caching

caching

Use when same data is fetched repeatedly. Use when database queries are slow. Use when implementing caching without invalidation strategy.

4Sterne
0Forks
Aktualisiert 1/23/2026
SKILL.md
readonlyread-only
name
caching
description

Use when same data is fetched repeatedly. Use when database queries are slow. Use when implementing caching without invalidation strategy.

Caching

Overview

Cache aggressively, but always have an invalidation strategy.

Caching improves performance dramatically, but stale data causes bugs. Every cache needs a plan for freshness.

When to Use

  • Same data fetched repeatedly
  • Expensive computations
  • Slow database queries
  • External API rate limits
  • Asked to "just add caching"

The Iron Rule

NEVER add a cache without defining its invalidation strategy.

No exceptions:

  • Not for "it rarely changes"
  • Not for "TTL is enough"
  • Not for "we'll figure it out later"
  • Not for "users can refresh"

Detection: Cache Without Strategy Smell

If cache has no invalidation plan, STOP:

// ❌ VIOLATION: Cache without invalidation strategy
const cache = new Map();

async function getUser(id: string) {
  if (cache.has(id)) {
    return cache.get(id);  // Could be stale forever!
  }
  
  const user = await db.users.findById(id);
  cache.set(id, user);  // When does this expire? When user updates?
  return user;
}

Problems:

  • User updates name → cache shows old name
  • No memory limit → memory leak
  • No TTL → stale forever
  • Distributed system → multiple stale copies

The Correct Pattern: Cache with Strategy

// ✅ CORRECT: Cache with TTL and invalidation

interface CacheEntry<T> {
  data: T;
  expiresAt: number;
}

class UserCache {
  private cache = new Map<string, CacheEntry<User>>();
  private TTL_MS = 5 * 60 * 1000; // 5 minutes
  
  async get(id: string): Promise<User> {
    const cached = this.cache.get(id);
    
    if (cached && cached.expiresAt > Date.now()) {
      return cached.data;
    }
    
    const user = await db.users.findById(id);
    this.set(id, user);
    return user;
  }
  
  set(id: string, user: User): void {
    this.cache.set(id, {
      data: user,
      expiresAt: Date.now() + this.TTL_MS
    });
  }
  
  // Explicit invalidation on updates
  invalidate(id: string): void {
    this.cache.delete(id);
  }
  
  invalidateAll(): void {
    this.cache.clear();
  }
}

// Usage with invalidation on write
async function updateUser(id: string, data: UpdateUserDto) {
  const user = await db.users.update(id, data);
  userCache.invalidate(id);  // Clear stale cache
  return user;
}

Cache Invalidation Strategies

1. Time-Based (TTL)

// Good for: data that can be slightly stale
const TTL = 60 * 1000; // 1 minute
cache.set(key, value, { ttl: TTL });

2. Write-Through

// Good for: data you control writes for
async function updateProduct(id, data) {
  const product = await db.products.update(id, data);
  await cache.set(`product:${id}`, product);  // Update cache on write
  return product;
}

3. Event-Based

// Good for: distributed systems
eventBus.on('user.updated', (userId) => {
  cache.delete(`user:${userId}`);
});

eventBus.on('product.priceChanged', (productId) => {
  cache.delete(`product:${productId}`);
});

4. Cache-Aside (Lazy)

// Good for: read-heavy, tolerance for staleness
async function getProduct(id) {
  let product = await cache.get(`product:${id}`);
  if (!product) {
    product = await db.products.findById(id);
    await cache.set(`product:${id}`, product, { ttl: 300 });
  }
  return product;
}

What to Cache

Good to Cache Bad to Cache
User profiles Session tokens
Product catalog Payment status
Configuration Real-time inventory
API responses User-specific calculations
Computed aggregates Rapidly changing data

Redis Example

import Redis from 'ioredis';

const redis = new Redis();

class ProductCache {
  private prefix = 'product:';
  private ttl = 300; // 5 minutes
  
  async get(id: string): Promise<Product | null> {
    const cached = await redis.get(this.prefix + id);
    return cached ? JSON.parse(cached) : null;
  }
  
  async set(id: string, product: Product): Promise<void> {
    await redis.setex(
      this.prefix + id,
      this.ttl,
      JSON.stringify(product)
    );
  }
  
  async invalidate(id: string): Promise<void> {
    await redis.del(this.prefix + id);
  }
  
  async invalidatePattern(pattern: string): Promise<void> {
    const keys = await redis.keys(this.prefix + pattern);
    if (keys.length) await redis.del(...keys);
  }
}

Pressure Resistance Protocol

1. "It Rarely Changes"

Pressure: "This data almost never updates"

Response: "Almost never" still means sometimes. When it does, stale cache = bugs.

Action: Add TTL at minimum. Add invalidation on write.

2. "TTL Is Enough"

Pressure: "We'll just expire after 5 minutes"

Response: 5 minutes of stale data might be unacceptable. User updates profile, sees old data.

Action: TTL + write-through invalidation.

3. "We'll Figure It Out Later"

Pressure: "Just add caching, we'll handle staleness if it's a problem"

Response: Staleness bugs are hard to debug. Design invalidation upfront.

Action: No cache without invalidation strategy defined.

Red Flags - STOP and Reconsider

  • Cache with no TTL
  • No invalidation on data updates
  • "Users can refresh to see new data"
  • In-memory cache in distributed system
  • Cache without memory limits

All of these mean: Define invalidation strategy.

Quick Reference

Pattern Use When Invalidation
TTL only Staleness OK Automatic expiry
Write-through You control writes Update cache on write
Event-based Distributed system Pub/sub on changes
Cache-aside Read-heavy TTL + manual invalidate

Common Rationalizations (All Invalid)

Excuse Reality
"Rarely changes" Rarely ≠ never. Plan for it.
"TTL is enough" TTL + invalidation is better.
"Figure it out later" Staleness bugs are hard to trace.
"Users can refresh" That's a bug, not a feature.
"It's just for performance" Stale data breaks functionality.

The Bottom Line

Every cache needs: TTL, size limit, and invalidation strategy.

Cache aggressively for performance. But always know how the cache gets invalidated when data changes. "It rarely changes" is not a strategy.

You Might Also Like

Related Skills

coding-agent

coding-agent

179Kdev-codegen

Run Codex CLI, Claude Code, OpenCode, or Pi Coding Agent via background process for programmatic control.

openclaw avataropenclaw
Holen
add-uint-support

add-uint-support

97Kdev-codegen

Add unsigned integer (uint) type support to PyTorch operators by updating AT_DISPATCH macros. Use when adding support for uint16, uint32, uint64 types to operators, kernels, or when user mentions enabling unsigned types, barebones unsigned types, or uint support.

pytorch avatarpytorch
Holen
at-dispatch-v2

at-dispatch-v2

97Kdev-codegen

Convert PyTorch AT_DISPATCH macros to AT_DISPATCH_V2 format in ATen C++ code. Use when porting AT_DISPATCH_ALL_TYPES_AND*, AT_DISPATCH_FLOATING_TYPES*, or other dispatch macros to the new v2 API. For ATen kernel files, CUDA kernels, and native operator implementations.

pytorch avatarpytorch
Holen
skill-writer

skill-writer

97Kdev-codegen

Guide users through creating Agent Skills for Claude Code. Use when the user wants to create, write, author, or design a new Skill, or needs help with SKILL.md files, frontmatter, or skill structure.

pytorch avatarpytorch
Holen

Implements JavaScript classes in C++ using JavaScriptCore. Use when creating new JS classes with C++ bindings, prototypes, or constructors.

oven-sh avataroven-sh
Holen

Creates JavaScript classes using Bun's Zig bindings generator (.classes.ts). Use when implementing new JS APIs in Zig with JSC integration.

oven-sh avataroven-sh
Holen