Node.js in Production

After this lesson, you will be able to: Run Node in production: process managers (PM2), clustering, memory profiling, graceful shutdown.

Node app + production = these four concerns. Without them you ship hot-mess infra.

Prerequisites:WebSockets

Process managers + clustering

PM2: classic process manager. Restart on crash, log rotation, multiple instances per core. Node's cluster module: built-in multi-process fork (one per CPU). Modern hosts (Docker, Kubernetes, Fargate, Vercel Functions) handle process management themselves, PM2 becomes optional. Default in 2026: one Node process per container, host orchestrator handles scaling.

Graceful shutdown (the must-have)

Without this, deploys drop in-flight requests.

python

import { createServer } from "node:http";
import express from "express";

const app = express();
const server = createServer(app);

server.listen(3000);

let shuttingDown = false;
app.get("/health", (_req, res) => {
  res.status(shuttingDown ? 503 : 200).json({ ok: !shuttingDown });
});

async function shutdown() {
  if (shuttingDown) return;
  shuttingDown = true;
  console.log("draining...");
  server.close(() => {
    console.log("closed");
    process.exit(0);
  });
  setTimeout(() => process.exit(1), 30_000);   // hard exit after 30s
}
process.on("SIGTERM", shutdown);
process.on("SIGINT", shutdown);

Memory + profiling

Memory leaks: usually long-lived listeners not removed; closures holding references; unbounded caches. Tools: `--inspect` flag → Chrome DevTools → Memory + Performance tabs. Heap snapshots: `process.kill -SIGUSR2 <pid>` (with --heap-prof or third-party tools). Production: monitor RSS + heap via process.memoryUsage(); alert on growth trend.

Health checks + observability

Every prod app needs these.

python

// Health endpoint (load balancer probes this)
app.get("/health", async (_req, res) => {
  try {
    await db.$queryRaw`SELECT 1`;
    res.json({ ok: true });
  } catch (e) {
    res.status(503).json({ ok: false, error: "db" });
  }
});

// Request logging (pino is the standard)
import pino from "pino";
const log = pino({ level: process.env.LOG_LEVEL ?? "info" });
app.use((req, _res, next) => {
  log.info({ method: req.method, url: req.url }, "req");
  next();
});

// Metrics: prom-client for Prometheus, OpenTelemetry for traces.

Common mistakes only experienced backend devs avoid

No SIGTERM handler. Deploys drop connections. console.log everywhere. Use pino (structured JSON). Trusting /health that returns 200 when DB is down. Check downstream dependencies. Unbounded in-memory caches. Heap grows forever.

Quick Check

Why is `process.on('SIGTERM', shutdown)` critical?

Pick the cleanest answer.

←WebSockets and Real-Time

Back to Node.js

Testing Node.js→