PM2 Cluster Mode on AWS EC2: A Production Setup Guide

Cluster mode is still a practical option when you need predictable cost and simple operations. The key is disciplined process management, graceful shutdown, and repeatable release scripts.

1) Use an explicit PM2 ecosystem file

1module.exports = {
2  apps: [
3    {
4      name: "api",
5      script: "./dist/server.js",
6      instances: "max",
7      exec_mode: "cluster",
8      listen_timeout: 10000,
9      kill_timeout: 5000,
10      max_memory_restart: "400M",
11      env: {
12        NODE_ENV: "production",
13        PORT: 4000
14      }
15    }
16  ]
17};

2) Graceful shutdown is mandatory

If your service does not stop cleanly, PM2 reload can still drop connections. Listen for SIGINT/SIGTERM and close HTTP server before exit.

1const server = app.listen(process.env.PORT || 4000);
2 
3function gracefulExit(signal) {
4  console.log("received", signal);
5  server.close(() => process.exit(0));
6  setTimeout(() => process.exit(1), 8000);
7}
8 
9process.on("SIGINT", () => gracefulExit("SIGINT"));
10process.on("SIGTERM", () => gracefulExit("SIGTERM"));

3) Keep Nginx config boring and explicit

1server {
2  listen 80;
3  server_name api.dudelemon.com;
4 
5  location / {
6    proxy_pass http://127.0.0.1:4000;
7    proxy_http_version 1.1;
8    proxy_set_header Upgrade $http_upgrade;
9    proxy_set_header Connection "upgrade";
10    proxy_set_header Host $host;
11    proxy_cache_bypass $http_upgrade;
12  }
13}

4) Release script for zero-downtime deploy

1#!/usr/bin/env bash
2set -euo pipefail
3 
4git fetch origin main
5git reset --hard origin/main
6npm ci
7npm run build
8pm2 reload ecosystem.config.cjs --update-env
9pm2 save

5) Infrastructure baseline before first production deploy

• Harden SSH access and disable password authentication.
• Set up CloudWatch/Datadog monitoring for CPU, memory, disk, and error rates.
• Provision TLS certificates and renewal automation.
• Back up environment and deployment secrets outside the instance.

6) Release safety and rollback strategy

A rollback plan must be executable in minutes. Keep last known good artifact references and a scriptable rollback path. If health checks fail after deploy, rollback should be automatic or one command away.

1#!/usr/bin/env bash
2set -euo pipefail
3 
4git fetch origin
5git checkout "$1"   # tag or commit hash
6npm ci
7npm run build
8pm2 reload ecosystem.config.cjs --update-env

7) Observability signals that matter most

• P95 and P99 latency by route.
• Non-2xx response rate over 5-minute windows.
• Process restarts and memory pressure per worker.
• Connection errors between app and dependent services.

PM2 on EC2 FAQ

Q: Should we run one big EC2 instance or multiple smaller nodes? A: Prefer multiple nodes for resilience and controlled failure domains.

Q: Is PM2 enough for all scaling needs? A: PM2 handles process-level scaling well, but infrastructure auto-scaling and load balancing still matter.