feat: add SSH metrics collection with ssh2 library and auto-scheduler

Replace sshpass with ssh2 Node.js library for reliable SSH connections. Add all 6 servers (CWP, EasyPanel, MCP Hub, Meet, WhatsApp, WhatSMS). Add 5-minute auto-collection scheduler in production mode. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-07 22:14:22 +00:00
parent 37164cf2ac
commit 10fc8f5ccc
4 changed files with 255 additions and 107 deletions
--- a/api/server.ts
+++ b/api/server.ts
@@ -13,6 +13,7 @@ import diagnosticRouter from './routes/diagnostic.js'
 import hetznerRouter from './routes/hetzner.js'
 import wpMonitorRouter from './routes/wp-monitor.js'
 import serverMetricsRouter from './routes/server-metrics.js'
+import { collectAllServerMetrics } from './services/server-metrics.js'

 const __filename = fileURLToPath(import.meta.url)
 const __dirname = path.dirname(__filename)
@@ -65,10 +66,29 @@ app.use((err: any, _req: express.Request, res: express.Response, _next: express.
 // Start server
 app.listen(PORT, () => {
  console.log('='.repeat(50))
-  console.log(`🚀 API Server running on http://localhost:${PORT}`)
-  console.log(`📊 Dashboard: http://localhost:${PORT}/api/dashboard`)
-  console.log(`🔍 Monitor: http://localhost:${PORT}/api/monitor`)
-  console.log(`🔧 Diagnostic: http://localhost:${PORT}/api/diagnostic`)
-  console.log(`☁️  Hetzner: http://localhost:${PORT}/api/hetzner`)
+  console.log(`API Server running on http://localhost:${PORT}`)
+  console.log(`Dashboard: http://localhost:${PORT}/api/dashboard`)
+  console.log(`Monitor: http://localhost:${PORT}/api/monitor`)
+  console.log(`Hetzner: http://localhost:${PORT}/api/hetzner`)
  console.log('='.repeat(50))
+
+  // Auto-collect server metrics every 5 minutes
+  if (isProduction) {
+    const INTERVAL = 5 * 60 * 1000
+    console.log('[SCHEDULER] Server metrics collection every 5min')
+
+    // Initial collection after 30s (let server stabilize)
+    setTimeout(() => {
+      collectAllServerMetrics().catch(err =>
+        console.error('[SCHEDULER] Initial collection failed:', err)
+      )
+    }, 30000)
+
+    // Recurring collection
+    setInterval(() => {
+      collectAllServerMetrics().catch(err =>
+        console.error('[SCHEDULER] Collection failed:', err)
+      )
+    }, INTERVAL)
+  }
 })
--- a/api/services/server-metrics.ts
+++ b/api/services/server-metrics.ts
@@ -1,18 +1,16 @@
 /**
 * Server Metrics Collector Service
- * Recolhe métricas de todos os servidores (Hetzner API + SSH)
+ * Recolhe métricas de todos os servidores via SSH (ssh2 library)
 * @author Descomplicar® | @link descomplicar.pt | @copyright 2026
 */
 import db from '../db.js'
-import { collectAllMetrics as collectHetznerMetrics } from './hetzner.js'
+import { Client } from 'ssh2'

-// Hetzner API Configuration (used by hetzner.ts service)
-
-// SSH Configuration (from MCP ssh-unified)
 interface SSHServer {
  name: string
  monitorName: string
  host: string
+  port: number
  user: string
  pass: string
 }
@@ -22,6 +20,7 @@ const SSH_SERVERS: SSHServer[] = [
    name: 'server',
    monitorName: 'CWP Server',
    host: process.env.SERVER_HOST || '176.9.3.158',
+    port: parseInt(process.env.SERVER_PORT || '9443'),
    user: process.env.SERVER_USER || 'root',
    pass: process.env.SERVER_PASS || ''
  },
@@ -29,19 +28,44 @@ const SSH_SERVERS: SSHServer[] = [
    name: 'easy',
    monitorName: 'EasyPanel',
    host: process.env.EASY_HOST || '178.63.18.51',
+    port: 22,
    user: process.env.EASY_USER || 'root',
    pass: process.env.EASY_PASS || ''
+  },
+  {
+    name: 'mcp-hub',
+    monitorName: 'MCP Hub',
+    host: process.env.MCPHUB_HOST || 'mcp-hub.descomplicar.pt',
+    port: 22,
+    user: process.env.MCPHUB_USER || 'root',
+    pass: process.env.MCPHUB_PASS || ''
+  },
+  {
+    name: 'meet',
+    monitorName: 'Meet',
+    host: process.env.MEET_HOST || 'meet.descomplicar.pt',
+    port: 22,
+    user: process.env.MEET_USER || 'root',
+    pass: process.env.MEET_PASS || ''
+  },
+  {
+    name: 'whatsapp',
+    monitorName: 'WhatsApp',
+    host: process.env.WHATSAPP_HOST || 'whatsapp.descomplicar.pt',
+    port: 22,
+    user: process.env.WHATSAPP_USER || 'root',
+    pass: process.env.WHATSAPP_PASS || ''
+  },
+  {
+    name: 'whatsms',
+    monitorName: 'WhatSMS',
+    host: process.env.WHATSMS_HOST || 'whatsms.descomplicar.pt',
+    port: 22,
+    user: process.env.WHATSMS_USER || 'root',
+    pass: process.env.WHATSMS_PASS || ''
  }
 ]

-// Hetzner server mapping to monitoring table
-const HETZNER_MAPPING: Record<string, string> = {
-  'gateway': 'MCP Hub',
-  'meet': 'Meet',
-  'whatsapp.descomplicar.pt': 'WhatsApp',
-  'whatsms': 'WhatSMS'
-}
-
 interface ServerMetrics {
  cpu: number
  ram: number
@@ -50,9 +74,6 @@ interface ServerMetrics {
  containers?: number
 }

-/**
- * Parse SSH metrics output
- */
 function parseSSHMetrics(output: string): ServerMetrics {
  const lines = output.split('\n')
  const metrics: ServerMetrics = { cpu: 0, ram: 0, disk: 0, load: 0 }
@@ -69,29 +90,67 @@ function parseSSHMetrics(output: string): ServerMetrics {
 }

 /**
- * Execute SSH command via subprocess (since we can't use MCP from here)
- * This is a placeholder - in production, use the MCP or a proper SSH library
+ * Execute SSH command via ssh2 library
 */
-async function executeSSH(server: SSHServer, command: string): Promise<string> {
-  // For now, we'll use the database values or call an external endpoint
-  // In production, you'd use ssh2 library or call the MCP endpoint
-  const { exec } = await import('child_process')
-  const { promisify } = await import('util')
-  const execAsync = promisify(exec)
+function executeSSH(server: SSHServer, command: string): Promise<string> {
+  return new Promise((resolve, reject) => {
+    const conn = new Client()
+    let output = ''
+    const timeout = setTimeout(() => {
+      conn.end()
+      reject(new Error(`SSH timeout for ${server.name}`))
+    }, 20000)

-  try {
-    // Use sshpass for password-based auth (not ideal but works for internal servers)
-    const sshCommand = `sshpass -p '${server.pass}' ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10 ${server.user}@${server.host} "${command}"`
-    const { stdout } = await execAsync(sshCommand, { timeout: 30000 })
-    return stdout
-  } catch (error) {
-    console.error(`SSH error for ${server.name}:`, error)
-    return ''
-  }
+    conn.on('ready', () => {
+      conn.exec(command, (err, stream) => {
+        if (err) {
+          clearTimeout(timeout)
+          conn.end()
+          reject(err)
+          return
+        }
+        stream.on('data', (data: Buffer) => {
+          output += data.toString()
+        })
+        stream.on('close', () => {
+          clearTimeout(timeout)
+          conn.end()
+          resolve(output)
+        })
+        stream.stderr.on('data', () => {
+          // ignore stderr
+        })
+      })
+    })
+
+    conn.on('error', (err) => {
+      clearTimeout(timeout)
+      reject(err)
+    })
+
+    conn.connect({
+      host: server.host,
+      port: server.port,
+      username: server.user,
+      password: server.pass,
+      readyTimeout: 15000,
+      algorithms: {
+        kex: [
+          'ecdh-sha2-nistp256',
+          'ecdh-sha2-nistp384',
+          'ecdh-sha2-nistp521',
+          'diffie-hellman-group-exchange-sha256',
+          'diffie-hellman-group14-sha256',
+          'diffie-hellman-group14-sha1',
+          'diffie-hellman-group1-sha1'
+        ]
+      }
+    })
+  })
 }

 /**
- * Collect metrics from SSH servers (CWP, EasyPanel)
+ * Collect metrics from all SSH servers
 */
 export async function collectSSHMetrics(): Promise<{ success: number; failed: number }> {
  let success = 0
@@ -100,6 +159,12 @@ export async function collectSSHMetrics(): Promise<{ success: number; failed: nu
  const metricsCommand = `echo "CPU:$(top -bn1 | grep 'Cpu(s)' | awk '{print $2}')"; echo "MEM:$(free -m | awk '/Mem:/ {printf "%.1f", $3/$2*100}')"; echo "DISK:$(df -h / | awk 'NR==2 {print $5}' | tr -d '%')"; echo "LOAD:$(cat /proc/loadavg | awk '{print $1}')"; echo "CONTAINERS:$(docker ps -q 2>/dev/null | wc -l || echo 0)"`

  for (const server of SSH_SERVERS) {
+    if (!server.pass) {
+      console.log(`[SSH] Skipping ${server.name}: no password configured`)
+      failed++
+      continue
+    }
+
    try {
      const output = await executeSSH(server, metricsCommand)
      if (!output) {
@@ -127,27 +192,30 @@ export async function collectSSHMetrics(): Promise<{ success: number; failed: nu

      // Update containers if EasyPanel
      if (server.name === 'easy' && metrics.containers !== undefined) {
-        // Get current container stats
-        const containerOutput = await executeSSH(server, 'docker ps -a --format "{{.Status}}" | grep -c "Up" || echo 0; docker ps -aq | wc -l')
-        const [up, total] = containerOutput.trim().split('\n').map(n => parseInt(n) || 0)
-        const down = total - up
+        try {
+          const containerOutput = await executeSSH(server, 'docker ps -a --format "{{.Status}}" | grep -c "Up" || echo 0; docker ps -aq | wc -l')
+          const [up, total] = containerOutput.trim().split('\n').map(n => parseInt(n) || 0)
+          const down = total - up

-        await db.query(
-          `UPDATE tbl_eal_monitoring
-           SET details = ?, status = ?, last_check = NOW()
-           WHERE category = 'container' AND name = 'EasyPanel Containers'`,
-          [
-            JSON.stringify({ total, up, down, restarting: 0 }),
-            down > 0 ? 'warning' : 'ok'
-          ]
-        )
+          await db.query(
+            `UPDATE tbl_eal_monitoring
+             SET details = ?, status = ?, last_check = NOW()
+             WHERE category = 'container' AND name = 'EasyPanel Containers'`,
+            [
+              JSON.stringify({ total, up, down, restarting: 0 }),
+              down > 0 ? 'warning' : 'ok'
+            ]
+          )
+        } catch {
+          // Container stats are optional
+        }
      }

      success++
-      console.log(`✅ ${server.monitorName}: CPU=${metrics.cpu}%, RAM=${metrics.ram}%, Disk=${metrics.disk}%`)
+      console.log(`[SSH] ${server.monitorName}: CPU=${metrics.cpu}%, RAM=${metrics.ram}%, Disk=${metrics.disk}%`)

    } catch (error) {
-      console.error(`❌ Failed to collect metrics from ${server.name}:`, error)
+      console.error(`[SSH] Failed ${server.name}:`, (error as Error).message)
      failed++
    }
  }
@@ -156,62 +224,20 @@ export async function collectSSHMetrics(): Promise<{ success: number; failed: nu
 }

 /**
- * Sync Hetzner metrics to monitoring table
- */
-export async function syncHetznerToMonitoring(): Promise<number> {
-  // First collect fresh Hetzner metrics
-  await collectHetznerMetrics()
-
-  // Then sync to monitoring table
-  let synced = 0
-
-  for (const [hetznerName, monitorName] of Object.entries(HETZNER_MAPPING)) {
-    const namePattern = hetznerName.includes('.') ? hetznerName : `${hetznerName}%`
-
-    const [result] = await db.query<any[]>(`
-      UPDATE tbl_eal_monitoring m
-      JOIN v_eal_hetzner_latest h ON h.name LIKE ?
-      SET m.details = JSON_OBJECT(
-          'cpu', ROUND(h.cpu_percent, 1),
-          'ram', 0,
-          'disk', 0,
-          'load', ROUND(h.cpu_percent / 25, 2),
-          'network_in', h.network_in_bps,
-          'network_out', h.network_out_bps,
-          'hetzner_id', h.hetzner_id
-      ),
-      m.status = 'up',
-      m.last_check = NOW()
-      WHERE m.category = 'server' AND m.name = ?
-    `, [namePattern, monitorName])
-
-    if ((result as any).affectedRows > 0) synced++
-  }
-
-  return synced
-}
-
-/**
- * Collect all server metrics (Hetzner + SSH)
+ * Collect all server metrics (SSH only - replaces Hetzner sync)
 */
 export async function collectAllServerMetrics(): Promise<{
-  hetzner: { success: number; failed: number }
  ssh: { success: number; failed: number }
-  synced: number
 }> {
-  console.log('[METRICS] Starting server metrics collection...')
-
-  // Collect SSH metrics (CWP, EasyPanel)
+  console.log('[METRICS] Collecting server metrics via SSH...')
  const ssh = await collectSSHMetrics()
-  console.log(`[SSH] ${ssh.success} OK, ${ssh.failed} failed`)
-
-  // Collect and sync Hetzner metrics
-  const synced = await syncHetznerToMonitoring()
-  console.log(`[HETZNER] ${synced} servers synced to monitoring`)
-
-  return {
-    hetzner: { success: synced, failed: 4 - synced },
-    ssh,
-    synced
-  }
+  console.log(`[METRICS] Done: ${ssh.success} OK, ${ssh.failed} failed`)
+  return { ssh }
+}
+
+// Keep for backward compatibility with routes
+export async function syncHetznerToMonitoring(): Promise<number> {
+  // Now handled by SSH collection
+  const result = await collectSSHMetrics()
+  return result.success
 }