スループット向け再利用可能なワーカープールの構築
負荷の高い状況でCPU使用率を最大化するため、スレッドを再利用するタスクキュー型ワーカープールを設計します。
「スループット向け再利用可能なワーカープールの構築」はCoddyKit上の無料Node.js Backend Development Bootcampレッスンです。 これはレッスン4/4です。 下記で完全なレッスンを無料で読むことができます。その後、ブラウザ内の組み込みコードエディタと24時間対応のAIチューターでハンズオン演習できます。 これはNode.js Backend Development Bootcamp学習パスの一部であり、ウェブとCoddyKitアプリ全体で進捗が同期されます。 Node.js Backend Development Bootcampコースには全4レッスンが含まれています。
このレッスンの一部はまだ翻訳されておらず、英語で表示されています。
Why a Worker Pool?
Node.js runs your JavaScript on a single event-loop thread. That is great for I/O, but a CPU-bound task (hashing, image resizing, parsing, compression) blocks the loop and stalls every other request.
The worker_threads module lets you run JavaScript on separate OS threads. But spawning a brand-new Worker for every task is wasteful: thread startup costs tens of milliseconds and memory.
- Goal: create a fixed set of long-lived workers once.
- Recycle them across many tasks via a queue.
- Maximize throughput by keeping every CPU core busy.
That recycled, queue-backed set of workers is a worker pool.
The Blocking Problem
Before building the pool, feel the pain. A synchronous CPU loop on the main thread freezes everything: timers, HTTP responses, even a simple setInterval heartbeat.
Run this and watch the heartbeat go silent while fib(42) burns the CPU.
function fib(n) {
return n < 2 ? n : fib(n - 1) + fib(n - 2);
}
let ticks = 0;
const timer = setInterval(() => {
console.log('heartbeat', ++ticks);
if (ticks >= 3) clearInterval(timer);
}, 50);
console.log('start blocking work');
console.log('fib(38) =', fib(38)); // blocks the event loop
console.log('done blocking work');A Single Worker Thread
The fix is to move CPU work off the main thread. A worker can be defined in the same file using isMainThread to branch behavior.
isMainThreadistruein the parent,falseinside the worker.parentPortis the message channel back to the parent.new Worker(__filename)re-runs this file on a new thread.
This is one worker for one task. The pool will generalize it.
const { Worker, isMainThread, parentPort } = require('node:worker_threads');
if (isMainThread) {
const worker = new Worker(__filename);
worker.on('message', (result) => {
console.log('result =', result);
worker.terminate();
});
worker.postMessage(40);
} else {
parentPort.on('message', (n) => {
const fib = (x) => (x < 2 ? x : fib(x - 1) + fib(x - 2));
parentPort.postMessage(fib(n));
});
}Designing the Pool's Pieces
A reusable pool needs four moving parts that work together:
- Workers array — a fixed number of long-lived threads, usually
os.cpus().length. - Idle list — workers ready to accept a task right now.
- Task queue — pending tasks waiting for a free worker.
- Pending map — links each busy worker to the Promise it must resolve.
The core invariant: a queued task only runs when an idle worker exists; when a worker finishes, it pulls the next task or returns to the idle list.
The Worker Script (worker.js)
Keep the worker logic in its own file so the pool can spawn many copies of it. The worker listens for messages, computes, and posts a structured reply that distinguishes success from error.
Always wrap the work in try/catch so a thrown error becomes a message rather than a crashed thread.
// worker.js
const { parentPort } = require('node:worker_threads');
function heavyTask(n) {
const fib = (x) => (x < 2 ? x : fib(x - 1) + fib(x - 2));
return fib(n);
}
parentPort.on('message', ({ id, payload }) => {
try {
const result = heavyTask(payload);
parentPort.postMessage({ id, result });
} catch (err) {
parentPort.postMessage({ id, error: err.message });
}
});Pool Skeleton: Spawning Workers
The pool constructor spawns N workers up front and tracks which are idle. Each task carries a unique id so replies map back to the right Promise.
Note the _tagWorker helper attaches a per-worker message/error listener exactly once, not once per task.
const { Worker } = require('node:worker_threads');
const os = require('node:os');
class WorkerPool {
constructor(workerPath, size = os.cpus().length) {
this.workerPath = workerPath;
this.idle = [];
this.queue = [];
this.pending = new Map(); // id -> { resolve, reject }
this.nextId = 0;
for (let i = 0; i < size; i++) this._spawn();
}
_spawn() {
const worker = new Worker(this.workerPath);
worker.on('message', (msg) => this._onResult(worker, msg));
worker.on('error', (err) => this._onError(worker, err));
this.idle.push(worker);
}
}Submitting Tasks and the Queue
run() returns a Promise and pushes a task onto the queue, then calls _dispatch(). Dispatch pairs a queued task with an idle worker; if none is free, the task simply waits.
- If
idleis empty, the task stays queued — no work is lost. - When a worker frees up, it drains the next queued task automatically.
This back-pressure is what keeps the pool stable under bursty load.
run(payload) {
return new Promise((resolve, reject) => {
const id = this.nextId++;
this.pending.set(id, { resolve, reject });
this.queue.push({ id, payload });
this._dispatch();
});
}
_dispatch() {
if (this.queue.length === 0 || this.idle.length === 0) return;
const worker = this.idle.pop();
const task = this.queue.shift();
worker._currentId = task.id;
worker.postMessage(task);
}Recycling: Handling Results
This is the heart of recycling. When a worker posts a result, the pool resolves the matching Promise, returns the worker to the idle list, and immediately tries to dispatch the next queued task.
The same worker handles task after task — no respawn — which is exactly what maximizes throughput.
_onResult(worker, msg) {
const { id, result, error } = msg;
const job = this.pending.get(id);
this.pending.delete(id);
worker._currentId = null;
this.idle.push(worker); // recycle the worker
this._dispatch(); // pull the next queued task
if (!job) return;
if (error) job.reject(new Error(error));
else job.resolve(result);
}Recycling on Failure
A worker can crash (uncaught exception, OOM). If you only handle message, a dead worker silently shrinks your pool and its in-flight Promise hangs forever.
On error, reject the in-flight task and respawn a replacement so the pool keeps its size. This self-healing behavior is essential for long-running services.
_onError(worker, err) {
const id = worker._currentId;
if (id != null && this.pending.has(id)) {
this.pending.get(id).reject(err);
this.pending.delete(id);
}
// remove the dead worker, keep pool size constant
this.idle = this.idle.filter((w) => w !== worker);
worker.terminate();
this._spawn();
this._dispatch();
}
async destroy() {
await Promise.all(this.idle.map((w) => w.terminate()));
}A Complete, Runnable Pool
Putting it together in a single file using isMainThread branching so it runs standalone. The pool fans 8 tasks across the available cores and resolves each via a Promise.
Notice every task resolves even though there are fewer workers than tasks — the queue handles the overflow.
const { Worker, isMainThread, parentPort } = require('node:worker_threads');
const os = require('node:os');
if (!isMainThread) {
const fib = (x) => (x < 2 ? x : fib(x - 1) + fib(x - 2));
parentPort.on('message', ({ id, payload }) => {
parentPort.postMessage({ id, result: fib(payload) });
});
} else {
class Pool {
constructor(size) {
this.idle = []; this.queue = []; this.pending = new Map(); this.id = 0;
for (let i = 0; i < size; i++) this._spawn();
}
_spawn() {
const w = new Worker(__filename);
w.on('message', ({ id, result }) => {
this.pending.get(id).resolve(result);
this.pending.delete(id);
this.idle.push(w); this._dispatch();
});
this.idle.push(w);
}
_dispatch() {
if (!this.queue.length || !this.idle.length) return;
const w = this.idle.pop(); const t = this.queue.shift();
w.postMessage(t);
}
run(payload) {
return new Promise((resolve) => {
const id = this.id++;
this.pending.set(id, { resolve });
this.queue.push({ id, payload }); this._dispatch();
});
}
destroy() { this.idle.forEach((w) => w.terminate()); }
}
const pool = new Pool(Math.min(4, os.cpus().length));
const jobs = [30, 31, 32, 33, 30, 31, 32, 33];
Promise.all(jobs.map((n) => pool.run(n))).then((results) => {
console.log('results:', results);
pool.destroy();
});
}Sizing and Tuning for Throughput
Pool size is a real decision, not a guess:
- CPU-bound work: size = number of physical cores (
os.cpus().length). More threads than cores just adds context-switch overhead. - Mixed work: a few extra workers can hide occasional I/O waits, but measure first.
- Transfer cost: large payloads serialize via structured clone; for big buffers use
postMessage(buf, [buf])to transfer ownership and avoid copying.
Always benchmark with realistic load. Throughput, not thread count, is the metric that matters.
const buf = new Uint8Array(1024 * 1024).fill(7);
// Transfer the buffer instead of copying it (zero-copy handoff)
worker.postMessage({ id, payload: buf }, [buf.buffer]);
// After transfer, buf is detached in the sender: buf.byteLength === 0Quick Check
Test your understanding of the pool's recycling design.
Recap
You designed a reusable worker pool that turns CPU-bound work into parallel throughput:
- Why: CPU-bound tasks block Node's single event loop;
worker_threadsmoves them to OS threads. - Pieces: a fixed workers array, an idle list, a task queue, and a pending map keyed by task id.
- Recycling: finished workers return to the idle list and pull the next queued task — no per-task respawn.
- Resilience: handle the
errorevent to reject the in-flight task and respawn a replacement so pool size stays constant. - Tuning: size to physical cores for CPU work, and transfer large buffers instead of copying them.
The result is a self-healing, back-pressured pool that keeps every core busy under load.
よくある質問
「スループット向け再利用可能なワーカープールの構築」レッスンは無料ですか?
はい。「スループット向け再利用可能なワーカープールの構築」の完全なテキストはこのウェブで無料で読めます。インタラクティブに演習し(組み込みコードエディタと24時間対応のAIチューター)、Node.js Backend Development Bootcampコースの残りをアンロックするには、CoddyKit PROにアップグレードしてください。 Node.js Backend Development Bootcampコースには全4レッスンが含まれています。
「スループット向け再利用可能なワーカープールの構築」で何を学びますか?
負荷の高い状況でCPU使用率を最大化するため、スレッドを再利用するタスクキュー型ワーカープールを設計します。 ブラウザで直接実行するハンズオンコードでNode.js Backend Development Bootcampを演習し、24時間対応のAIチューターがレッスンを進める中での質問に答えます。
Node.js Backend Development Bootcampを始めるのに経験は必要ですか?
事前経験は必要ありません。CoddyKitのNode.js Backend Development Bootcampは初級者から上級者向けに構成されているため、ここから始めるか最初から始めて、自分のペースで進むことができます。 これはレッスン4/4です。
「スループット向け再利用可能なワーカープールの構築」レッスンにはどのくらい時間がかかりますか?
ほとんどのCoddyKitレッスンは約5~10分かかります。各レッスンはコンパクトでインタラクティブなので、着実に進歩し、ウェブとアプリ全体で正確に前回の場所から再開できます。
このNode.js Backend Development Bootcampレッスンでコードを書いて実行できますか?
はい。すべてのNode.js Backend Development Bootcampレッスンに組み込みコードエディタが含まれているため、ブラウザでリアルコードを書いて実行し、即座のAIフィードバックを取得できます。ローカル設定は不要です。
このコースのすべてのレッスン
- CPUバウンド処理でイベントループが停止する理由
- Worker Threadsの生成とメッセージの受け渡し
- SharedArrayBufferとAtomicsによるメモリ共有
- スループット向け再利用可能なワーカープールの構築