Node.js

Streams process data in chunks instead of loading it all into memory, keeping usage flat for large or unbounded data. There are four types: Readable (source, e.g. a file read), Writable (sink, e.g. a file write), Duplex (both, e.g. a TCP socket), and Transform (a Duplex that modifies chunks, e.g. gzip). pipe connects them and propagates backpressure so a fast source can't overwhelm a slow sink.

Node.js shines for I/O-bound workloads — APIs, proxies, real-time apps — because its single-threaded event loop handles thousands of concurrent connections without blocking while it waits on the network or disk. It is a poor fit for CPU-bound work (heavy computation, image/video processing): a long synchronous task blocks the event loop and stalls all other requests until it finishes.

Both add parallelism beyond the single main thread. worker_threads run JS on threads inside the same process, sharing memory cheaply via SharedArrayBuffer and communicating with low overhead — ideal for CPU-bound work. child_process (spawn/fork/exec) launches separate OS processes with isolated memory, communicating over IPC or pipes — better for running other programs or full isolation. Use a cluster to spread an HTTP server across cores.

The libuv event loop cycles through ordered phases: timers (setTimeout), pending callbacks, poll (I/O), check (setImmediate), and close callbacks. Between every callback it drains two microtask queues: process.nextTick first, then promise callbacks.

Graceful shutdown means that on a termination signal (SIGTERM/SIGINT) the process stops accepting new work, finishes in-flight requests, closes resources (DB pools, sockets, the HTTP server), then exits. It matters because an abrupt exit drops live requests, leaks connections, and can corrupt partial writes. A timeout forces exit if cleanup hangs, so a stuck connection can't block the shutdown forever.