Worker Threads in Node.js: When and How to Use Them for CPU-Intensive Tasks (2026)

#backend #javascript #node #performance

Your API hashes a password. The event loop blocks. Every other request waits. Node is single-threaded, but it does not have to be.

The Problem

CPU-bound work (hashing, image processing, JSON parsing large payloads, compression) blocks the event loop. While Node processes that work, it cannot handle incoming requests, timers, or I/O callbacks.

Basic Worker Thread

import { Worker, isMainThread, parentPort, workerData } from "worker_threads";
if (isMainThread) {
  const worker = new Worker(__filename, { workerData: { input: "heavy-task" } });
  worker.on("message", (result) => console.log("Result:", result));
  worker.on("error", (err) => console.error("Worker error:", err));
} else {
  // This runs in a separate thread
  const result = heavyComputation(workerData.input);
  parentPort?.postMessage(result);
}

Worker Pool

Creating a new worker per request is expensive. Pool them:

import { Worker } from "worker_threads";
import os from "os";
class WorkerPool {
  private workers: Worker[] = [];
  private queue: Array<{ data: unknown; resolve: Function; reject: Function }> = [];
  private available: Worker[] = [];
  constructor(script: string, size = os.cpus().length) {
    for (let i = 0; i < size; i++) {
      const w = new Worker(script);
      this.workers.push(w);
      this.available.push(w);
    }
  }
  run(data: unknown): Promise<unknown> {
    return new Promise((resolve, reject) => {
      const worker = this.available.pop();
      if (\!worker) { this.queue.push({ data, resolve, reject }); return; }
      worker.once("message", (result) => {
        this.available.push(worker);
        const next = this.queue.shift();
        if (next) this.run(next.data).then(next.resolve, next.reject);
        resolve(result);
      });
      worker.postMessage(data);
    });
  }
}

When to Use Worker Threads

Yes: Password hashing, image/video processing, CSV parsing, compression, encryption, CPU-intensive validation.

No: Database queries, HTTP requests, file I/O. These are already async and non-blocking. Worker threads add overhead without benefit.

SharedArrayBuffer for Zero-Copy

For large data, postMessage copies the data. Use SharedArrayBuffer to share memory between threads without copying. Use Atomics for synchronization.

Common Mistakes

Using workers for I/O: Database calls are already non-blocking
Creating a worker per request: Spawning is expensive, use a pool
Not handling worker crashes: Always listen for error and exit events
Too many workers: More workers than CPU cores causes context switching overhead

Part of my Production Backend Patterns series. Follow for more practical backend engineering.