epic-caching（史诗级缓存）

Epic Stack: Caching

When to use this skill

Use this skill when you need to:

• Cache results of expensive queries
• Cache responses from external APIs
• Optimize performance of data that doesn't change frequently
• Implement stale-while-revalidate
• Manage cache invalidation
• Integrate cache with server timing

Patterns and conventions

Caching Philosophy

Following Epic Web principles:

Weigh the cost-benefit of performance optimizations - Caching adds complexity. Only add cache when there's a clear, measurable benefit. Don't cache "just in case" - cache when you have a real performance problem that caching solves.

When NOT to use cache:

• Data that changes frequently (cache invalidation becomes a problem)
• Data that's already fast to fetch (no measurable benefit)
• Data that's only fetched once (no benefit from caching)
• Simple queries that don't need optimization
• When cache invalidation logic becomes more complex than the problem it solves

Example - Evaluating cost-benefit:

// ✅ Good - Cache expensive external API call
export async function getGitHubEvents({
	username,
	timings,
}: {
	username: string
	timings?: Timings
}) {
	return await cachified({
		key: `github:${username}:events`,
		cache,
		timings,
		getFreshValue: async () => {
			// Expensive: External API call, rate limits, network latency
			const response = await fetch(
				`https://api.github.com/users/${username}/events/public`,
			)
			return await response.json()
		},
		checkValue: GitHubEventSchema.array(),
		ttl: 1000 * 60 * 60, // 1 hour - reasonable for external data
	})
}

// ❌ Avoid - Caching simple, fast database query
export async function getUser({ userId }: { userId: string }) {
	// This query is already fast - caching adds complexity without benefit
	return await cachified({
		key: `user:${userId}`,
		cache,
		getFreshValue: async () => {
			// Simple query, already fast
			return await prisma.user.findUnique({
				where: { id: userId },
				select: { id: true, username: true },
			})
		},
		ttl: 1000 * 60 * 5,
	})
	// Better: Just query directly without cache
}

Two Types of Cache

Epic Stack provides two types of cache:

1. SQLite Cache - Long-lived, replicated with LiteFS

- Persistent across restarts - Replicated across all instances - Ideal for data that changes infrequently

1. LRU Cache - Short-lived, in-memory

- Cleared on restart - Not replicated (only on current instance) - Ideal for deduplication and temporary cache

Using cachified

Epic Stack uses @epic-web/cachified as an abstraction for cache management.

Basic import:

import { cachified, cache } from '#app/utils/cache.server.ts'
import { type Timings } from '#app/utils/timing.server.ts'

Basic structure:

export async function getCachedData({
	timings,
}: {
	timings?: Timings
} = {}) {
	return await cachified({
		key: 'my-cache-key',
		cache,
		timings,
		getFreshValue: async () => {
			// Get fresh data
			return await fetchDataFromAPI()
		},
		checkValue: z.object({
			/* schema */
		}), // Validation with Zod
		ttl: 1000 * 60 * 60 * 24, // 24 hours
		staleWhileRevalidate: 1000 * 60 * 60 * 24 * 30, // 30 days
	})
}

Cache Keys

Naming conventions:

• Use format: entity:identifier:data
• Examples:

- user:${userId}:profile - note:${noteId}:full - api:github:events - tito:scheduled-events

Avoid:

• Keys that are too long
• Keys with special characters
• Keys that don't clearly identify the content

TTL (Time To Live)

Define TTL:

await cachified({
	key: 'my-key',
	cache,
	getFreshValue: () => fetchData(),
	ttl: 1000 * 60 * 60 * 24, // 24 hours in milliseconds
})

Null TTL to never expire:

ttl: null, // Never expires (not recommended unless necessary)

Stale-While-Revalidate (SWR)

SWR allows returning stale data while fresh data is fetched in the background.

Example:

await cachified({
	key: 'my-key',
	cache,
	getFreshValue: () => fetchData(),
	ttl: 1000 * 60 * 60 * 24, // 24 hours - after this it's considered stale
	staleWhileRevalidate: 1000 * 60 * 60 * 24 * 30, // 30 days - up to here returns stale while revalidating
})

Behavior:

• Less than 24h: Returns cache, no request made
• 24h - 30 days: Returns stale cache immediately, updates in background
• More than 30 days: Waits for fresh data before returning

Validation with Zod

Always validate cached data with Zod:

import { z } from 'zod'

const EventSchema = z.object({
	id: z.string(),
	title: z.string(),
	date: z.string(),
})

export async function getEvents({ timings }: { timings?: Timings } = {}) {
	return await cachified({
		key: 'events:all',
		cache,
		timings,
		getFreshValue: async () => {
			const response = await fetch('https://api.example.com/events')
			return await response.json()
		},
		checkValue: EventSchema.array(), // Validates it's an array of events
		ttl: 1000 * 60 * 60 * 24, // 24 hours
	})
}

If cached data doesn't pass validation, fresh data is fetched.

Server Timing Integration

Integrate cache with server timing for monitoring:

import { type Timings } from '#app/utils/timing.server.ts'

export async function loader({ request }: Route.LoaderArgs) {
	const timings: Timings = {}

	const events = await getEvents({ timings })

	// Timings are automatically added to headers
	return json(
		{ events },
		{
			headers: combineServerTimings(timings),
		},
	)
}

Cache Invalidation

Invalidate by key:

import { cache } from '#app/utils/cache.server.ts'

await cache.delete('user:123:profile')

Invalidate multiple keys:

// Search and delete matching keys
import { searchCacheKeys } from '#app/utils/cache.server.ts'

const keys = await searchCacheKeys('user:123', 100)
await Promise.all(keys.map((key) => cache.delete(key)))

Invalidate entire SQLite cache:

// Use admin dashboard or
await cache.clear() // If available

Using LRU Cache

For temporary data, use LRU cache directly:

import { lru } from '#app/utils/cache.server.ts'

// LRU cache is useful for:
// - Request deduplication
// - Very temporary cache (< 5 minutes)
// - Data that doesn't need to persist

const cachedValue = lru.get('temp-key')
if (!cachedValue) {
	const freshValue = await computeExpensiveValue()
	lru.set('temp-key', freshValue, { ttl: 1000 * 60 * 5 }) // 5 minutes
	return freshValue
}
return cachedValue

Multi-Region Cache

With LiteFS, SQLite cache is automatically replicated:

Behavior:

• Only the primary instance writes to cache
• Replicas can read from cache
• Writes are automatically synchronized

Best practices:

• Don't assume all writes are immediate
• Use ensurePrimary() if you need to guarantee writes

import { ensurePrimary } from '#app/utils/litefs.server.ts'

export async function action({ request }: Route.ActionArgs) {
	await ensurePrimary() // Ensure we're on primary instance

	// Invalidate cache
	await cache.delete('my-key')

	// ...
}

Error Handling

Handle errors in getFreshValue:

await cachified({
	key: 'my-key',
	cache,
	getFreshValue: async () => {
		try {
			return await fetchData()
		} catch (error) {
			console.error('Failed to fetch fresh data:', error)
			throw error // Re-throw so cachified handles it
		}
	},
	// If getFreshValue fails and there's stale cache, it returns it
	fallbackToCache: true, // Default: true
})

Cache Admin Dashboard

Epic Stack includes a dashboard to manage cache:

Route: /admin/cache

Features:

• View all cache keys
• Search keys
• View details of a key
• Delete keys
• Clear entire cache

Common examples

Example 1: Cache external API response

// app/utils/api.server.ts
import { cachified, cache } from '#app/utils/cache.server.ts'
import { type Timings } from '#app/utils/timing.server.ts'
import { z } from 'zod'

const GitHubEventSchema = z.object({
	id: z.string(),
	type: z.string(),
	actor: z.object({
		login: z.string(),
	}),
	created_at: z.string(),
})

export async function getGitHubEvents({
	username,
	timings,
}: {
	username: string
	timings?: Timings
}) {
	return await cachified({
		key: `github:${username}:events`,
		cache,
		timings,
		getFreshValue: async () => {
			const response = await fetch(
				`https://api.github.com/users/${username}/events/public`,
			)
			if (!response.ok) {
				throw new Error(`GitHub API error: ${response.statusText}`)
			}
			const data = await response.json()
			return data
		},
		checkValue: GitHubEventSchema.array(),
		ttl: 1000 * 60 * 60, // 1 hour
		staleWhileRevalidate: 1000 * 60 * 60 * 24, // 24 hours
	})
}

Example 2: Cache Prisma query

// app/utils/user.server.ts
import { cachified, cache } from '#app/utils/cache.server.ts'
import { prisma } from '#app/utils/db.server.ts'
import { z } from 'zod'

const UserStatsSchema = z.object({
	totalNotes: z.number(),
	totalLikes: z.number(),
	joinDate: z.string(),
})

export async function getUserStats({
	userId,
	timings,
}: {
	userId: string
	timings?: Timings
}) {
	return await cachified({
		key: `user:${userId}:stats`,
		cache,
		timings,
		getFreshValue: async () => {
			const [totalNotes, totalLikes, user] = await Promise.all([
				prisma.note.count({ where: { ownerId: userId } }),
				prisma.like.count({ where: { userId } }),
				prisma.user.findUnique({
					where: { id: userId },
					select: { createdAt: true },
				}),
			])

			return {
				totalNotes,
				totalLikes,
				joinDate: user?.createdAt.toISOString() ?? '',
			}
		},
		checkValue: UserStatsSchema,
		ttl: 1000 * 60 * 5, // 5 minutes
		staleWhileRevalidate: 1000 * 60 * 60, // 1 hour
	})
}

Example 3: Invalidate cache after mutation

// app/routes/users/$username/notes/new.tsx
export async function action({ request }: Route.ActionArgs) {
	const userId = await requireUserId(request)
	const formData = await request.formData()

	// ... validate and create note

	const note = await prisma.note.create({
		data: {
			title,
			content,
			ownerId: userId,
		},
		include: { owner: true },
	})

	// Invalidate related cache
	await Promise.all([
		cache.delete(`user:${userId}:notes`),
		cache.delete(`user:${userId}:stats`),
		cache.delete(`note:${note.id}:full`),
	])

	return redirect(`/users/${note.owner.username}/notes/${note.id}`)
}

Example 4: Cache with dependencies

export async function getUserWithNotes({
	userId,
	timings,
}: {
	userId: string
	timings?: Timings
}) {
	const user = await cachified({
		key: `user:${userId}:profile`,
		cache,
		timings,
		getFreshValue: async () => {
			return await prisma.user.findUnique({
				where: { id: userId },
				select: {
					id: true,
					username: true,
					name: true,
				},
			})
		},
		checkValue: z
			.object({
				id: z.string(),
				username: z.string(),
				name: z.string().nullable(),
			})
			.nullable(),
		ttl: 1000 * 60 * 30, // 30 minutes
	})

	const notes = await cachified({
		key: `user:${userId}:notes`,
		cache,
		timings,
		getFreshValue: async () => {
			return await prisma.note.findMany({
				where: { ownerId: userId },
				select: {
					id: true,
					title: true,
					updatedAt: true,
				},
				orderBy: { updatedAt: 'desc' },
			})
		},
		checkValue: z.array(
			z.object({
				id: z.string(),
				title: z.string(),
				updatedAt: z.date(),
			}),
		),
		ttl: 1000 * 60 * 10, // 10 minutes
	})

	return { user, notes }
}

Example 5: Use LRU for deduplication

// Avoid multiple simultaneous requests to the same URL
const requestCache = new Map<string, Promise<any>>()

export async function fetchWithDedup(url: string) {
	if (requestCache.has(url)) {
		return requestCache.get(url)
	}

	const promise = fetch(url).then((res) => res.json())
	requestCache.set(url, promise)

	// Clean up after 1 second
	setTimeout(() => {
		requestCache.delete(url)
	}, 1000)

	return promise
}

Common mistakes to avoid

• ❌ Caching without measuring benefit: Only add cache when there's a clear, measurable performance problem
• ❌ Caching simple, fast queries: Don't cache data that's already fast to fetch - it adds complexity without benefit
• ❌ Caching frequently changing data: Cache invalidation becomes more complex than the problem it solves
• ❌ Caching sensitive data: Never cache passwords, tokens, or sensitive personal data
• ❌ TTL too long: Avoid very long TTLs (> 1 week) unless absolutely necessary
• ❌ Not validating cached data: Always use checkValue with Zod to validate data
• ❌ Forgetting to invalidate cache: Invalidate cache after mutations
• ❌ Assuming cache always works: Cache can fail, always handle errors
• ❌ Keys too long or ambiguous: Use consistent and descriptive format
• ❌ Not using timings: Integrate with server timing for monitoring
• ❌ Forgetting stale-while-revalidate: Use SWR for better UX when appropriate
• ❌ Over-caching: Too much caching makes the system harder to understand and debug

References

• Epic Stack Caching Docs
• Epic Web Principles
• @epic-web/cachified
• app/utils/cache.server.ts - Cache implementation
• app/routes/admin/cache/ - Admin dashboard
• app/utils/timing.server.ts - Server timing utilities