mcp-log-query-server 3.2.0 → 3.4.0

package/index.js

@@ -34,6 +34,23 @@ import {
   listLokiEnvironments as getLokiEnvList, listLokiServices as getLokiSvcList
 } from './loki-client.js';
 
+// 超时配置
+const REQUEST_TIMEOUT = 60000;  // MCP 请求兜底超时 60s
+const WATCHDOG_TIMEOUT = 120000; // 进程看门狗 120s，卡死则强制退出
+
+function withTimeout(promise, ms, label) {
+  return Promise.race([
+    promise,
+    new Promise((_, reject) =>
+      setTimeout(() => reject(new Error(`${label} 超时(${ms}ms)`)), ms)
+    ),
+  ]);
+}
+
+// 进程级安全网
+process.on('unhandledRejection', (err) => console.error('[unhandledRejection]', err));
+process.on('uncaughtException', (err) => { console.error('[uncaughtException]', err); process.exit(1); });
+
 // 创建 MCP Server
 const server = new Server(
   {
@@ -320,7 +337,38 @@ server.setRequestHandler(ListToolsRequestSchema, async () => {
 // 处理工具调用
 server.setRequestHandler(CallToolRequestSchema, async (request) => {
   const { name, arguments: args } = request.params;
+  const startTime = Date.now();
+
+  // 进程级看门狗：如果整个请求卡死超时，强制重启进程
+  const watchdog = setTimeout(() => {
+    console.error(`[Watchdog] ${name} 卡死超过 ${WATCHDOG_TIMEOUT}ms，强制退出进程`);
+    process.exit(1);
+  }, WATCHDOG_TIMEOUT);
+  watchdog.unref();
+
+  try {
+    const result = await withTimeout(
+      handleToolCall(name, args),
+      REQUEST_TIMEOUT,
+      name
+    );
+    clearTimeout(watchdog);
+    console.error(`[MCP] ${name} ${Date.now() - startTime}ms`);
+    return result;
+  } catch (error) {
+    clearTimeout(watchdog);
+    console.error(`[Error] ${name} ${Date.now() - startTime}ms: ${error.message}`);
+    return {
+      content: [{ type: 'text', text: `## 执行错误\n\n❌ ${error.message}` }],
+      isError: true
+    };
+  }
+});
 
+/**
+ * 实际的工具调用处理逻辑
+ */
+async function handleToolCall(name, args) {
   try {
     switch (name) {
       case 'query_log': {
@@ -622,25 +670,43 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
 
         console.error(`[MCP] 追踪日志: traceId=${traceId}, namespace=${targetNamespace || 'default'}, 服务数=${servicesToSearch.length}`);
 
+        const TRACE_TOTAL_TIMEOUT = 50000;  // 总耗时上限 50s
+        const TRACE_PER_SERVICE = 10000;    // 单服务超时 10s
+        const traceStart = Date.now();
         const results = [];
+        let searched = 0;
+        let skipped = 0;
+
         for (const serviceName of servicesToSearch) {
+          // 总耗时检查
+          if (Date.now() - traceStart > TRACE_TOTAL_TIMEOUT) {
+            skipped = servicesToSearch.length - searched;
+            console.error(`[MCP] trace_log 总耗时超过 ${TRACE_TOTAL_TIMEOUT}ms，跳过剩余 ${skipped} 个服务`);
+            break;
+          }
+
           const service = findService(serviceName, targetNamespace);
-          if (!service) continue;
+          if (!service) { searched++; continue; }
 
           try {
             const command = `grep -i -C ${contextLines} "${traceId}"`;
-            const result = await queryLog(service, command);
+            const result = await queryLog(service, command, { timeout: TRACE_PER_SERVICE });
 
             if (result && result.trim() && !result.includes('未找到')) {
               results.push({ service: serviceName, namespace: service.namespace, logs: result });
             }
           } catch (err) {
-            console.error(`[MCP] 搜索 ${serviceName} 失败: ${err.message}`);
+            // 快速跳过失败/超时的服务
+            console.error(`[MCP] ${serviceName} 跳过: ${err.message.substring(0, 80)}`);
           }
+          searched++;
         }
 
+        const elapsed = Date.now() - traceStart;
+        const timeNote = skipped > 0 ? `\n**注意**: 已搜索 ${searched}/${servicesToSearch.length} 个服务（耗时 ${elapsed}ms，跳过 ${skipped} 个）` : '';
+
         if (results.length === 0) {
-          return { content: [{ type: 'text', text: `## TraceId 追踪结果\n\n**traceId**: ${traceId}\n**namespace**: ${targetNamespace || '默认'}\n\n❌ 未在任何服务中找到匹配的日志` }] };
+          return { content: [{ type: 'text', text: `## TraceId 追踪结果\n\n**traceId**: ${traceId}\n**namespace**: ${targetNamespace || '默认'}\n\n❌ 未在已搜索的 ${searched} 个服务中找到匹配的日志${timeNote}` }] };
         }
 
         const output = results.map(r => `### ${r.service} (${r.namespace})\n\`\`\`\n${r.logs}\n\`\`\``).join('\n\n');
@@ -648,7 +714,7 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
         return {
           content: [{
             type: 'text',
-            text: `## TraceId 追踪结果\n\n**traceId**: ${traceId}\n**namespace**: ${targetNamespace || '默认'}\n**匹配服务数**: ${results.length}\n\n${output}`
+            text: `## TraceId 追踪结果\n\n**traceId**: ${traceId}\n**namespace**: ${targetNamespace || '默认'}\n**匹配服务数**: ${results.length}${timeNote}\n\n${output}`
           }]
         };
       }
@@ -741,7 +807,7 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
         };
     }
   } catch (error) {
-    console.error(`[MCP] 错误: ${error.message}`);
+    console.error(`[MCP] 工具内部错误: ${error.message}`);
     return {
       content: [{
         type: 'text',
@@ -750,13 +816,13 @@ server.setRequestHandler(CallToolRequestSchema, async (request) => {
       isError: true
     };
   }
-});
+}
 
 // 启动服务器
 async function main() {
   const transport = new StdioServerTransport();
   await server.connect(transport);
-  console.error('[MCP] Log Query Server v3.1.0 已启动 (支持 Loki 生产环境日志)');
+  console.error('[MCP] Log Query Server v3.2.0 已启动 (支持超时保护 + 进程看门狗)');
 }
 
 main().catch((error) => {

package/loki-client.js

@@ -11,6 +11,9 @@
 
 import { LOKI_ENVIRONMENTS, LOKI_DEFAULTS } from './config.js';
 
+// Loki 查询超时（毫秒）
+const LOKI_FETCH_TIMEOUT = 30000;
+
 // 时间范围自动递进策略（毫秒）
 const AUTO_RANGE_STEPS = [
   { range: 5 * 60 * 1000,            label: '5 分钟' },
@@ -63,11 +66,27 @@ export async function queryLoki(envName, expr, options = {}) {
 
   console.error(`[Loki] 查询: env=${envName}, expr=${expr}`);
 
-  const resp = await fetch(url, {
-    method: 'POST',
-    headers: buildHeaders(env),
-    body: JSON.stringify(body)
-  });
+  // 带超时的 fetch
+  const controller = new AbortController();
+  const timer = setTimeout(() => controller.abort(), LOKI_FETCH_TIMEOUT);
+
+  let resp;
+  try {
+    resp = await fetch(url, {
+      method: 'POST',
+      headers: buildHeaders(env),
+      body: JSON.stringify(body),
+      signal: controller.signal
+    });
+  } catch (e) {
+    clearTimeout(timer);
+    if (e.name === 'AbortError') {
+      throw new Error(`Loki 查询超时（${LOKI_FETCH_TIMEOUT}ms）`);
+    }
+    throw e;
+  } finally {
+    clearTimeout(timer);
+  }
 
   if (!resp.ok) {
     const text = await resp.text();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mcp-log-query-server",
-  "version": "3.2.0",
+  "version": "3.4.0",
   "description": "MCP Server for querying server logs via SSH jump host and Grafana Loki API",
   "main": "index.js",
   "type": "module",