Cache Strategies - 🌲 Sonamu

Effective caching requires choosing strategies that match data characteristics. This guide covers various cache strategies and how to use them.

TTL (Time To Live) Strategy

What is TTL?

TTL is the duration a cache remains valid. Once TTL expires, the cache is invalidated and fresh data is fetched.

@cache({ ttl: "10m" })  // Expires after 10 minutes
async getData() {
  return await this.expensiveQuery();
}

TTL Settings by Data Characteristics

Static Data
Semi-static Data
Dynamic Data
Very Short-lived Data

Data that rarely changes

// Configuration data (almost never changes)
@cache({ ttl: "forever", tags: ["config"] })
async getConfig() {
  return this.findOne(['key', 'app_config']);
}

// Categories (occasionally changes)
@cache({ ttl: "1d", tags: ["category"] })
async getCategoryTree() {
  return this.buildCategoryTree();
}

TTL: "forever", "1d", "1w"

Data that changes periodically

// User profile (occasionally changes)
@cache({ ttl: "1h", tags: ["user"] })
async getUserProfile(id: number) {
  return this.findOne(['id', id]);
}

// Product info (inventory/price changes)
@cache({ ttl: "30m", tags: ["product"] })
async getProduct(id: number) {
  return this.findOne(['id', id]);
}

TTL: "30m", "1h", "6h"

Data that changes frequently

// Post list (frequently added)
@cache({ ttl: "5m", tags: ["post", "list"] })
async getPostList() {
  return this.findMany({ num: 20, page: 1 });
}

// Real-time stats (constantly changing)
@cache({ ttl: "1m", tags: ["stats"] })
async getLiveStats() {
  return this.calculateStats();
}

TTL: "1m", "5m", "10m"

Data that changes by the second

// API rate limiting
@cache({ ttl: "10s" })
async getRateLimitStatus(userId: number) {
  return this.checkRateLimit(userId);
}

// Real-time ranking
@cache({ ttl: "5s", tags: ["ranking"] })
async getRealtimeRanking() {
  return this.calculateRanking();
}

TTL: "5s", "10s", "30s"

TTL Units

// Seconds
@cache({ ttl: "30s" })

// Minutes
@cache({ ttl: "10m" })

// Hours
@cache({ ttl: "2h" })

// Days
@cache({ ttl: "7d" })

// Weeks
@cache({ ttl: "2w" })

// Permanent
@cache({ ttl: "forever" })

// Milliseconds (number)
@cache({ ttl: 60000 })  // 60000ms = 1 minute

Grace Period (Stale-While-Revalidate)

What is Grace?

Grace is a strategy that returns stale cache even after TTL expiration while refreshing in the background.

@cache({
  ttl: "5m",     // Expires after 5 minutes
  grace: "1h"    // Use stale value for 1 hour after expiration
})
async getExpensiveData() {
  return await this.heavyComputation();
}

How It Works

When to Use Grace

Use Grace ✅

Heavy computations/queries

Aggregate statistics
Complex join queries
External API calls
Large data processing

@cache({
  ttl: "10m",
  grace: "1h"
})
async getDashboard() {
  return await this.complexAggregation();
}

Grace Unnecessary ❌

Fast queries

Simple SELECT
Index lookups
Cached data

@cache({ ttl: "5m" })
// No grace
async getUser(id: number) {
  return this.findOne(['id', id]);
}

Grace Practical Examples

class AnalyticsModelClass extends BaseModelClass {
  // Dashboard stats (heavy aggregation)
  @cache({
    ttl: "5m",      // Refresh every 5 minutes
    grace: "2h",    // Allow stale for 2 hours
    tags: ["analytics"]
  })
  @api()
  async getDashboardStats() {
    const [userCount, orderCount, revenue] = await Promise.all([
      this.countUsers(),
      this.countOrders(),
      this.calculateRevenue(),
    ]);

    return { userCount, orderCount, revenue };
  }

  // Real-time ranking (very heavy)
  @cache({
    ttl: "1m",      // Refresh every minute
    grace: "10m",   // Allow stale for 10 minutes
    tags: ["ranking"]
  })
  @api()
  async getRealTimeRanking() {
    // Aggregate millions of records
    return await this.calculateRankingWithHeavyComputation();
  }
}

Grace vs Long TTL

Grace Strategy:

@cache({ ttl: "5m", grace: "1h" })

Mostly fresh data (within 5 minutes)
Returns stale immediately on expiration (fast)
Background refresh

Long TTL:

@cache({ ttl: "1h" })

Data up to 1 hour old
Recalculates on expiration (slow)

Recommendation: Use Grace for heavy operations

Namespace Strategy

What is Namespace?

Namespace is a feature that logically isolates caches.

const userCache = Sonamu.cache.namespace("user:123");
const adminCache = Sonamu.cache.namespace("admin");

// Same key but different namespace means different cache
await userCache.set({ key: "data", value: "user data" });
await adminCache.set({ key: "data", value: "admin data" });

await userCache.get({ key: "data" });   // "user data"
await adminCache.get({ key: "data" });  // "admin data"

Per-User Isolation

class UserDataModelClass extends BaseModelClass {
  @api()
  async getMyData(ctx: Context) {
    const userId = ctx.user.id;

    // Per-user namespace
    const userCache = Sonamu.cache.namespace(`user:${userId}`);

    return userCache.getOrSet({
      key: "preferences",
      ttl: "1h",
      factory: async () => {
        return this.getUserPreferences(userId);
      }
    });
  }

  @api()
  async updateMyData(ctx: Context, data: any) {
    const userId = ctx.user.id;
    const result = await this.saveUserData(userId, data);

    // Delete only that user's cache
    const userCache = Sonamu.cache.namespace(`user:${userId}`);
    await userCache.clear();

    return result;
  }
}

Advantages:

User A’s changes don’t affect User B
Selective invalidation possible

Multi-Tenant

class TenantServiceModelClass extends BaseModelClass {
  @api()
  async getData(tenantId: number) {
    // Per-tenant namespace
    const tenantCache = Sonamu.cache.namespace(`tenant:${tenantId}`);

    return tenantCache.getOrSet({
      key: "service-data",
      ttl: "1h",
      tags: ["service"],
      factory: async () => {
        return this.findMany({
          where: [['tenant_id', tenantId]]
        });
      }
    });
  }

  @api()
  async clearTenantCache(tenantId: number) {
    const tenantCache = Sonamu.cache.namespace(`tenant:${tenantId}`);

    // Delete only specific tenant's cache
    await tenantCache.clear();
  }
}

Per-Feature Isolation

// Authentication related
const authCache = Sonamu.cache.namespace("auth");
await authCache.set({ key: `session:${sessionId}`, value: session, ttl: "1h" });

// Statistics related
const statsCache = Sonamu.cache.namespace("stats");
await statsCache.set({ key: "daily", value: stats, ttl: "1d" });

// API rate limiting
const rateLimitCache = Sonamu.cache.namespace("ratelimit");
await rateLimitCache.set({ key: `user:${userId}`, value: count, ttl: "1m" });

Cache Patterns

1. Cache-Aside (Lazy Loading)

Most basic pattern: Query and cache when needed

async getData(id: number) {
  // 1. Check cache
  const cached = await Sonamu.cache.get({ key: `data:${id}` });
  if (cached) return cached;

  // 2. DB query
  const data = await this.findOne(['id', id]);

  // 3. Cache
  await Sonamu.cache.set({
    key: `data:${id}`,
    value: data,
    ttl: "10m"
  });

  return data;
}

@cache decorator automatically implements this pattern

2. Write-Through

Update cache on write: Update cache simultaneously with data change

@api()
async update(id: number, data: Partial<DataSave>) {
  // 1. DB update
  const result = await this.updateOne(['id', id], data);

  // 2. Cache refresh
  await Sonamu.cache.set({
    key: `data:${id}`,
    value: result,
    ttl: "10m",
    tags: ["data"]
  });

  return result;
}

Advantage: Cache always up-to-date Disadvantage: Write performance degradation

3. Write-Behind (Write-Back)

Write to cache first: Update cache then asynchronously save to DB

@api()
async updateCounter(id: number) {
  const key = `counter:${id}`;

  // 1. Increment cache
  let count = await Sonamu.cache.get({ key });
  count = (count ?? 0) + 1;
  await Sonamu.cache.set({ key, value: count, ttl: "10m" });

  // 2. Async DB save (queue it)
  await this.queueCounterUpdate(id, count);

  return count;
}

Advantage: Fast response Disadvantage: Complex implementation, potential data loss

4. Refresh-Ahead

Pre-refresh before expiration: Prepare fresh data before TTL ends

@cache({
  ttl: "10m",
  grace: "1h"  // Grace is similar to Refresh-Ahead
})
async getData() {
  return await this.expensiveQuery();
}

Grace period serves this role

Combined Strategies

TTL by Layer

class ProductModelClass extends BaseModelClass {
  // L1 (memory): Short TTL
  // L2 (Redis): Long TTL
  @cache({
    ttl: "5m",      // L1 is 5 minutes
    // L2 automatically retained longer
    tags: ["product"]
  })
  @api()
  async findById(id: number) {
    return this.findOne(['id', id]);
  }
}

Strategies by Priority

Critical Data
Important Data
Nice-to-have Data

Data that must always be accurate

// Payment info (no caching or short TTL)
@cache({ ttl: "30s", grace: false })
async getPaymentInfo(id: number) {
  return this.findOne(['id', id]);
}

// Inventory info (short TTL + immediate invalidation)
@cache({ ttl: "1m", tags: ["inventory"] })
async getInventory(productId: number) {
  return this.getStock(productId);
}

Accuracy important but slight delay acceptable

// User profile (moderate TTL)
@cache({ ttl: "10m", tags: ["user"] })
async getUserProfile(id: number) {
  return this.findOne(['id', id]);
}

// Order status (short TTL + Grace)
@cache({ ttl: "2m", grace: "10m", tags: ["order"] })
async getOrderStatus(id: number) {
  return this.findOne(['id', id]);
}

Data that can be stale

// Statistics (long TTL + long Grace)
@cache({ ttl: "1h", grace: "6h", tags: ["stats"] })
async getStatistics() {
  return this.calculateStats();
}

// Popular posts (long TTL)
@cache({ ttl: "6h", tags: ["popular"] })
async getPopularPosts() {
  return this.findPopular();
}

Time-Based Strategy

class DynamicCacheModelClass extends BaseModelClass {
  @api()
  async getData() {
    const hour = new Date().getHours();

    // Peak hours (9 AM ~ 6 PM): Short TTL
    const ttl = (hour >= 9 && hour <= 18) ? "5m" : "30m";

    return Sonamu.cache.getOrSet({
      key: "dynamic-data",
      ttl,
      tags: ["data"],
      factory: async () => {
        return this.expensiveQuery();
      }
    });
  }
}

Performance Optimization

Cache Warming

Pre-populate cache on server start.

class CacheWarmerModelClass extends BaseModelClass {
  async warmupCache() {
    console.log("Warming up cache...");

    // Pre-cache frequently used data
    await Promise.all([
      this.getCategoryTree(),  // @cache applied
      this.getPopularProducts(),
      this.getConfig(),
    ]);

    console.log("Cache warmed up!");
  }

  @cache({ ttl: "1d", tags: ["category"] })
  async getCategoryTree() {
    return this.buildTree();
  }

  @cache({ ttl: "1h", tags: ["popular"] })
  async getPopularProducts() {
    return this.findPopular();
  }
}

// Run on server start
await cacheWarmerModel.warmupCache();

Batch Caching

Cache multiple items at once.

class BatchCacheModelClass extends BaseModelClass {
  async batchCache(ids: number[]) {
    // 1. Find uncached IDs
    const cacheKeys = ids.map(id => `item:${id}`);
    const cached = await Promise.all(
      cacheKeys.map(key => Sonamu.cache.get({ key }))
    );

    const uncachedIds = ids.filter((_, i) => !cached[i]);

    if (uncachedIds.length === 0) {
      return cached.filter(Boolean);
    }

    // 2. Query from DB
    const items = await this.findMany({
      where: [['id', 'in', uncachedIds]]
    });

    // 3. Batch cache
    await Promise.all(
      items.rows.map(item =>
        Sonamu.cache.set({
          key: `item:${item.id}`,
          value: item,
          ttl: "10m"
        })
      )
    );

    // 4. Return results
    return [...cached.filter(Boolean), ...items.rows];
  }
}

Conditional Caching

Decide whether to cache based on conditions.

class ConditionalCacheModelClass extends BaseModelClass {
  @api()
  async getData(id: number, useCache: boolean = true) {
    const cacheKey = `data:${id}`;

    // Choose whether to use cache
    if (!useCache) {
      return this.findOne(['id', id]);
    }

    return Sonamu.cache.getOrSet({
      key: cacheKey,
      ttl: "10m",
      factory: async () => {
        return this.findOne(['id', id]);
      }
    });
  }
}

Cautions

Cache Strategy Selection Cautions:

TTL Too Long: Serves stale data

// ❌ Frequently changing but TTL too long
@cache({ ttl: "1d" })
async getLiveStock() { ... }

// ✅ Appropriate TTL
@cache({ ttl: "1m" })
async getLiveStock() { ... }

TTL Too Short: Reduced cache effectiveness

// ❌ Rarely changes but TTL too short
@cache({ ttl: "10s" })
async getConfig() { ... }

// ✅ Appropriate TTL
@cache({ ttl: "1d", tags: ["config"] })
async getConfig() { ... }

Grace Overuse: Grace unnecessary for fast queries

// ❌ Unnecessary Grace
@cache({ ttl: "5m", grace: "1h" })
async getUser(id: number) {
  return this.findOne(['id', id]);  // Fast query
}

// ✅ Sufficient without Grace
@cache({ ttl: "5m" })
async getUser(id: number) { ... }

Namespace Overuse: Too many namespaces are hard to manage

// ❌ Excessive namespace
const cache1 = Sonamu.cache.namespace("a:b:c:d:e");

// ✅ Appropriate namespace
const userCache = Sonamu.cache.namespace(`user:${userId}`);

Strategy Summary Table

Data Type	TTL	Grace	Tags	Invalidation
Static Config	`forever`	❌	`["config"]`	Manual
User Profile	`1h`	❌	`["user"]`	On change
Product Info	`30m`	`1h`	`["product"]`	On change
Post List	`5m`	`30m`	`["post", "list"]`	On create/update
Real-time Stats	`1m`	`10m`	`["stats"]`	TTL dependent
Aggregated Dashboard	`10m`	`2h`	`["dashboard"]`	TTL dependent

Next Steps

Cache Configuration

Configure Stores and Drivers

@cache Decorator

Apply caching to methods

Cache Invalidation

Tag-based cache invalidation

​TTL (Time To Live) Strategy

​What is TTL?

​TTL Settings by Data Characteristics

​TTL Units

​Grace Period (Stale-While-Revalidate)

​What is Grace?

​How It Works

​When to Use Grace

Use Grace ✅

Grace Unnecessary ❌

​Grace Practical Examples

​Grace vs Long TTL

​Namespace Strategy

​What is Namespace?

​Per-User Isolation

​Multi-Tenant

​Per-Feature Isolation

​Cache Patterns

​1. Cache-Aside (Lazy Loading)

​2. Write-Through

​3. Write-Behind (Write-Back)

​4. Refresh-Ahead

​Combined Strategies

​TTL by Layer

​Strategies by Priority

​Time-Based Strategy

​Performance Optimization

​Cache Warming

​Batch Caching

​Conditional Caching

​Cautions

​Strategy Summary Table

​Next Steps