Offloading CPU-bound work with Worker Threads

What's really being asked

The question separates candidates who understand that Node's asynchrony only helps I/O from those who think wrapping work in a Promise magically parallelizes it. CPU-bound code blocks the one JavaScript thread regardless of async syntax.

The full answer

Node runs your JavaScript on a single thread. A tight CPU loop, such as hashing, image resizing, or large JSON crunching, monopolizes that thread, so the event loop never advances to other phases. Every other in-flight request, timer, and I/O callback is frozen until the computation returns, which destroys throughput and latency. The correct fix is to move the work to a Worker Thread, which has its own V8 instance and event loop. The main thread posts input data with postMessage, the worker computes, and it posts results back. For shared numeric buffers you can use SharedArrayBuffer to avoid copying. In production you maintain a worker pool so spawning cost is amortized and concurrency is bounded.

The mistakes people make

Claiming async/await or wrapping the function in a Promise solves it, since neither yields the thread during synchronous computation. Suggesting the cluster module as the primary tool when the real need is sharing a result for one request, or spawning an unbounded worker per request, which exhausts memory.

What usually comes next

Worker pool sizing relative to CPU cores, the cost of structured cloning large payloads, when to prefer a separate microservice or child_process, and how SharedArrayBuffer plus Atomics avoids copies.

A concrete example

An endpoint generates a PDF that takes two seconds of pure CPU. Inline, those two seconds block all other users. Instead, a pool of four workers each handle one PDF job; the handler posts the document spec to an idle worker, awaits the result message, and responds. The main loop stays free to accept and route other requests throughout.