,9 +2,13 @@ import WebSocket from 'ws';
 import ctx from '../context.js';
 import { shouldReconnectForHeartbeat } from './heartbeat-policy.js';
 
+const AGENT_HEARTBEAT_INTERVAL_MS = 25000;
+const HEARTBEAT_STALL_LOG_MS = 5000;
+
 export function startAgentHeartbeat() {
   stopAgentHeartbeat();
   ctx.lastPongAt = Date.now();
+  ctx.lastHeartbeatTickAt = Date.now();
 
   // 监听 pong 帧
   if (ctx.ws) {
@@ -18,8 +22,15 @@ export function startAgentHeartbeat() {
 
     // 检查上次 pong 是否超时
     const now = Date.now();
+    const timerDriftMs = Math.max(0, now - (ctx.lastHeartbeatTickAt || now) - AGENT_HEARTBEAT_INTERVAL_MS);
+    ctx.lastHeartbeatTickAt = now;
     const sincePong = now - ctx.lastPongAt;
-    if (shouldReconnectForHeartbeat(ctx.lastPongAt, now)) {
+    if (timerDriftMs > HEARTBEAT_STALL_LOG_MS) {
+      ctx.lastHeartbeatStallAt = now;
+      ctx.lastHeartbeatStallMs = timerDriftMs;
+      console.warn(`[Heartbeat] Agent event loop delayed heartbeat by ${Math.round(timerDriftMs)}ms; deferring reconnect`);
+    }
+    if (shouldReconnectForHeartbeat(ctx.lastPongAt, now, undefined, { timerDriftMs })) {
       console.warn(`[Heartbeat] No pong for ${Math.round(sincePong / 1000)}s, reconnecting...`);
       ctx.ws.terminate();
       return;
@@ -30,7 +41,7 @@ export function startAgentHeartbeat() {
     } catch (e) {
       console.warn('[Heartbeat] Failed to send ping:', e.message);
     }
-  }, 25000);
+  }, AGENT_HEARTBEAT_INTERVAL_MS);
 }
 
 export function stopAgentHeartbeat() {

package/connection/message-router.js

@@ -329,7 +329,7 @@ export async function handleMessage(msg) {
       break;
 
     case 'restart_agent':
-      handleRestartAgent();
+      await handleRestartAgent();
       break;
 
     case 'upgrade_agent':

package/connection/upgrade.js

@@ -64,9 +64,9 @@ function cleanupAndExit(exitCode) {
   }, 500);
 }
 
-export function handleRestartAgent() {
+export async function handleRestartAgent() {
   console.log('[Agent] Restart requested, shutting down for PM2/systemd restart...');
-  sendToServer({ type: 'restart_agent_ack' });
+  await sendToServer({ type: 'restart_agent_ack' });
   cleanupAndExit(1);
 }
 
@@ -162,7 +162,7 @@ export async function handleUpgradeAgent() {
     });
     if (latestVersion === ctx.agentVersion) {
       console.log(`[Agent] Already at latest version (${ctx.agentVersion}), skipping upgrade.`);
-      sendToServer({ type: 'upgrade_agent_ack', success: true, alreadyLatest: true, version: ctx.agentVersion });
+      await sendToServer({ type: 'upgrade_agent_ack', success: true, alreadyLatest: true, version: ctx.agentVersion });
       return;
     }
 
@@ -175,7 +175,7 @@ export async function handleUpgradeAgent() {
     if (requiredNode && !nodeRangeSatisfied(currentNode, requiredNode)) {
       const msg = `Node ${currentNode} does not satisfy required ${requiredNode} for ${pkgName}@${latestVersion}`;
       console.warn(`[Agent] Upgrade aborted: ${msg}`);
-      sendToServer({
+      await sendToServer({
         type: 'upgrade_agent_ack',
         success: false,
         reason: 'node_incompatible',
@@ -197,7 +197,7 @@ export async function handleUpgradeAgent() {
     if (!isNpmInstall) {
       // 源码运行不支持远程升级（代码在 git repo 中，需要手动 git pull）
       console.log('[Agent] Source-based install detected, remote upgrade not supported.');
-      sendToServer({ type: 'upgrade_agent_ack', success: false, error: 'Source-based install: please use git pull to upgrade' });
+      await sendToServer({ type: 'upgrade_agent_ack', success: false, error: 'Source-based install: please use git pull to upgrade' });
       return;
     }
 
@@ -217,9 +217,9 @@ export async function handleUpgradeAgent() {
     const instanceId = resolveInstanceId();
 
     if (isWindows) {
-      spawnWindowsUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId);
+      await spawnWindowsUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId);
     } else {
-      spawnUnixUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId);
+      await spawnUnixUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId);
     }
 
     // On PM2: delete the app BEFORE exiting so PM2 won't auto-restart the old version.
@@ -238,11 +238,11 @@ export async function handleUpgradeAgent() {
     cleanupAndExit(0);
   } catch (e) {
     console.error('[Agent] Upgrade failed:', e.message);
-    sendToServer({ type: 'upgrade_agent_ack', success: false, error: e.message });
+    await sendToServer({ type: 'upgrade_agent_ack', success: false, error: e.message });
   }
 }
 
-function spawnWindowsUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId = DEFAULT_INSTANCE_ID) {
+async function spawnWindowsUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId = DEFAULT_INSTANCE_ID) {
   const pid = process.pid;
   const configDir = getConfigDir(instanceId);
   mkdirSync(configDir, { recursive: true });
@@ -364,7 +364,7 @@ function spawnWindowsUpgradeScript(pkgName, installDir, isGlobalInstall, latestV
   }).unref();
 
   console.log(`[Agent] Spawned upgrade via VBScript (PID wait for ${pid}, pm2=${isPm2}, dir=${installDir}): ${batPath}`);
-  sendToServer({ type: 'upgrade_agent_ack', success: true, version: latestVersion, pendingRestart: true });
+  await sendToServer({ type: 'upgrade_agent_ack', success: true, version: latestVersion, pendingRestart: true });
 }
 
 /**
@@ -526,7 +526,7 @@ export function buildUnixUpgradeScript({
   return shLines.join('\n');
 }
 
-function spawnUnixUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId = DEFAULT_INSTANCE_ID) {
+async function spawnUnixUpgradeScript(pkgName, installDir, isGlobalInstall, latestVersion, instanceId = DEFAULT_INSTANCE_ID) {
   const configDir = getConfigDir(instanceId);
   // Create logs/ too: the generated script's `exec > "$LOGFILE" 2>&1` aborts
   // the whole upgrade silently if that directory is missing.
@@ -551,5 +551,5 @@ function spawnUnixUpgradeScript(pkgName, installDir, isGlobalInstall, latestVers
   });
   child.unref();
   console.log(`[Agent] Spawned upgrade script: ${shPath}`);
-  sendToServer({ type: 'upgrade_agent_ack', success: true, version: latestVersion, pendingRestart: true });
+  await sendToServer({ type: 'upgrade_agent_ack', success: true, version: latestVersion, pendingRestart: true });
 }

package/context.js

@@ -28,6 +28,11 @@ export default {
   pendingAuthTempId: null,
   agentHeartbeatTimer: null,
   lastPongAt: 0,
+  lastHeartbeatTickAt: 0,
+  lastHeartbeatStallAt: 0,
+  lastHeartbeatStallMs: 0,
+  outboundSendQueue: [],
+  outboundSendQueueActive: false,
   // 断连期间的消息缓冲队列（重连后 flush）
   messageBuffer: [],
   messageBufferMaxSize: 5000, // 防止内存无限增长

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yeaft/webchat-agent",
-  "version": "1.0.64",
+  "version": "1.0.65",
   "description": "Remote worker agent for Yeaft Web Code Agent — connects the native Yeaft engine, CLI providers, and workbench tools",
   "main": "index.js",
   "type": "module",