Cache strategy

/cache-strategy

Implement a permanent cache-first strategy: read from cache, fall back to DB on miss, write to cache, invalidate only on data mutation. No expiry timers — the cache stays valid until the underlying data changes.

Philosophy

The goal is one DB read per data item, ever (until it changes):

READ:   cache hit? → return cached value
        cache miss? → query DB → write to cache → return value

WRITE:  update DB → invalidate (or update) cache entry → done

This is the Play Framework model: no TTL, no polling, no background refresh. The cache is always warm and always correct because it is invalidated precisely when the source changes.

Why permanent over TTL?

• TTL caches trade stale reads for simplicity — every expiry is a guaranteed DB hit even if nothing changed
• Permanent caches with invalidation give you zero unnecessary DB hits and zero stale reads
• TTL is only appropriate when you can't know when data changes (third-party APIs, etc.)

Step 1: Detect cache infrastructure

# Check for Redis
grep -rn "redis\|Redis\|REDIS_URL" --include="*.json" --include="*.env*" --include="*.yml" --include="*.yaml" --include="*.rb" --include="*.py" --include="*.js" --include="*.ts" . 2>/dev/null | grep -v node_modules | head -10

# Check for Memcached
grep -rn "memcached\|Memcached\|MEMCACHE" --include="*.json" --include="*.env*" --include="*.yml" . 2>/dev/null | grep -v node_modules | head -5

# Check for in-process cache (node-cache, lru-cache, etc.)
grep -rn "lru-cache\|node-cache\|memory-cache\|caffeine\|guava.*cache" --include="*.json" --include="*.js" --include="*.ts" --include="*.java" . 2>/dev/null | grep -v node_modules | head -5

Report what's available. If no cache layer exists, recommend Redis as the default and offer to add the client library.

Step 2: Identify cacheable DB reads

Scan for DB query patterns and classify by cacheability:

High-value cache candidates (read often, change rarely):

• User profiles, settings, preferences
• Product/item details
• Configuration tables
• Reference data (categories, tags, enums stored in DB)
• Aggregates that are expensive to compute (counts, sums)

Poor cache candidates (skip these):

• Real-time data (live inventory counts, chat messages)
• Per-session data
• Data that changes on every write (e.g., counters that increment on every request)

# Find the most-called query patterns
grep -rn "\.find\b\|\.findById\|\.findOne\|\.where\|SELECT" \
  --include="*.rb" --include="*.py" --include="*.js" --include="*.ts" \
  . 2>/dev/null | grep -v node_modules | grep -v test | grep -v spec | head -30

For each candidate, note:

• Cache key pattern (e.g., user:{id}, product:{slug})
• Where it's written/updated (to know where to place invalidation)

Step 3: Implement cache-first reads

For each cacheable read, apply this pattern:

Redis (Node.js / TypeScript)

async function getUser(id: string) {
  const cacheKey = `user:${id}`;
  const cached = await redis.get(cacheKey);
  if (cached) return JSON.parse(cached);

  const user = await db.users.findById(id);
  await redis.set(cacheKey, JSON.stringify(user)); // no TTL — permanent until invalidated
  return user;
}

Redis (Ruby / Rails)

def find_user(id)
  Rails.cache.fetch("user:#{id}") do
    User.find(id)  # only called on cache miss
  end
end

Redis (Python)

def get_user(user_id: int):
    cache_key = f"user:{user_id}"
    cached = redis_client.get(cache_key)
    if cached:
        return json.loads(cached)
    user = db.session.get(User, user_id)
    redis_client.set(cache_key, json.dumps(user.to_dict()))  # no expiry
    return user

Play Framework (Scala)

def getUser(id: Long): Future[User] = {
  cache.getOrElseUpdate(s"user:$id") {
    userRepository.findById(id)
  }
}

Step 4: Implement cache invalidation on writes

For every place that mutates cached data, add invalidation immediately after the DB write succeeds:

Node.js / TypeScript

async function updateUser(id: string, data: Partial<User>) {
  const user = await db.users.update(id, data);
  await redis.del(`user:${id}`); // invalidate on change
  return user;
}

Ruby / Rails

def update_user(id, attrs)
  user = User.find(id).tap { |u| u.update!(attrs) }
  Rails.cache.delete("user:#{id}")
  user
end

Important: invalidate on delete too:

async function deleteUser(id: string) {
  await db.users.delete(id);
  await redis.del(`user:${id}`);
}

For lists/collections, invalidate the collection key when any item in it changes:

await redis.del(`user:${id}`);       // the specific item
await redis.del(`users:list`);       // any cached list that included this item

Step 5: Cache key conventions

Use a consistent key naming scheme across the codebase:

<entity>:<identifier>          → user:123, product:slug-name
<entity>:list:<filter>         → products:list:category-5
<entity>:<id>:<relation>       → user:123:orders
aggregate:<entity>:<filter>    → aggregate:orders:user-123-total

Document the convention in a comment at the top of your cache module:

// Cache key conventions:
// user:{id}                — single user record
// users:list               — paginated user list (invalidate on any user change)
// user:{id}:orders         — orders belonging to a user

Step 6: Cache warming (optional)

For critical data that must never have a cold-start miss, add a warm-up step on app boot:

async function warmCache() {
  const activeUsers = await db.users.findActiveUsers();
  await Promise.all(
    activeUsers.map(u => redis.set(`user:${u.id}`, JSON.stringify(u)))
  );
}

Step 7: Summary

Report what was implemented:

## Cache Strategy Applied

### Cache layer: Redis (permanent, no TTL)

### Reads cached (X total)
- user:{id} — UserService.getUser()
- product:{slug} — ProductService.findBySlug()
- ...

### Invalidation points added (Y total)
- user:{id} — UserService.updateUser(), UserService.deleteUser()
- product:{slug} — ProductService.updateProduct()
- ...

### Estimated DB load reduction
Before: ~X DB queries/min
After:  ~Y DB queries/min (first-read only, then cache)

Note: for data that comes from third-party APIs or sources where you can't intercept writes, use a long TTL (hours, not minutes) as a fallback — but flag these separately so they can be revisited if an invalidation webhook becomes available.