pm2-perfmonitor 2.1.6 → 2.3.0

package/lib/alert.js

@@ -0,0 +1,29 @@
+const { getExeca } = require('./execa-helper')
+
+/**
+ * 发送警告
+ * @param { object } options
+ * @param { string } options.cmd - bash 脚本 path（默认值：/var/job/alert.sh）
+ * @param { string } options.env - 环境（默认值: prod）
+ * @param { string } options.level - 报警级别（默认值: Sev-2）
+ * @param { string } options.title - 报警标题
+ * @param { string } options.content - 报警正文
+ */
+const sendAlert = async (options) => {
+  try {
+    const execa = await getExeca()
+
+    const args = [options.env, options.level, options.title, options.content]
+
+    await execa(options.cmd, args)
+
+    return true
+  } catch (err) {
+    console.error('[Send Alert Error]:', err)
+    return false
+  }
+}
+
+module.exports = {
+  sendAlert,
+}

package/lib/app.js

@@ -6,10 +6,12 @@ const {
   parseParamToNumber,
   parseBool,
   sleepAsync,
+  getSysCpuUsageByPid,
 } = require('./utils')
 const { defaultOptions } = require('./defaults')
 const { sendMessage } = require('./message')
 const { performPerfSampling } = require('./perf-sampler')
+const { sendAlert } = require('./alert')
 
 const conf = pmx.initModule({}, (err, incomingConf) => {
   if (err) {
@@ -50,6 +52,10 @@ const enableNodeInspectorCollection = parseBool(
 const nodeInspectorSampleDuration = parseParamToNumber(
   conf.nodeInspectorSampleDuration,
 )
+const enableAlert = parseBool(conf.enableAlert)
+const alertCmdPath = conf.alertCmdPath
+const alertEnv = conf.alertEnv
+const alertLevel = conf.alertLevel
 
 // 存储每个进程的 CPU 采样历史（pm_id -> [cpu1, cpu2, ...]）
 const zombieCpuHistory = new Map()
@@ -60,6 +66,8 @@ const cpuOverloadHistory = new Map()
 const cpuOverloadRestartHistory = new Map()
 const cpuOverloadRestartFailedHistory = new Map()
 
+let isProcessCheckerRunning = false
+
 /**
  * perf 样本是否采集中
  * @type { Map<number,boolean> }
@@ -158,7 +166,22 @@ const setCpuOverloadRestartFailedHistory = (pm_id) => {
   }
 }
 
-let isProcessCheckerRunning = false
+/**
+ * 发送重启警告
+ * @param {string} title
+ * @param {string} content
+ */
+const sendRestartAlert = async (title, content) => {
+  if (!enableAlert) return
+
+  return await sendAlert({
+    cmd: alertCmdPath,
+    env: alertEnv,
+    level: alertLevel,
+    title: `[${MODULE_NAME}] Alert: ${title}`,
+    content,
+  })
+}
 
 /**
  * check process
@@ -172,10 +195,12 @@ const processChecker = async () => {
     const apps = await listAppsAsync()
 
     for (const app of apps) {
-      const { name, pm_id, monit, pm2_env, pid } = app
+      const { name, pid, pm_id, monit, pm2_env } = app
+
+      const sysCpuUsage = await getSysCpuUsageByPid(pid)
 
       const appStatus = pm2_env?.status
-      const appCpuUsage = monit?.cpu || 0
+      const appCpuUsage = sysCpuUsage
 
       // 非目标应用，跳过
       if (
@@ -240,6 +265,14 @@ const processChecker = async () => {
               `[ZOMBIE] Restarted ${name} (pm_id: ${pm_id}) successfully!!! Restarted ${zombieRestartHistory.get(pm_id)} times`,
             )
 
+            await sendRestartAlert(
+              `The zombie process has been restarted!`,
+              `appName: ${name}, \n 
+                pid: ${pid}, \n 
+                pm_id: ${pm_id}, \n 
+                restarted: ${zombieRestartHistory.get(pm_id)} times`,
+            )
+
             // 重启后清除该进程的历史记录，避免刚重启又被判定为僵尸
             zombieCpuHistory.delete(pm_id)
           } catch (restartErr) {
@@ -294,6 +327,14 @@ const processChecker = async () => {
             `[CPU OVERLOAD] Restarted ${name} (pm_id: ${pm_id}) successfully!!! Restarted ${cpuOverloadRestartHistory.get(pm_id)} times`,
           )
 
+          await sendRestartAlert(
+            `CPU overload process restarted!`,
+            `appName: ${name}, \n 
+              pid: ${pid}, \n 
+              pm_id: ${pm_id}, \n 
+              restarted: ${cpuOverloadRestartHistory.get(pm_id)} times`,
+          )
+
           cpuOverloadHistory.delete(pm_id)
         } catch (restartErr) {
           logger(

package/lib/defaults.js

@@ -78,6 +78,23 @@ const defaultOptions = {
    * node:inspector 性能采集持续时间 (s)
    */
   nodeInspectorSampleDuration: 10,
+
+  /**
+   * 指定是否开启报警
+   */
+  enableAlert: false,
+  /**
+   * 指定报警执行 bash 脚本位置
+   */
+  alertCmdPath: '/var/job/alert.sh',
+  /**
+   * 指定报警环境类别
+   */
+  alertEnv: 'prod',
+  /**
+   * 指定报警级别
+   */
+  alertLevel: 'Sev-2',
 }
 
 module.exports = {

package/lib/execa-helper.js

@@ -0,0 +1,17 @@
+let execaCommandCache
+
+/**
+ * 获取 execa 函数（缓存）
+ * @returns { import('execa')['execa'] }
+ */
+const getExeca = async () => {
+  if (!execaCommandCache) {
+    const execaModule = await import('execa')
+    execaCommandCache = execaModule.execa
+  }
+  return execaCommandCache
+}
+
+module.exports = {
+  getExeca,
+}

package/lib/perf-sampler.js

@@ -1,19 +1,6 @@
 const fs = require('fs-extra')
 const path = require('path')
-
-let execaCommandCache
-
-/**
- * 获取 execa 函数（缓存）
- * @returns { import('execa')['execa'] }
- */
-const getExeca = async () => {
-  if (!execaCommandCache) {
-    const execaModule = await import('execa')
-    execaCommandCache = execaModule.execa
-  }
-  return execaCommandCache
-}
+const { getExeca } = require('./execa-helper')
 
 /**
  * 执行命令（不通过 shell，直接使用参数数组）

package/lib/utils.js

@@ -1,3 +1,5 @@
+const pidusage = require('pidusage')
+
 const parseParamToArray = (value, defaultVal = []) => {
   if (Array.isArray(value)) return value
 
@@ -37,9 +39,21 @@ const sleepAsync = (duration = 0) => {
   })
 }
 
+/**
+ * 获取指定进程的CPU使用率
+ * @param {string| number} pid
+ * @returns { Promise<number> } CPU 使用率
+ */
+const getSysCpuUsageByPid = async (pid) => {
+  const stats = await pidusage(pid)
+
+  return stats.cpu
+}
+
 module.exports = {
   parseParamToArray,
   parseParamToNumber,
   parseBool,
   sleepAsync,
+  getSysCpuUsageByPid,
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pm2-perfmonitor",
-  "version": "2.1.6",
+  "version": "2.3.0",
   "description": "A pm2 module for performance monitoring. Automatically detect zombie processes and restart it",
   "author": {
     "name": "elenh",
@@ -44,6 +44,7 @@
   "dependencies": {
     "execa": "^9.6.1",
     "fs-extra": "^11.3.4",
+    "pidusage": "^4.0.1",
     "pm2": "latest",
     "pmx": "latest"
   },
@@ -52,4 +53,4 @@
     "cz-conventional-changelog": "^3.3.0",
     "minimist": "^1.2.8"
   }
-}
+}