Loading…

Memory Management Guidelines

Production memory management patterns for Node.js/TypeScript — GC fundamentals, leak detection, weak references, caching strategies, object pooling, stream backpressure, and closure memory for V8

OpenCodeCodexClaude Code

bash

npx skills add m10rten/typescript-bits --skill memory-management-guidelines

skill.md · 482 lines~4k

Memory Management Guidelines

Practical, production-oriented memory management for Node.js/TypeScript on V8. Apply these when writing performance-sensitive code, investigating leaks, or reviewing resource-heavy logic.

GC Fundamentals

V8 Heap: Structure & Triggers

Generation	Size	Algorithm	Trigger
Young (nursery)	~16 MB	Scavenge (copying)	Young gen fills
Old	~1.4 GB	Mark-sweep / mark-compact	Old gen grows past limit
Large object	> 64 KB	Direct allocation	On demand

Young gen: Most objects die here. Fast ~1-5ms pauses.

Old gen: Mark-sweep identifies live; mark-compact defragments. Pauses can exceed 100ms.

Large objects: > 64 KB outside normal heap — compaction is expensive so V8 avoids it.

Other triggers: allocation pressure, idle-time GC (Node 14+), `global.gc()` (`--expose-gc`, never in production)

import { performance } from "node:perf_hooks";
const observer = new PerformanceObserver((list) => {
  for (const entry of list.getEntries()) console.log(`GC: ${entry.name}, ${entry.duration}ms`);
});
observer.observe({ entryTypes: ["gc"] });

GC Tuning Flags

Flag	Effect
`--max-old-space-size=4096`	Increase old gen limit (MB)
`--max-semi-space-size=64`	Increase young gen size (MB)
`--trace-gc`	Log all GC events to stderr
`--trace-gc-verbose`	Detailed GC stats

Memory Leak Patterns

Closures Holding References

Closures capture the entire scope chain — not just the variables they use:

// ❌ Retains entire `largeData` in closure scope
function createHandlers(largeData: big[]) {
  return { onClick: () => console.log(largeData.length) };
}
// ✅ Extract only what's needed
function createHandlers(largeData: big[]) {
  const length = largeData.length;
  return { onClick: () => console.log(length) };
}

Forgotten Timers and Intervals

Node.js won't exit while `setInterval` / `setTimeout` is active. The callback retains its entire scope. Always store the timer handle and clear it on cleanup.

EventEmitter Listeners

Every `.on()` is a strong reference from the emitter to the listener. Without removal, listeners accumulate and prevent GC of their scope:

// ❌ Accumulates — never removed
source.on("data", handleData);
// ✅ Remove on cleanup (pass signal or return teardown)
source.on("data", handleData);
signal.addEventListener("abort", () => source.off("data", handleData));
// ✅ Use once for one-shots
source.once("end", () => console.log("done"));

Set `maxListeners` to detect accidental leaks: `emitter.setMaxListeners(50)`.

Global Caches

Module-level caches that grow unbounded — every unique key lives forever:

// ❌ Unbounded
const cache = new Map<string, big[]>();
function fetchData(key: string): big[] {
  if (!cache.has(key)) cache.set(key, expensiveLoad(key));
  return cache.get(key)!;
}
// ✅ Bounded with LRU or TTL
const cache = new LRUCache<string, big[]>({ maxSize: 1000 });

Map/Set With Object Keys

Object keys in `Map` / `Set` are held by strong reference. Use `WeakMap` / `WeakSet` for ephemeral associations:

// ✅ GC collects the entry when the key object is no longer reachable
const sessions = new WeakMap<object, SessionData>();

Detecting Leaks

Heap Snapshots

Generate and compare snapshots over time:

import { writeHeapSnapshot } from "node:v8";
const snapshotPath = writeHeapSnapshot();

Chrome DevTools Memory Tab

1. Start Node with `--inspect`: `node --inspect app.mjs`

2. Open `chrome://inspect` in Chrome

3. Look for: detached DOM trees, shallow vs retained size, constructor retention

process.memoryUsage()

Quick health check — log at intervals to spot growth trends:

import { memoryUsage } from "node:process";
function logMemory(): void {
  const { rss, heapUsed, heapTotal, external } = memoryUsage();
  console.log({
    rss: `${(rss / 1024 / 1024).toFixed(1)} MB`,
    heapUsed: `${(heapUsed / 1024 / 1024).toFixed(1)} MB`,
    heapTotal: `${(heapTotal / 1024 / 1024).toFixed(1)} MB`,
    external: `${(external / 1024 / 1024).toFixed(1)} MB`,
  });
}

Metric	Measures	Leak indicator
`rss`	Total memory (heap + stack + code + C++)	Steady growth over time
`heapUsed`	V8 heap actively in use	Sawtooth (GC) then climb
`heapTotal`	V8 heap allocated	Not releasing after GC
`external`	C++ objects (Buffers, WASM)	Growth unrelated to JS heap

3-Snapshot Technique

1. Snapshot A at startup (baseline)

2. Snapshot B after suspected leak

3. Snapshot C after cleanup / repeat

4. Objects in B but not freed in C (absent from A) are likely leaked

Node --trace-gc

Run `node --trace-gc app.mjs`. Look for increasing old-gen size after major GC, longer scavenge pauses, or increasing GC frequency (allocation pressure).

Production Monitoring

Monitor heap growth in production with periodic `process.memoryUsage()` checks:

setInterval(() => {
  const { heapUsed } = process.memoryUsage();
  if (heapUsed / 1024 / 1024 > 1024) console.error(`heap > 1 GB`);
}, 30_000);

Thresholds: >512 MB log/metric, >1 GB alert/snapshot, >2 GB restart/scale.

Weak References

WeakMap / WeakSet

Keys are held weakly — if no other reference to the key exists, the entry is GC'd:

const metadata = new WeakMap<object, Metadata>();
function setMeta(obj: object, meta: Metadata): void {
  metadata.set(obj, meta);
}
function getMeta(obj: object): Metadata | undefined {
  return metadata.get(obj);
}

Structure	Key type	Value GC'd with key?	Use case
`Map`	Any	No (must delete)	Long-lived associations
`WeakMap`	Object only	Yes	Ephemeral metadata, membrane
`Set`	Any	No (must delete)	Membership tracking
`WeakSet`	Object only	Yes	Tagging, visited tracking

WeakRef

Holds a weak reference — does not prevent GC. Dereference with `.deref()`:

const cache = new Map<string, WeakRef<ExpensiveObject>>();
function getOrCreate(key: string): ExpensiveObject {
  const ref = cache.get(key);
  const existing = ref?.deref();
  if (existing) return existing;
  const obj = new ExpensiveObject(key);
  cache.set(key, new WeakRef(obj));
  return obj;
}

Caveats: `.deref()` can return `undefined` after GC — always check. GC is non-deterministic. Avoid in hot paths.

FinalizationRegistry

const registry = new FinalizationRegistry<string>((resourceId: string) => {
  nativeCleanup(resourceId);
});
class FileHandle {
  readonly #id: string;
  constructor(path: string) {
    this.#id = nativeOpen(path);
    registry.register(this, this.#id);
  }
}

Caveats: Callback runs on a different tick — don't rely on timing. Not called during process exit — explicit cleanup still needed.

Buffer Management

Buffer.alloc vs Buffer.from

Method	Behavior	When to use
`Buffer.alloc(size)`	Zero-filled, safe	New allocations
`Buffer.allocUnsafe(size)`	Uninitialized, faster	Performance-critical, immediately overwrite
`Buffer.from(data)`	Allocates and copies	Converting strings, arrays

const safe = Buffer.alloc(1024);
const fast = Buffer.allocUnsafe(1024);
fast.fill(0); // must fill before reading

Avoid Large String Concatenation

Creates multiple intermediate copies. For large payloads, use arrays or buffers:

// ❌ O(n²) memory — multiple intermediate strings
let result = "";
for (const chunk of largeArray) result += chunk;

// ✅ Single allocation
const parts: string[] = [];
for (const chunk of largeArray) parts.push(chunk);
const result = parts.join("");

// ✅ For binary data
const result = Buffer.concat(largeArray.map((s) => Buffer.from(s)));

Detached Buffer Gotcha

`postMessage()` transfers `ArrayBuffer` ownership — source `byteLength` becomes 0:

const buf = new ArrayBuffer(1024);
postMessage(buf, [buf]);
buf.byteLength; // 0 — detached!

Caching Strategies

LRU Cache

Evicts least-recently-used entries when full:

class LRUCache<K, V> {
  readonly #max: number;
  readonly #map = new Map<K, V>();
  constructor(max: number) {
    this.#max = max;
  }
  get(key: K): V | undefined {
    const value = this.#map.get(key);
    if (value !== undefined) {
      this.#map.delete(key);
      this.#map.set(key, value);
    }
    return value;
  }
  set(key: K, value: V): void {
    if (this.#map.has(key)) this.#map.delete(key);
    this.#map.set(key, value);
    if (this.#map.size > this.#max) {
      const first = this.#map.keys().next();
      if (!first.done) this.#map.delete(first.value);
    }
  }
}

TTL Cache

Entries expire after a set duration regardless of access:

class TTLCache<K, V> {
  readonly #map = new Map<K, { value: V; expiresAt: number }>();
  constructor(private readonly ttlMs: number) {}
  get(key: K): V | undefined {
    const entry = this.#map.get(key);
    if (!entry) return;
    if (Date.now() > entry.expiresAt) {
      this.#map.delete(key);
      return;
    }
    return entry.value;
  }
  set(key: K, value: V): void {
    this.#map.set(key, { value, expiresAt: Date.now() + this.ttlMs });
  }
}

Eviction Comparison

Strategy	Eviction policy	Memory bound	Overhead	Use case
LRU	Least recently used	Hard	O(1)	General caching
TTL	Time-based	Soft	O(1)	Stale-data tolerance
LFU	Least frequently used	Hard	O(log n)	Hot data prioritization
FIFO	First in, first out	Hard	O(1)	Simple, predictable

Which Cache to Use

Pattern	Best pick
Fixed-size hot data	FIFO
Recency-skewed access	LRU
Frequency-skewed	LFU
Time-bounded validity	TTL

Object Pooling

Reuse objects to reduce GC pressure — allocation is cheap, but GC of short-lived objects is not free:

class ConnectionPool<T> {
  readonly #available: T[] = [];
  readonly #max: number;
  readonly #factory: () => T;
  readonly #reset: (item: T) => void;

  constructor(max: number, factory: () => T, reset: (item: T) => void) {
    this.#max = max;
    this.#factory = factory;
    this.#reset = reset;
  }
  acquire(): T {
    return this.#available.pop() ?? this.#factory();
  }
  release(item: T): void {
    this.#reset(item);
    if (this.#available.length < this.#max) this.#available.push(item);
  }
}

When to Pool

Pool candidate	NOT worth pooling
Database connections	Small temporary objects
Large buffers (> 64 KB)	Short-lived primitives
WebSocket / TCP sockets	Objects pooled by V8 already
Expensive-to-create objects	Objects with complex reset logic

Pool Gotchas

Stale state — always reset before reuse

Held references — pooled objects in closures prevent pool GC

Growth under load — set hard `max`, not a soft target

Fragmentation — varying-size objects waste space; use fixed-size pools

Streams and Backpressure

Memory-Safe Streaming

Read streams buffer data. Without backpressure, entire payloads fit in memory:

// ❌ Loads entire file
const data = await readFile("large.log", "utf-8");
// ✅ Constant memory — one line at a time
const rl = createInterface({ input: createReadStream("large.log") });
for await (const line of rl) processLine(line);

highWaterMark

Controls internal buffer size. Tune for throughput vs memory:

// Lower memory, more I/O
const slow = createReadStream("file", { highWaterMark: 16 * 1024 }); // 16 KB
// Higher throughput, more memory
const fast = createReadStream("file", { highWaterMark: 1024 * 1024 }); // 1 MB

highWaterMark	Memory/stream	Throughput	Use case
16 KB	~16 KB	Lower	Memory-constrained
64 KB	~64 KB	Medium	Default (balanced)
1 MB	~1 MB	Higher	High-throughput

Drain Events

When `Writable.write()` returns `false`, wait for `drain` before continuing:

async function writeLarge(output: Writable, items: Buffer[]): Promise<void> {
  for (const item of items) {
    const canContinue = output.write(item);
    if (!canContinue) await new Promise<void>((resolve) => output.once("drain", resolve));
  }
  output.end();
}

Closure Memory

Scope Capture

A closure captures the entire `[[Scope]]` — every variable in every enclosing function, not just referenced ones:

// ❌ Closure retains both `massive` and `tiny` — only `tiny` needed
function outer() {
  const massive = new Array(1_000_000).fill("data");
  const tiny = "hello";
  return function inner() {
    return tiny;
  };
}

// ✅ Separate scopes so `massive` isn't captured
function makeGetter() {
  const tiny = "hello";
  return function inner() {
    return tiny;
  };
}
function createMassiveData(): void {
  const massive = new Array(1_000_000).fill("data"); // GC'd after return
}

Holding `this` in Callbacks

Arrow functions capture `this` — if `this` is a large object, it's retained for the callback's lifetime. Extract only what's needed instead.

Accidental Retention in Loops

Callbacks in loops capture the entire iteration scope. If only a primitive is needed, extract it.

Common Mistakes

Mistake	Fix
Unbounded `Map`/`Set` with no eviction	Add LRU, TTL, or bounded size
Not removing EventEmitter listeners	Track with cleanup; use `once` for one-shots
Closures capturing large scope	Extract only needed variables
Using `WeakRef` without `.deref()` null check	Always guard: `ref?.deref() ?? createNew()`
Large string concatenation in loops	Use `Array.join()` or `Buffer.concat()`
No `highWaterMark` tuning on streams	Set based on throughput vs memory trade-off
`Buffer.allocUnsafe` without filling before read	`fill(0)` or immediately overwrite
Ignoring `external` memory growth	Profile native bindings, Buffer usage, WASM
No heap snapshot comparison in investigations	Use 3-snapshot technique

Loading…

Memory Management Guidelines

Production memory management patterns for Node.js/TypeScript — GC fundamentals, leak detection, weak references, caching strategies, object pooling, stream backpressure, and closure memory for V8

OpenCodeCodexClaude Code

bash

npx skills add m10rten/typescript-bits --skill memory-management-guidelines

skill.md · 482 lines~4k

Memory Management Guidelines

Practical, production-oriented memory management for Node.js/TypeScript on V8. Apply these when writing performance-sensitive code, investigating leaks, or reviewing resource-heavy logic.

GC Fundamentals

V8 Heap: Structure & Triggers

Generation	Size	Algorithm	Trigger
Young (nursery)	~16 MB	Scavenge (copying)	Young gen fills
Old	~1.4 GB	Mark-sweep / mark-compact	Old gen grows past limit
Large object	> 64 KB	Direct allocation	On demand

Young gen: Most objects die here. Fast ~1-5ms pauses.

Old gen: Mark-sweep identifies live; mark-compact defragments. Pauses can exceed 100ms.

Large objects: > 64 KB outside normal heap — compaction is expensive so V8 avoids it.

Other triggers: allocation pressure, idle-time GC (Node 14+), `global.gc()` (`--expose-gc`, never in production)

import { performance } from "node:perf_hooks";
const observer = new PerformanceObserver((list) => {
  for (const entry of list.getEntries()) console.log(`GC: ${entry.name}, ${entry.duration}ms`);
});
observer.observe({ entryTypes: ["gc"] });

GC Tuning Flags

Flag	Effect
`--max-old-space-size=4096`	Increase old gen limit (MB)
`--max-semi-space-size=64`	Increase young gen size (MB)
`--trace-gc`	Log all GC events to stderr
`--trace-gc-verbose`	Detailed GC stats

Memory Leak Patterns

Closures Holding References

Closures capture the entire scope chain — not just the variables they use:

// ❌ Retains entire `largeData` in closure scope
function createHandlers(largeData: big[]) {
  return { onClick: () => console.log(largeData.length) };
}
// ✅ Extract only what's needed
function createHandlers(largeData: big[]) {
  const length = largeData.length;
  return { onClick: () => console.log(length) };
}

Forgotten Timers and Intervals

Node.js won't exit while `setInterval` / `setTimeout` is active. The callback retains its entire scope. Always store the timer handle and clear it on cleanup.

EventEmitter Listeners

Every `.on()` is a strong reference from the emitter to the listener. Without removal, listeners accumulate and prevent GC of their scope:

// ❌ Accumulates — never removed
source.on("data", handleData);
// ✅ Remove on cleanup (pass signal or return teardown)
source.on("data", handleData);
signal.addEventListener("abort", () => source.off("data", handleData));
// ✅ Use once for one-shots
source.once("end", () => console.log("done"));

Set `maxListeners` to detect accidental leaks: `emitter.setMaxListeners(50)`.

Global Caches

Module-level caches that grow unbounded — every unique key lives forever:

// ❌ Unbounded
const cache = new Map<string, big[]>();
function fetchData(key: string): big[] {
  if (!cache.has(key)) cache.set(key, expensiveLoad(key));
  return cache.get(key)!;
}
// ✅ Bounded with LRU or TTL
const cache = new LRUCache<string, big[]>({ maxSize: 1000 });

Map/Set With Object Keys

Object keys in `Map` / `Set` are held by strong reference. Use `WeakMap` / `WeakSet` for ephemeral associations:

// ✅ GC collects the entry when the key object is no longer reachable
const sessions = new WeakMap<object, SessionData>();

Detecting Leaks

Heap Snapshots

Generate and compare snapshots over time:

import { writeHeapSnapshot } from "node:v8";
const snapshotPath = writeHeapSnapshot();

Chrome DevTools Memory Tab

1. Start Node with `--inspect`: `node --inspect app.mjs`

2. Open `chrome://inspect` in Chrome

3. Look for: detached DOM trees, shallow vs retained size, constructor retention

process.memoryUsage()

Quick health check — log at intervals to spot growth trends:

import { memoryUsage } from "node:process";
function logMemory(): void {
  const { rss, heapUsed, heapTotal, external } = memoryUsage();
  console.log({
    rss: `${(rss / 1024 / 1024).toFixed(1)} MB`,
    heapUsed: `${(heapUsed / 1024 / 1024).toFixed(1)} MB`,
    heapTotal: `${(heapTotal / 1024 / 1024).toFixed(1)} MB`,
    external: `${(external / 1024 / 1024).toFixed(1)} MB`,
  });
}

Metric	Measures	Leak indicator
`rss`	Total memory (heap + stack + code + C++)	Steady growth over time
`heapUsed`	V8 heap actively in use	Sawtooth (GC) then climb
`heapTotal`	V8 heap allocated	Not releasing after GC
`external`	C++ objects (Buffers, WASM)	Growth unrelated to JS heap

3-Snapshot Technique

1. Snapshot A at startup (baseline)

2. Snapshot B after suspected leak

3. Snapshot C after cleanup / repeat

4. Objects in B but not freed in C (absent from A) are likely leaked

Node --trace-gc

Run `node --trace-gc app.mjs`. Look for increasing old-gen size after major GC, longer scavenge pauses, or increasing GC frequency (allocation pressure).

Production Monitoring

Monitor heap growth in production with periodic `process.memoryUsage()` checks:

setInterval(() => {
  const { heapUsed } = process.memoryUsage();
  if (heapUsed / 1024 / 1024 > 1024) console.error(`heap > 1 GB`);
}, 30_000);

Thresholds: >512 MB log/metric, >1 GB alert/snapshot, >2 GB restart/scale.

Weak References

WeakMap / WeakSet

Keys are held weakly — if no other reference to the key exists, the entry is GC'd:

const metadata = new WeakMap<object, Metadata>();
function setMeta(obj: object, meta: Metadata): void {
  metadata.set(obj, meta);
}
function getMeta(obj: object): Metadata | undefined {
  return metadata.get(obj);
}

Structure	Key type	Value GC'd with key?	Use case
`Map`	Any	No (must delete)	Long-lived associations
`WeakMap`	Object only	Yes	Ephemeral metadata, membrane
`Set`	Any	No (must delete)	Membership tracking
`WeakSet`	Object only	Yes	Tagging, visited tracking

WeakRef

Holds a weak reference — does not prevent GC. Dereference with `.deref()`:

const cache = new Map<string, WeakRef<ExpensiveObject>>();
function getOrCreate(key: string): ExpensiveObject {
  const ref = cache.get(key);
  const existing = ref?.deref();
  if (existing) return existing;
  const obj = new ExpensiveObject(key);
  cache.set(key, new WeakRef(obj));
  return obj;
}

Caveats: `.deref()` can return `undefined` after GC — always check. GC is non-deterministic. Avoid in hot paths.

FinalizationRegistry

const registry = new FinalizationRegistry<string>((resourceId: string) => {
  nativeCleanup(resourceId);
});
class FileHandle {
  readonly #id: string;
  constructor(path: string) {
    this.#id = nativeOpen(path);
    registry.register(this, this.#id);
  }
}

Caveats: Callback runs on a different tick — don't rely on timing. Not called during process exit — explicit cleanup still needed.

Buffer Management

Buffer.alloc vs Buffer.from

Method	Behavior	When to use
`Buffer.alloc(size)`	Zero-filled, safe	New allocations
`Buffer.allocUnsafe(size)`	Uninitialized, faster	Performance-critical, immediately overwrite
`Buffer.from(data)`	Allocates and copies	Converting strings, arrays

const safe = Buffer.alloc(1024);
const fast = Buffer.allocUnsafe(1024);
fast.fill(0); // must fill before reading

Avoid Large String Concatenation

Creates multiple intermediate copies. For large payloads, use arrays or buffers:

// ❌ O(n²) memory — multiple intermediate strings
let result = "";
for (const chunk of largeArray) result += chunk;

// ✅ Single allocation
const parts: string[] = [];
for (const chunk of largeArray) parts.push(chunk);
const result = parts.join("");

// ✅ For binary data
const result = Buffer.concat(largeArray.map((s) => Buffer.from(s)));

Detached Buffer Gotcha

`postMessage()` transfers `ArrayBuffer` ownership — source `byteLength` becomes 0:

const buf = new ArrayBuffer(1024);
postMessage(buf, [buf]);
buf.byteLength; // 0 — detached!

Caching Strategies

LRU Cache

Evicts least-recently-used entries when full:

class LRUCache<K, V> {
  readonly #max: number;
  readonly #map = new Map<K, V>();
  constructor(max: number) {
    this.#max = max;
  }
  get(key: K): V | undefined {
    const value = this.#map.get(key);
    if (value !== undefined) {
      this.#map.delete(key);
      this.#map.set(key, value);
    }
    return value;
  }
  set(key: K, value: V): void {
    if (this.#map.has(key)) this.#map.delete(key);
    this.#map.set(key, value);
    if (this.#map.size > this.#max) {
      const first = this.#map.keys().next();
      if (!first.done) this.#map.delete(first.value);
    }
  }
}

TTL Cache

Entries expire after a set duration regardless of access:

class TTLCache<K, V> {
  readonly #map = new Map<K, { value: V; expiresAt: number }>();
  constructor(private readonly ttlMs: number) {}
  get(key: K): V | undefined {
    const entry = this.#map.get(key);
    if (!entry) return;
    if (Date.now() > entry.expiresAt) {
      this.#map.delete(key);
      return;
    }
    return entry.value;
  }
  set(key: K, value: V): void {
    this.#map.set(key, { value, expiresAt: Date.now() + this.ttlMs });
  }
}

Eviction Comparison

Strategy	Eviction policy	Memory bound	Overhead	Use case
LRU	Least recently used	Hard	O(1)	General caching
TTL	Time-based	Soft	O(1)	Stale-data tolerance
LFU	Least frequently used	Hard	O(log n)	Hot data prioritization
FIFO	First in, first out	Hard	O(1)	Simple, predictable

Which Cache to Use

Pattern	Best pick
Fixed-size hot data	FIFO
Recency-skewed access	LRU
Frequency-skewed	LFU
Time-bounded validity	TTL

Object Pooling

Reuse objects to reduce GC pressure — allocation is cheap, but GC of short-lived objects is not free:

class ConnectionPool<T> {
  readonly #available: T[] = [];
  readonly #max: number;
  readonly #factory: () => T;
  readonly #reset: (item: T) => void;

  constructor(max: number, factory: () => T, reset: (item: T) => void) {
    this.#max = max;
    this.#factory = factory;
    this.#reset = reset;
  }
  acquire(): T {
    return this.#available.pop() ?? this.#factory();
  }
  release(item: T): void {
    this.#reset(item);
    if (this.#available.length < this.#max) this.#available.push(item);
  }
}

When to Pool

Pool candidate	NOT worth pooling
Database connections	Small temporary objects
Large buffers (> 64 KB)	Short-lived primitives
WebSocket / TCP sockets	Objects pooled by V8 already
Expensive-to-create objects	Objects with complex reset logic

Pool Gotchas

Stale state — always reset before reuse

Held references — pooled objects in closures prevent pool GC

Growth under load — set hard `max`, not a soft target

Fragmentation — varying-size objects waste space; use fixed-size pools

Streams and Backpressure

Memory-Safe Streaming

Read streams buffer data. Without backpressure, entire payloads fit in memory:

// ❌ Loads entire file
const data = await readFile("large.log", "utf-8");
// ✅ Constant memory — one line at a time
const rl = createInterface({ input: createReadStream("large.log") });
for await (const line of rl) processLine(line);

highWaterMark

Controls internal buffer size. Tune for throughput vs memory:

// Lower memory, more I/O
const slow = createReadStream("file", { highWaterMark: 16 * 1024 }); // 16 KB
// Higher throughput, more memory
const fast = createReadStream("file", { highWaterMark: 1024 * 1024 }); // 1 MB

highWaterMark	Memory/stream	Throughput	Use case
16 KB	~16 KB	Lower	Memory-constrained
64 KB	~64 KB	Medium	Default (balanced)
1 MB	~1 MB	Higher	High-throughput

Drain Events

When `Writable.write()` returns `false`, wait for `drain` before continuing:

async function writeLarge(output: Writable, items: Buffer[]): Promise<void> {
  for (const item of items) {
    const canContinue = output.write(item);
    if (!canContinue) await new Promise<void>((resolve) => output.once("drain", resolve));
  }
  output.end();
}

Closure Memory

Scope Capture

A closure captures the entire `[[Scope]]` — every variable in every enclosing function, not just referenced ones:

// ❌ Closure retains both `massive` and `tiny` — only `tiny` needed
function outer() {
  const massive = new Array(1_000_000).fill("data");
  const tiny = "hello";
  return function inner() {
    return tiny;
  };
}

// ✅ Separate scopes so `massive` isn't captured
function makeGetter() {
  const tiny = "hello";
  return function inner() {
    return tiny;
  };
}
function createMassiveData(): void {
  const massive = new Array(1_000_000).fill("data"); // GC'd after return
}

Holding `this` in Callbacks

Arrow functions capture `this` — if `this` is a large object, it's retained for the callback's lifetime. Extract only what's needed instead.

Accidental Retention in Loops

Callbacks in loops capture the entire iteration scope. If only a primitive is needed, extract it.

Common Mistakes

Mistake	Fix
Unbounded `Map`/`Set` with no eviction	Add LRU, TTL, or bounded size
Not removing EventEmitter listeners	Track with cleanup; use `once` for one-shots
Closures capturing large scope	Extract only needed variables
Using `WeakRef` without `.deref()` null check	Always guard: `ref?.deref() ?? createNew()`
Large string concatenation in loops	Use `Array.join()` or `Buffer.concat()`
No `highWaterMark` tuning on streams	Set based on throughput vs memory trade-off
`Buffer.allocUnsafe` without filling before read	`fill(0)` or immediately overwrite
Ignoring `external` memory growth	Profile native bindings, Buffer usage, WASM
No heap snapshot comparison in investigations	Use 3-snapshot technique