pm2-perfmonitor 2.3.1 → 2.4.0

package/lib/app.js

@@ -12,6 +12,7 @@ const { defaultOptions } = require('./defaults')
 const { sendMessage } = require('./message')
 const { performPerfSampling } = require('./perf-sampler')
 const { sendAlert } = require('./alert')
+const { getJobConfHostName } = require('./job-conf')
 
 const conf = pmx.initModule({}, (err, incomingConf) => {
   if (err) {
@@ -56,6 +57,7 @@ const enableAlert = parseBool(conf.enableAlert)
 const alertCmdPath = conf.alertCmdPath
 const alertEnv = conf.alertEnv
 const alertLevel = conf.alertLevel
+const jobHostNameConfPath = conf.jobHostNameConfPath
 
 // 存储每个进程的 CPU 采样历史（pm_id -> [cpu1, cpu2, ...]）
 const zombieCpuHistory = new Map()
@@ -169,17 +171,20 @@ const setCpuOverloadRestartFailedHistory = (pm_id) => {
 /**
  * 发送重启警告
  * @param {string} title
- * @param {string} content
+ * @param {string} message
  */
-const sendRestartAlert = async (title, content) => {
+const sendRestartAlert = async (title, message) => {
   if (!enableAlert) return
 
+  const datetime = new Date().toLocaleString()
+  const jobHostName = getJobConfHostName(jobHostNameConfPath)
+
   return await sendAlert({
     cmd: alertCmdPath,
     env: alertEnv,
     level: alertLevel,
-    title: `[${MODULE_NAME}] Alert: ${title}`,
-    content,
+    title: `[${MODULE_NAME}] [${datetime}] Alert: ${title}`,
+    content: `[${jobHostName}] - ${message}`,
   })
 }
 
@@ -197,12 +202,7 @@ const processChecker = async () => {
     for (const app of apps) {
       const { name, pid, pm_id, monit, pm2_env } = app
 
-      const sysCpuUsage = await getSysCpuUsageByPid(pid)
-      const pm2CpuUsage = monit?.cpu
-
       const appStatus = pm2_env?.status
-      const appCpuUsage =
-        typeof sysCpuUsage === 'number' ? sysCpuUsage : pm2CpuUsage
 
       // 非目标应用，跳过
       if (
@@ -222,6 +222,11 @@ const processChecker = async () => {
         continue
       }
 
+      const pm2CpuUsage = monit?.cpu
+      const sysCpuUsage = await getSysCpuUsageByPid(pid)
+      const appCpuUsage =
+        typeof sysCpuUsage === 'number' ? sysCpuUsage : pm2CpuUsage
+
       const history = setZombieCpuHistory(pm_id, appCpuUsage)
       const history2 = setCpuOverloadHistory(pm_id, appCpuUsage)
 
@@ -269,10 +274,7 @@ const processChecker = async () => {
 
             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`,
+              `appName: ${name}, pid: ${pid}, pm_id: ${pm_id}, 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
@@ -331,10 +333,7 @@ const processChecker = async () => {
 
           await sendRestartAlert(
             `CPU overload process restarted!`,
-            `appName: ${name}, \n 
-              pid: ${pid}, \n 
-              pm_id: ${pm_id}, \n 
-              restarted: ${cpuOverloadRestartHistory.get(pm_id)} times`,
+            `appName: ${name}, pid: ${pid}, pm_id: ${pm_id}, restarted: ${cpuOverloadRestartHistory.get(pm_id)} times`,
           )
 
           cpuOverloadHistory.delete(pm_id)

package/lib/defaults.js

@@ -95,6 +95,8 @@ const defaultOptions = {
    * 指定报警级别
    */
   alertLevel: 'Sev-2',
+
+  jobHostNameConfPath: '/var/job/host_name.conf',
 }
 
 module.exports = {

package/lib/job-conf.js

@@ -0,0 +1,39 @@
+import fs from 'node:fs'
+import path from 'node:path'
+
+/**
+ * @param {string} filePath
+ * @returns { { success: boolean, data: string | null }}
+ */
+const readFileContent = (filePath) => {
+  try {
+    const absolutePath = path.resolve(filePath)
+    const data = fs.readFileSync(absolutePath, 'utf8')
+    return { success: true, data }
+  } catch {
+    return {
+      success: false,
+      data: null,
+    }
+  }
+}
+
+/**
+ * @param {string} path
+ */
+const getJobConfHostName = (path) => {
+  const result = readFileContent(path)
+
+  if (result.success && result.data) {
+    const arr = result.data.split('=')
+
+    return arr[1]
+  }
+
+  return 'unknown'
+}
+
+module.exports = {
+  readFileContent,
+  getJobConfHostName,
+}

package/lib/utils.js

@@ -45,9 +45,12 @@ const sleepAsync = (duration = 0) => {
  * @returns { Promise<number> } CPU 使用率
  */
 const getSysCpuUsageByPid = async (pid) => {
-  const stats = await pidusage(pid)
-
-  return stats.cpu
+  try {
+    const stats = await pidusage(pid)
+    return stats.cpu
+  } catch (err) {
+    console.error('Call pidusage error:', err)
+  }
 }
 
 module.exports = {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pm2-perfmonitor",
-  "version": "2.3.1",
+  "version": "2.4.0",
   "description": "A pm2 module for performance monitoring. Automatically detect zombie processes and restart it",
   "author": {
     "name": "elenh",