Technical Documentation

March 29, 2026 · View on GitHub

Overview

The tiny-lru library provides a high-performance, memory-efficient Least Recently Used (LRU) cache implementation with optional Time-To-Live (TTL) support. This document covers the technical architecture, data flow, modern usage patterns, and security considerations for production applications.

Table of Contents

Architecture Overview

The LRU cache implements a doubly-linked list combined with a hash map for O(1) operations on both insertion and retrieval.

Accessibility Note: All diagrams use WCAG AA compliant colors with sufficient contrast ratios (>4.5:1) that work on both light and dark backgrounds. Colors are combined with distinct shapes and borders to ensure accessibility for users with color vision deficiencies.

graph TD
    A["LRU Cache Instance"] --> B["Hash Map<br/>items: Object"]
    A --> C["Doubly Linked List"]

    B --> D["Key-Value Mapping<br/>O(1) Access"]
    C --> E["Node: first"]
    C --> F["Node: last"]

    E --> G["prev: null<br/>next: Node<br/>key: string<br/>value: any<br/>expiry: number"]
    F --> H["prev: Node<br/>next: null<br/>key: string<br/>value: any<br/>expiry: number"]

    G --> I["Middle Nodes"]
    I --> F

    style A fill:#2563eb,stroke:#1e40af,stroke-width:2px,color:#ffffff
    style B fill:#7c3aed,stroke:#5b21b6,stroke-width:2px,color:#ffffff
    style C fill:#059669,stroke:#047857,stroke-width:2px,color:#ffffff

Core Components

  • Hash Map (items): Object.create(null) providing O(1) key-based access to cache nodes
  • Doubly Linked List: Maintains LRU order with first and last pointers for O(1) insertion/deletion
  • Cache Nodes: Store key, value, expiry timestamp, and linked list pointers (prev/next)
  • LRU Class Properties:
    • first: Pointer to least recently used item
    • last: Pointer to most recently used item
    • items: Hash map for O(1) key lookup
    • max: Maximum cache size (0 = unlimited)
    • size: Current number of items
    • ttl: Time-to-live in milliseconds (0 = no expiration)
    • resetTTL: Whether to reset TTL on set() and setWithEvicted() operations (not on get())

Data Flow

Cache Hit Scenario

sequenceDiagram
    participant Client
    participant LRU as LRU Cache
    participant HashMap as Hash Map
    participant LinkedList as Linked List

    Client->>LRU: get(key)
    LRU->>HashMap: items[key]
    HashMap-->>LRU: node reference

    alt TTL enabled
        LRU->>LRU: check item.expiry <= Date.now()
        alt expired
            LRU->>HashMap: delete items[key]
            LRU->>LinkedList: remove node
            LRU-->>Client: undefined
        else not expired
            LRU->>LRU: moveToEnd(item) - O(1) optimization
            LRU-->>Client: item.value
        end
    else no TTL
        LRU->>LRU: moveToEnd(item) - O(1) optimization
        LRU-->>Client: item.value
    end

Cache Miss and Eviction Scenario

sequenceDiagram
    participant Client
    participant LRU as LRU Cache
    participant HashMap as Hash Map
    participant LinkedList as Linked List

    Client->>LRU: set(key, value)
    LRU->>HashMap: check items[key]
    HashMap-->>LRU: undefined (miss)

    alt cache full (size === max)
        LRU->>LRU: evict() - remove this.first
        LRU->>HashMap: delete items[first.key]
        LRU->>LinkedList: update first pointer
        Note over LRU: LRU eviction complete
    end

    LRU->>HashMap: items[key] = new item
    LRU->>LinkedList: set item.prev = this.last
    LRU->>LinkedList: update this.last = item
    LRU->>LRU: increment size
    LRU-->>Client: this (chainable)

Core Operations

Time Complexity

OperationAverage CaseWorst CaseSpaceDescription
get(key)O(1)O(1)O(1)Retrieve value and move to end
set(key, value)O(1)O(1)O(1)Store value, evict if needed
setWithEvicted(key, value)O(1)O(1)O(1)Store value, return evicted item
delete(key)O(1)O(1)O(1)Remove item from cache
has(key)O(1)O(1)O(1)Check key existence
peek(key)O(1)O(1)O(1)Retrieve value without LRU update
clear()O(n)O(n)O(1)Reset all pointers, nullify node links
evict()O(1)O(1)O(1)Remove least recently used item
expiresAt(key)O(1)O(1)O(1)Get expiration timestamp
moveToEnd(item)O(1)O(1)O(1)Internal: optimize LRU positioning
keys()O(n)O(n)O(n)Array of all keys in LRU order
values(keys?)O(n)O(n)O(n)Array of values for specified keys
entries([keys])O(n)O(n)O(n)Array of [key, value] pairs

Memory Usage

  • Per Node: ~120 bytes (key + value + pointers + metadata)
  • Base Overhead: ~200 bytes (class instance + hash map)
  • Total: base + (nodes × 120) bytes approximately

Mathematical Foundation

Core Operations

The LRU cache maintains a doubly-linked list LL and a hash table HH for O(1) operations:

Data Structure:

  • L=(first,last,size)L = (first, last, size) - Doubly-linked list with head/tail pointers
  • H:K→{key:K,value:V,prev:Object,next:Object,expiry:N0}H: K \rightarrow \{key: K, value: V, prev: Object, next: Object, expiry: \mathbb{N}_0\} - Hash table mapping keys to item objects
  • max∈N0max \in \mathbb{N}_0 - Maximum cache size (0 = unlimited)
  • ttl∈N0ttl \in \mathbb{N}_0 - Time-to-live in milliseconds
  • resetTtl∈{true,false}resetTtl \in \{\text{true}, \text{false}\} - Whether to reset TTL on set() operations

Core Methods:

Note: The mathematical notation uses create(k, v) to represent the item creation logic that is inline in the actual implementation.

Set Operation: set(k,v)→LRUset(k, v) \rightarrow \text{LRU}

set(k,v)={update(k,v)if k∈Hinsert(k,v)if k∉Hupdate(k,v)=H[k].value←v∧moveToEnd(H[k])∧{H[k].expiry←tnow+ttlif resetTtl=true∧ttl>0no-opotherwiseinsert(k,v)={evict()∧create(k,v)if max>0∧size=maxcreate(k,v)otherwisecreate(k,v)=H[k]←{key:k,value:v,prev:last,next:null,expiry:ttl>0?tnow+ttl:0}∧last←H[k]∧size←size+1∧{first←H[k]if size=1last.next←H[k]otherwise\begin{align} set(k, v) &= \begin{cases} update(k, v) & \text{if } k \in H \\ insert(k, v) & \text{if } k \notin H \end{cases} \\ update(k, v) &= H[k].value \leftarrow v \land moveToEnd(H[k]) \\ & \quad \land \begin{cases} H[k].expiry \leftarrow t_{now} + ttl & \text{if } resetTtl = true \land ttl > 0 \\ \text{no-op} & \text{otherwise} \end{cases} \\ insert(k, v) &= \begin{cases} evict() \land create(k, v) & \text{if } max > 0 \land size = max \\ create(k, v) & \text{otherwise} \end{cases} \\ create(k, v) &= H[k] \leftarrow \{key: k, value: v, prev: last, next: null, expiry: ttl > 0 ? t_{now} + ttl : 0\} \\ & \quad \land last \leftarrow H[k] \land size \leftarrow size + 1 \\ & \quad \land \begin{cases} first \leftarrow H[k] & \text{if } size = 1 \\ last.next \leftarrow H[k] & \text{otherwise} \end{cases} \end{align}

Time Complexity: O(1)O(1) amortized

Set With Evicted Operation: setWithEvicted(k,v)→{key:K,value:V,expiry:N0}∪{⊥}setWithEvicted(k, v) \rightarrow \{key: K, value: V, expiry: \mathbb{N}_0\} \cup \{\bot\}

setWithEvicted(k,v)={update(k,v)∧⊥if k∈Hevicted∧create(k,v)if k∉H∧max>0∧size=max⊥∧create(k,v)if k∉H∧(max=0∨size<max)update(k,v)=H[k].value←v∧moveToEnd(H[k])∧{H[k].expiry←tnow+ttlif resetTtl=true∧ttl>0no-opotherwisewhere evicted={{key:this.first.key,value:this.first.value,expiry:this.first.expiry}if size>0⊥otherwise\begin{align} setWithEvicted(k, v) &= \begin{cases} update(k, v) \land \bot & \text{if } k \in H \\ evicted \land create(k, v) & \text{if } k \notin H \land max > 0 \land size = max \\ \bot \land create(k, v) & \text{if } k \notin H \land (max = 0 \lor size < max) \end{cases} \\ update(k, v) &= H[k].value \leftarrow v \land moveToEnd(H[k]) \\ & \quad \land \begin{cases} H[k].expiry \leftarrow t_{now} + ttl & \text{if } resetTtl = true \land ttl > 0 \\ \text{no-op} & \text{otherwise} \end{cases} \\ \text{where } evicted &= \begin{cases} \{key: this.first.key, value: this.first.value, expiry: this.first.expiry\} & \text{if } size > 0 \\ \bot & \text{otherwise} \end{cases} \end{align}

Time Complexity: O(1)O(1) amortized

Get Operation: get(k)→V∪{⊥}get(k) \rightarrow V \cup \{\bot\}

get(k)={cleanup(k)∧moveToEnd(H[k])∧H[k].valueif k∈H∧(ttl=0∨H[k].expiry>tnow)⊥otherwise\begin{align} get(k) &= \begin{cases} cleanup(k) \land moveToEnd(H[k]) \land H[k].value & \text{if } k \in H \land (ttl = 0 \lor H[k].expiry > t_{now}) \\ \bot & \text{otherwise} \end{cases} \end{align}

Note: get() operations never reset TTL, regardless of the resetTtl setting.

Time Complexity: O(1)O(1)

Delete Operation: delete(k)→LRUdelete(k) \rightarrow \text{LRU}

delete(k)={removeFromList(H[k])∧H∖{k}∧size←size−1if k∈Hno-opotherwiseremoveFromList(item)={item.prev.next←item.next∧item.next.prev←item.prev∧first←item.next∧last←item.previf item.prev≠null∧item.next≠nullitem.prev.next←item.next∧first←item.next∧last←nullif item.prev≠null∧item.next=nullitem.next.prev←item.prev∧first←item.next∧last←nullif item.prev=null∧item.next≠nullfirst←null∧last←nullif item.prev=null∧item.next=null\begin{align} delete(k) &= \begin{cases} removeFromList(H[k]) \land H \setminus \{k\} \land size \leftarrow size - 1 & \text{if } k \in H \\ \text{no-op} & \text{otherwise} \end{cases} \\ removeFromList(item) &= \begin{cases} item.prev.next \leftarrow item.next \land item.next.prev \leftarrow item.prev \land first \leftarrow item.next \land last \leftarrow item.prev & \text{if } item.prev \neq null \land item.next \neq null \\ item.prev.next \leftarrow item.next \land first \leftarrow item.next \land last \leftarrow null & \text{if } item.prev \neq null \land item.next = null \\ item.next.prev \leftarrow item.prev \land first \leftarrow item.next \land last \leftarrow null & \text{if } item.prev = null \land item.next \neq null \\ first \leftarrow null \land last \leftarrow null & \text{if } item.prev = null \land item.next = null \end{cases} \end{align}

Time Complexity: O(1)O(1)

Move to End: moveToEnd(item)moveToEnd(item)

moveToEnd(item)={no-opif item=lastitem.prev.next←item.next∧item.next.prev←item.prev∧first←item.next∧item.prev←last∧last.next←item∧last←itemif item≠last\begin{align} moveToEnd(item) &= \begin{cases} \text{no-op} & \text{if } item = last \\ item.prev.next \leftarrow item.next \land item.next.prev \leftarrow item.prev \land first \leftarrow item.next \land item.prev \leftarrow last \land last.next \leftarrow item \land last \leftarrow item & \text{if } item \neq last \end{cases} \end{align}

Edge Case: When item is the only node in the list (item.prev=null∧item.next=nullitem.prev = null \land item.next = null), the condition item=lastitem = last is true since first=last=itemfirst = last = item, so the operation is a no-op.

Time Complexity: O(1)O(1)

Eviction Policy

LRU Eviction: When max>0∧size=maxmax > 0 \land size = max and inserting a new item:

evict()={no-opif size=0first←first.next∧first.prev←null∧H∖{first.key}∧size←size−1∧item.next←nullif size>0evict() = \begin{cases} \text{no-op} & \text{if } size = 0 \\ first \leftarrow first.next \land first.prev \leftarrow null \land H \setminus \{first.key\} \land size \leftarrow size - 1 \land item.next \leftarrow null & \text{if } size > 0 \end{cases}

TTL Expiration

Expiration Check: For any operation accessing key kk:

isExpired(k)=ttl>0∧H[k].expiry≤tnowisExpired(k) = ttl > 0 \land H[k].expiry \leq t_{now}

Automatic Cleanup: Expired items are removed on access:

cleanup(k)={delete(k)if isExpired(k)no-opotherwisecleanup(k) = \begin{cases} delete(k) & \text{if } isExpired(k) \\ \text{no-op} & \text{otherwise} \end{cases}

TTL Reset Behavior:

  • TTL is reset during set() and setWithEvicted() operations when resetTtl = true
  • get() operations never reset TTL, regardless of the resetTtl setting

Space Complexity

  • Worst Case: O(n)O(n) where n=min⁡(size,max)n = \min(size, max)
  • Hash Table: O(n)O(n) for key-value storage
  • Linked List: O(n)O(n) for LRU ordering
  • Per Item Overhead: Constant space for prev/next pointers and metadata

Invariants

  1. Size Constraint: $0 \leq size \leq max(when(whenmax > 0$)
  2. List Consistency: first≠null  ⟺  last≠null  ⟺  size>0first \neq null \iff last \neq null \iff size > 0
  3. Hash Consistency: ∣H∣=size|H| = size
  4. LRU Order: Items in list are ordered from least to most recently used
  5. TTL Validity: (ttl=0⇒∀k∈H:H[k].expiry=0)∧(ttl>0⇒∀k∈H:H[k].expiry≥tnow)(ttl = 0 \Rightarrow \forall k \in H: H[k].expiry = 0) \land (ttl > 0 \Rightarrow \forall k \in H: H[k].expiry \geq t_{now})
  6. TTL Reset Invariant: TTL is only reset during set() and setWithEvicted() operations when resetTtl = true

TypeScript Support

The library includes comprehensive TypeScript definitions with generic type support for type-safe value storage.

Interface Definitions

// Factory function with optional generic type
export function lru<T = any>(max?: number, ttl?: number, resetTtl?: boolean): LRU<T>;

// Main LRU class with generic value type
export class LRU<T> {
	constructor(max?: number, ttl?: number, resetTtl?: boolean);

	// Instance properties
	first: LRUItem<T> | null;
	items: Record<any, LRUItem<T>>;
	last: LRUItem<T> | null;
	max: number;
	resetTtl: boolean;
	size: number;
	ttl: number;

	// Core methods
	clear(): this;
	delete(key: any): this;
	entries(keys?: any[]): [any, T | undefined][];
	evict(): this;
	expiresAt(key: any): number | undefined;
	get(key: any): T | undefined;
	has(key: any): boolean;
	keys(): any[];
	peek(key: any): T | undefined;
	set(key: any, value: T): this;
	setWithEvicted(key: any, value: T): EvictedItem<T> | null;
	values(keys?: any[]): (T | undefined)[];
}

// Internal item structure
interface LRUItem<T> {
	expiry: number;
	key: any;
	prev: LRUItem<T> | null;
	next: LRUItem<T> | null;
	value: T;
}

TypeScript Usage Examples

import { LRU, lru } from "tiny-lru";

// Type-safe cache for user objects
interface User {
	id: number;
	name: string;
	email: string;
}

const userCache = new LRU<User>(1000, 300000); // 1000 users, 5 min TTL
userCache.set("user_123", { id: 123, name: "John", email: "john@example.com" });
const user: User | undefined = userCache.get("user_123"); // Fully typed

// Type-safe cache for API responses
interface APIResponse<T> {
	data: T;
	status: number;
	timestamp: number;
}

const apiCache = lru<APIResponse<any>>(500, 600000); // 10 min TTL
apiCache.set("endpoint_abc", {
	data: { results: [] },
	status: 200,
	timestamp: Date.now(),
});

// String cache with factory function
const stringCache = lru<string>(100);
stringCache.set("key1", "value1");
const value: string | undefined = stringCache.get("key1");

Modern Usage Patterns

1. LLM Response Caching

Cache expensive AI model responses with content-based keys and reasonable TTLs.

import { LRU } from "tiny-lru";

class LLMCache {
	constructor() {
		// Cache up to 1000 responses for 1 hour
		this.cache = new LRU(1000, 3600000); // 1 hour TTL
	}

	/**
	 * Generate cache key for LLM request
	 * @param {string} model - Model identifier
	 * @param {string} prompt - User prompt
	 * @param {object} params - Model parameters
	 */
	generateKey(model, prompt, params = {}) {
		const paramsHash = this.hashObject(params);
		const promptHash = this.hashString(prompt);
		return `llm:${model}:${promptHash}:${paramsHash}`;
	}

	async getResponse(model, prompt, params = {}) {
		const key = this.generateKey(model, prompt, params);

		// Check cache first
		const cached = this.cache.get(key);
		if (cached) {
			return { ...cached, fromCache: true };
		}

		// Make expensive API call
		const response = await this.callLLMAPI(model, prompt, params);

		// Cache the response
		this.cache.set(key, {
			response: response.text,
			tokens: response.tokens,
			timestamp: Date.now(),
		});

		return { ...response, fromCache: false };
	}

	hashString(str) {
		// Simple hash function for demonstration
		let hash = 0;
		for (let i = 0; i < str.length; i++) {
			const char = str.charCodeAt(i);
			hash = (hash << 5) - hash + char;
			hash = hash & hash; // Convert to 32-bit integer
		}
		return Math.abs(hash).toString(36);
	}

	hashObject(obj) {
		return this.hashString(JSON.stringify(obj, Object.keys(obj).sort()));
	}
}

2. API Response Caching with Rate Limiting

Cache external API responses with different TTLs based on data sensitivity.

import { LRU } from "tiny-lru";

class APICache {
	constructor() {
		this.caches = {
			// Fast-changing data: 5 minutes
			realtime: new LRU(500, 300000),
			// Moderate data: 30 minutes
			standard: new LRU(1000, 1800000),
			// Stable data: 24 hours
			stable: new LRU(2000, 86400000),
		};
	}

	async fetchUserProfile(userId, domain = "users") {
		const key = `${domain}:profile:${userId}`;
		const cache = this.caches.standard;

		const cached = cache.get(key);
		if (cached) {
			return cached;
		}

		const profile = await fetch(`/api/users/${userId}`).then((r) => r.json());

		// Use setWithEvicted to track what gets evicted for analytics
		const evicted = cache.setWithEvicted(key, profile);
		if (evicted) {
			console.log(`Evicted user profile: ${evicted.key}`);
		}

		return profile;
	}

	async fetchRealtimeData(symbol, domain = "market") {
		const key = `${domain}:price:${symbol}`;
		const cache = this.caches.realtime;

		const cached = cache.get(key);
		if (cached) {
			return cached;
		}

		const data = await fetch(`/api/market/${symbol}`).then((r) => r.json());
		cache.set(key, data);

		return data;
	}
}

3. Database Query Result Caching

Cache expensive database queries with intelligent cache invalidation.

import { LRU } from "tiny-lru";

class QueryCache {
	constructor() {
		// No TTL - manual invalidation
		this.cache = new LRU(10000);
		this.dependencyMap = new Map(); // Track query dependencies
	}

	async query(sql, params = [], dependencies = []) {
		const key = this.generateQueryKey(sql, params);

		const cached = this.cache.get(key);
		if (cached) {
			return cached;
		}

		const result = await this.executeQuery(sql, params);
		this.cache.set(key, result);

		// Track dependencies for invalidation
		this.trackDependencies(key, dependencies);

		return result;
	}

	generateQueryKey(sql, params) {
		const sqlHash = this.hashString(sql.replace(/\s+/g, " ").trim());
		const paramsHash = this.hashString(JSON.stringify(params));
		return `query:${sqlHash}:${paramsHash}`;
	}

	// Invalidate cache when specific tables change
	invalidateByTable(tableName) {
		const keysToDelete = [];

		for (const [key, deps] of this.dependencyMap.entries()) {
			if (deps.includes(tableName)) {
				keysToDelete.push(key);
			}
		}

		keysToDelete.forEach((key) => {
			this.cache.delete(key);
			this.dependencyMap.delete(key);
		});
	}

	// Get cache statistics using modern API methods
	getCacheStats() {
		const allKeys = this.cache.keys();
		const allEntries = this.cache.entries(); // Gets [key, value] pairs
		const dependentQueries = this.cache.values(
			allKeys.filter((key) => this.dependencyMap.has(key)),
		);

		return {
			totalQueries: this.cache.size,
			queryKeys: allKeys,
			dependentQueryCount: dependentQueries.length,
			lruOrder: allEntries.map(([key, _]) => key), // Least to most recent
		};
	}

	trackDependencies(key, dependencies) {
		if (dependencies.length > 0) {
			this.dependencyMap.set(key, dependencies);
		}
	}
}

4. Session and Authentication Caching

Cache user sessions and authentication tokens with proper security.

import { LRU } from "tiny-lru";

class AuthCache {
	constructor() {
		// Session cache: 30 minutes with TTL reset on access
		this.sessions = new LRU(10000, 1800000, true);
		// Token validation cache: 5 minutes, no reset
		this.tokens = new LRU(5000, 300000, false);
		// Permission cache: 15 minutes
		this.permissions = new LRU(5000, 900000);
	}

	cacheSession(sessionId, userData, domain = "app") {
		const key = `${domain}:session:${sessionId}`;
		this.sessions.set(key, {
			userId: userData.userId,
			permissions: userData.permissions,
			loginTime: Date.now(),
			lastActivity: Date.now(),
		});

		// Log when this session will expire
		const expiryTime = this.sessions.expiresAt(key);
		if (expiryTime) {
			console.log(`Session ${sessionId} expires at: ${new Date(expiryTime)}`);
		}
	}

	getSession(sessionId, domain = "app") {
		const key = `${domain}:session:${sessionId}`;
		return this.sessions.get(key);
	}

	cacheTokenValidation(tokenHash, isValid, userId = null) {
		const key = `auth:token:${tokenHash}`;
		this.tokens.set(key, { isValid, userId, validatedAt: Date.now() });
	}

	isTokenValid(tokenHash) {
		const key = `auth:token:${tokenHash}`;
		const cached = this.tokens.get(key);
		return cached?.isValid || false;
	}
}

Security Considerations

Multi-Domain Key Convention

Implement a hierarchical key naming convention to prevent cross-domain data leakage and improve operational security.

Key Format Specification

{domain}:{service}:{resource}:{identifier}[:{version}]

Domain Categories

const DOMAIN_PREFIXES = {
	// User-related data
	USER: "usr",
	// Authentication & authorization
	AUTH: "auth",
	// External API responses
	API: "api",
	// Database query results
	DB: "db",
	// Application logic
	APP: "app",
	// System/infrastructure
	SYS: "sys",
	// Analytics & metrics
	ANALYTICS: "analytics",
	// Machine learning / AI
	ML: "ml",
};

const SERVICE_PREFIXES = {
	// Authentication services
	LOGIN: "login",
	LOGOUT: "logout",
	REFRESH: "refresh",
	// Data services
	PROFILE: "profile",
	SETTINGS: "settings",
	// External integrations
	PAYMENT: "payment",
	EMAIL: "email",
	SMS: "sms",
};

Implementation Example

class SecureKeyManager {
	constructor(domain, service) {
		this.domain = domain;
		this.service = service;
		this.separator = ":";
	}

	generateKey(resource, identifier, version = null) {
		const parts = [this.domain, this.service, resource, identifier];
		if (version) parts.push(version);

		return parts.join(this.separator);
	}

	parseKey(key) {
		const parts = key.split(this.separator);
		return {
			domain: parts[0],
			service: parts[1],
			resource: parts[2],
			identifier: parts[3],
			version: parts[4] || null,
		};
	}

	// Validate key follows security convention
	validateKey(key) {
		const parts = key.split(this.separator);

		if (parts.length < 4 || parts.length > 5) {
			throw new Error("Invalid key format: must have 4-5 parts");
		}

		if (!Object.values(DOMAIN_PREFIXES).includes(parts[0])) {
			throw new Error(`Invalid domain prefix: ${parts[0]}`);
		}

		return true;
	}
}

// Usage examples
const authKeys = new SecureKeyManager(DOMAIN_PREFIXES.AUTH, SERVICE_PREFIXES.LOGIN);
const userKeys = new SecureKeyManager(DOMAIN_PREFIXES.USER, SERVICE_PREFIXES.PROFILE);
const mlKeys = new SecureKeyManager(DOMAIN_PREFIXES.ML, "llm");

// Generate secure keys
const sessionKey = authKeys.generateKey("session", "abc123def456");
// Result: "auth:login:session:abc123def456"

const profileKey = userKeys.generateKey("data", "12345", "v2");
// Result: "usr:profile:data:12345:v2"

const llmKey = mlKeys.generateKey("response", "gpt4-prompt-hash");
// Result: "ml:llm:response:gpt4-prompt-hash"

Cache Isolation

class IsolatedCacheManager {
	constructor() {
		this.caches = new Map();
	}

	getCache(domain, service, maxSize = 1000, ttl = 0) {
		const cacheKey = `${domain}:${service}`;

		if (!this.caches.has(cacheKey)) {
			this.caches.set(cacheKey, new LRU(maxSize, ttl));
		}

		return this.caches.get(cacheKey);
	}

	// Secure method to access only your domain's cache
	accessCache(domain, service, operation, ...args) {
		if (!this.isAuthorized(domain, service)) {
			throw new Error("Unauthorized cache access");
		}

		const cache = this.getCache(domain, service);
		return cache[operation](...args);
	}

	isAuthorized(domain, service) {
		// Implement your authorization logic
		return true; // Simplified for example
	}

	// Clear caches by domain for security incidents
	clearDomain(domain) {
		for (const [key, cache] of this.caches.entries()) {
			if (key.startsWith(`${domain}:`)) {
				cache.clear();
			}
		}
	}
}

Performance Characteristics

Benchmarking Setup

// Performance testing setup
import { LRU } from "tiny-lru";
import { performance } from "perf_hooks";

class PerformanceTester {
	constructor() {
		this.cache = new LRU(10000);
	}

	async benchmarkOperations(iterations = 100000) {
		const results = {};

		// Benchmark set operations
		const setStart = performance.now();
		for (let i = 0; i < iterations; i++) {
			this.cache.set(`key_${i}`, `value_${i}`);
		}
		const setEnd = performance.now();
		results.setOpsPerSecond = iterations / ((setEnd - setStart) / 1000);

		// Benchmark get operations
		const getStart = performance.now();
		for (let i = 0; i < iterations; i++) {
			this.cache.get(`key_${i % 5000}`); // Mix hits and misses
		}
		const getEnd = performance.now();
		results.getOpsPerSecond = iterations / ((getEnd - getStart) / 1000);

		return results;
	}
}

Memory Usage Patterns

graph LR
    A["Memory Usage"] --> B["Linear Growth"]
    B --> C["Until Max Size"]
    C --> D["Constant Memory"]
    D --> E["With Eviction"]

    F["GC Pressure"] --> G["Low"]
    F --> H["Object Reuse"]
    F --> I["Minimal Allocation"]

    style A fill:#2563eb,stroke:#1e40af,stroke-width:2px,color:#ffffff
    style F fill:#ea580c,stroke:#c2410c,stroke-width:2px,color:#ffffff

Integration Examples

Express.js Middleware

import { LRU } from "tiny-lru";

function createCacheMiddleware(options = {}) {
	const cache = new LRU(
		options.maxSize || 1000,
		options.ttl || 300000, // 5 minutes default
	);

	return (req, res, next) => {
		// Generate cache key from request
		const key = `http:${req.method}:${req.path}:${JSON.stringify(req.query)}`;

		// Check cache
		const cached = cache.get(key);
		if (cached) {
			res.set(cached.headers);
			return res.status(cached.status).json(cached.data);
		}

		// Intercept response
		const originalSend = res.json;
		res.json = function (data) {
			// Cache successful responses
			if (res.statusCode < 400) {
				cache.set(key, {
					status: res.statusCode,
					headers: res.getHeaders(),
					data: data,
				});
			}

			return originalSend.call(this, data);
		};

		next();
	};
}

// Usage
app.use("/api/users", createCacheMiddleware({ ttl: 600000 })); // 10 minutes

React Cache Hook

import { useMemo, useRef } from "react";
import { LRU } from "tiny-lru";

function useCache(maxSize = 100, ttl = 300000) {
	const cache = useRef(null);

	if (!cache.current) {
		cache.current = new LRU(maxSize, ttl);
	}

	const memoizedCache = useMemo(
		() => ({
			get: (key) => cache.current.get(key),
			set: (key, value) => cache.current.set(key, value),
			delete: (key) => cache.current.delete(key),
			clear: () => cache.current.clear(),
			has: (key) => cache.current.has(key),
		}),
		[],
	);

	return memoizedCache;
}

// Usage in component
function UserProfile({ userId }) {
	const cache = useCache(50, 600000); // 10 minutes
	const [profile, setProfile] = useState(null);

	useEffect(() => {
		const cacheKey = `usr:profile:data:${userId}`;
		const cached = cache.get(cacheKey);

		if (cached) {
			setProfile(cached);
		} else {
			fetchUserProfile(userId).then((data) => {
				cache.set(cacheKey, data);
				setProfile(data);
			});
		}
	}, [userId, cache]);

	return <div>{/* render profile */}</div>;
}

Best Practices

1. Cache Sizing Strategy

flowchart TD
    A["Determine Cache Size"] --> B{"Memory Available?"}
    B -->|Limited| C["Conservative Sizing<br/>~1000 items"]
    B -->|Abundant| D["Aggressive Sizing<br/>~10000+ items"]

    C --> E["Monitor Hit Rate"]
    D --> E

    E --> F{"Hit Rate < 80%?"}
    F -->|Yes| G["Increase Size<br/>or Adjust TTL"]
    F -->|No| H["Optimize Keys<br/>or Logic"]

    G --> I["Re-evaluate"]
    H --> I

    style A fill:#2563eb,stroke:#1e40af,stroke-width:2px,color:#ffffff
    style I fill:#059669,stroke:#047857,stroke-width:2px,color:#ffffff
    style B fill:#7c3aed,stroke:#5b21b6,stroke-width:2px,color:#ffffff
    style F fill:#7c3aed,stroke:#5b21b6,stroke-width:2px,color:#ffffff

2. TTL Selection Guidelines

Data TypeRecommended TTLRationale
User Sessions30-60 minutesBalance security and UX
API Responses5-30 minutesDepends on data freshness needs
Database Queries15-60 minutesBased on update frequency
LLM Responses1-24 hoursExpensive to regenerate
Static Content24+ hoursRarely changes

3. Error Handling

class RobustCache {
	constructor(maxSize, ttl) {
		try {
			this.cache = new LRU(maxSize, ttl);
		} catch (error) {
			console.error("Cache initialization failed:", error);
			this.cache = null;
		}
	}

	safeGet(key) {
		try {
			return this.cache?.get(key);
		} catch (error) {
			console.error("Cache get failed:", error);
			return undefined;
		}
	}

	safeSet(key, value) {
		try {
			return this.cache?.set(key, value);
		} catch (error) {
			console.error("Cache set failed:", error);
			return false;
		}
	}
}

4. Monitoring and Metrics

class MonitoredCache {
	constructor(maxSize, ttl) {
		this.cache = new LRU(maxSize, ttl);
		this.metrics = {
			hits: 0,
			misses: 0,
			sets: 0,
			deletes: 0,
			evictions: 0,
		};
	}

	get(key) {
		const value = this.cache.get(key);
		this.metrics[value !== undefined ? "hits" : "misses"]++;
		return value;
	}

	set(key, value) {
		const hadKey = this.cache.has(key);
		const wasAtCapacity = this.cache.size >= this.cache.max;

		const result = this.cache.set(key, value);

		this.metrics.sets++;
		if (!hadKey && wasAtCapacity && this.cache.max > 0) {
			this.metrics.evictions++;
		}

		return result;
	}

	getHitRate() {
		const total = this.metrics.hits + this.metrics.misses;
		return total > 0 ? this.metrics.hits / total : 0;
	}

	getMetrics() {
		return {
			...this.metrics,
			hitRate: this.getHitRate(),
			size: this.cache.size,
			maxSize: this.cache.max,
		};
	}
}

Build Configuration and Distribution

The library uses Rollup for building and distributing multiple module formats to support different JavaScript environments.

Build Process

// rollup.config.js
export default [
	{
		input: "./src/lru.js",
		output: [
			// CommonJS for Node.js
			{ format: "cjs", file: "dist/tiny-lru.cjs" },
			// ES Modules for modern bundlers
			{ format: "esm", file: "dist/tiny-lru.js" },
			// Minified ES Modules
			{ format: "esm", file: "dist/tiny-lru.min.js", plugins: [terser()] },
		],
	},
];

Package Exports

{
	"source": "src/lru.js",
	"main": "dist/tiny-lru.cjs",
	"exports": {
		"types": "./types/lru.d.ts",
		"import": "./dist/tiny-lru.js",
		"require": "./dist/tiny-lru.cjs"
	},
	"type": "module",
	"types": "types/lru.d.ts"
}

Available Formats

  • ESM: dist/tiny-lru.js - ES Modules for modern bundlers
  • CommonJS: dist/tiny-lru.cjs - Node.js compatible format
  • Minified: dist/tiny-lru.min.js - Minified ESM
  • TypeScript: types/lru.d.ts - Complete type definitions

Development Commands

# Build all distribution formats
npm run build

# Run tests with coverage
npm test

# Lint code
npm run lint

# Run benchmarks
npm run benchmark:all

File Structure

tiny-lru/
├── src/
│   └── lru.js              # Source implementation
├── types/
│   └── lru.d.ts           # TypeScript definitions
├── dist/                   # Built distributions (generated)
│   ├── tiny-lru.js        # ES Modules
│   ├── tiny-lru.cjs       # CommonJS
│   └── tiny-lru.min.js    # Minified ESM
├── tests/
│   ├── unit/              # Unit tests with Node.js built-in test runner
│   └── integration/       # Integration tests
├── benchmarks/            # Performance benchmarks
└── docs/                  # Documentation

Conclusion

The tiny-lru library provides a robust foundation for caching in modern applications. By following these patterns, security guidelines, and best practices, you can build efficient, secure, and maintainable caching solutions for contemporary use cases including AI/ML applications, API gateways, and high-performance web services.

Remember to:

  • Use proper key namespacing for security
  • Monitor cache performance and hit rates
  • Implement appropriate TTL strategies
  • Handle errors gracefully
  • Consider memory constraints in your sizing decisions
  • Choose the appropriate module format for your environment
  • Leverage TypeScript support for type safety