pm2-perfmonitor 2.1.1 → 2.1.3

package/lib/app.js

@@ -54,10 +54,11 @@ const nodeInspectorSampleDuration = parseParamToNumber(
 // 存储每个进程的 CPU 采样历史（pm_id -> [cpu1, cpu2, ...]）
 const zombieCpuHistory = new Map()
 const zombieRestartHistory = new Map()
-const restartFailedHistory = new Map()
+const zombieRestartFailedHistory = new Map()
 
 const cpuOverloadHistory = new Map()
 const cpuOverloadRestartHistory = new Map()
+const cpuOverloadRestartFailedHistory = new Map()
 
 /**
  * @param {'log' | 'info' | 'error' | 'warn'} type
@@ -129,11 +130,25 @@ const setCpuOverloadHistory = (pm_id, appCpuUsage) => {
   return history
 }
 
-const setRestartFailedHistory = (pm_id) => {
-  if (!restartFailedHistory.has(pm_id)) {
-    restartFailedHistory.set(pm_id, 1)
+const setZombieRestartFailedHistory = (pm_id) => {
+  if (!zombieRestartFailedHistory.has(pm_id)) {
+    zombieRestartFailedHistory.set(pm_id, 1)
   } else {
-    restartFailedHistory.set(pm_id, restartFailedHistory.get(pm_id) + 1)
+    zombieRestartFailedHistory.set(
+      pm_id,
+      zombieRestartFailedHistory.get(pm_id) + 1,
+    )
+  }
+}
+
+const setCpuOverloadRestartFailedHistory = (pm_id) => {
+  if (!cpuOverloadRestartFailedHistory.has(pm_id)) {
+    cpuOverloadRestartFailedHistory.set(pm_id, 1)
+  } else {
+    cpuOverloadRestartFailedHistory.set(
+      pm_id,
+      cpuOverloadRestartFailedHistory.get(pm_id) + 1,
+    )
   }
 }
 
@@ -222,7 +237,7 @@ const processChecker = async () => {
               restartErr,
             )
 
-            setRestartFailedHistory(pm_id)
+            setZombieRestartFailedHistory(pm_id)
           }
         }
       }
@@ -271,7 +286,7 @@ const processChecker = async () => {
             restartErr,
           )
 
-          setRestartFailedHistory(pm_id)
+          setCpuOverloadRestartFailedHistory(pm_id)
         }
       }
     }
@@ -323,7 +338,7 @@ const runModule = () => {
     value: () => {
       const res = []
 
-      for (const [k, v] of restartFailedHistory) {
+      for (const [k, v] of zombieRestartFailedHistory) {
         if (v > 0) {
           res.push([k, v])
         }
@@ -385,6 +400,23 @@ const runModule = () => {
       return res.join(',')
     },
   })
+
+  Probe.metric({
+    name: 'CPU Overload Restarts (failed)',
+    value: () => {
+      const res = []
+
+      for (const [k, v] of cpuOverloadRestartFailedHistory) {
+        if (v > 0) {
+          res.push([k, v])
+        }
+      }
+
+      if (!res.length) return 'N/A'
+
+      return res.map((v) => `[${v[0]}]:${v[1]}`).join(' ; ')
+    },
+  })
 }
 
 runModule()

package/lib/perf-sampler.js

@@ -25,10 +25,9 @@ const getExeca = async () => {
 const execCommand = async (cmd, args, options = {}) => {
   try {
     const execa = await getExeca()
-    await execa(cmd, args, {
-      stdio: 'inherit',
-      ...options,
-    })
+
+    await execa(cmd, args, options)
+
     return true
   } catch (err) {
     console.error(`Command failed: ${cmd} ${args.join(' ')}`, err.message)
@@ -122,10 +121,10 @@ const performPerfSampling = async ({
     logger('error', `Perf permission check failed: ${err.message}`)
     logger(
       'error',
-      'Please ensure the perf command is installed and the user has permission to run it.\n' +
-        'You can configure the system to allow non-root perf by setting:\n' +
-        '  echo -1 | sudo tee /proc/sys/kernel/perf_event_paranoid\n' +
-        '  sudo setcap cap_sys_admin+ep $(which perf)',
+      `Please ensure the perf command is installed and the user has permission to run it.\n
+        You can configure the system to allow non-root perf by setting:\n
+          "echo -1 | sudo tee /proc/sys/kernel/perf_event_paranoid"\n
+          "sudo setcap cap_sys_admin+ep $(which perf)"`,
     )
     return // 无权限则直接退出
   }
@@ -189,6 +188,7 @@ const performPerfSampling = async ({
     const scriptOk = await execCommand('perf', ['script', '-i', perfDataFile], {
       stdout: fs.createWriteStream(perfStacksFile),
     })
+
     if (!scriptOk) return
 
     logger('info', `PID:${pidNum} Perf sampling completed: ${perfStacksFile}`)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pm2-perfmonitor",
-  "version": "2.1.1",
+  "version": "2.1.3",
   "description": "A pm2 module for performance monitoring. Automatically detect zombie processes and restart it",
   "author": {
     "name": "elenh",