npx skills add ...
npx skills add affaan-m/ecc --skill redis-patterns
Redis data structure patterns, caching strategies, distributed locks, rate limiting, pub/sub, and connection management for production applications. Use when adding caching, a distributed lock, rate limiting, or pub/sub with Redis, or when key design needs review.
npx skills add affaan-m/ecc --skill redis-patterns
Quick reference for Redis best practices across common backend use cases.
Redis is an in-memory data structure store that supports strings, hashes, lists, sets, sorted sets, streams, and more. Individual Redis commands are atomic on a single instance; multi-step workflows require Lua scripts, MULTI/EXEC transactions, or explicit synchronization to stay atomic. Data is optionally persisted via RDB snapshots or AOF logs. Clients communicate over TCP using the RESP protocol; connection pools are essential to avoid per-request handshake overhead.
| Use Case | Structure | Example Key |
|---|---|---|
| Simple cache | String | product:123 |
| User session | Hash | session:abc |
| Leaderboard | Sorted Set | scores:weekly |
| Unique visitors | Set | visitors:2024-01-01 |
| Activity feed | List | feed:user:456 |
| Event stream | Stream | events:orders |
| Counters / rate limits | String (INCR) | ratelimit:user:123 |
| Bloom filter / HLL | HyperLogLog | hll:pageviews |
For multi-node setups use the
redlock-pylibrary which implements the full Redlock algorithm.
Prefer Streams over Pub/Sub when you need delivery guarantees, consumer groups, or replay.
| Data Type | Suggested TTL |
|---|---|
| User session | 24h (86400) |
| API response cache | 5–15 min |
| Rate limit window | Match window size |
| Short-lived tokens | 5–10 min |
| Leaderboard | 1h–24h |
| Static/reference data | 1h–1 week |
Always set a TTL. Keys without TTL accumulate indefinitely and cause memory pressure.
| Policy | Behavior | Best For |
|---|---|---|
noeviction | Error on write when full | Queues / critical data |
allkeys-lru | Evict least recently used | General cache |
volatile-lru | LRU only among keys with TTL | Mixed data store |
allkeys-lfu | Evict least frequently used | Skewed access patterns |
volatile-ttl | Evict soonest-to-expire | Prioritize long-lived data |
Set via redis.conf: maxmemory-policy allkeys-lru
| Anti-Pattern | Problem | Fix |
|---|---|---|
| Keys with no TTL | Memory grows unbounded | Always set TTL |
KEYS * in production | Blocks the server (O(N)) | Use SCAN cursor |
| Storing large blobs (>100KB) | Slow serialization, memory pressure | Store reference + fetch from object store |
| Single Redis for everything | No isolation between cache & queue | Use separate DBs or instances |
| Ignoring connection pool limits | Connection exhaustion under load | Size pool to workload |
| Not handling cache miss stampede | Thundering herd on cold start | Use locks or probabilistic early expiry |
FLUSHALL without thought | Wipes entire instance | Scope deletes by key pattern |
Note: for multi-process deployments, replace the in-process lock with
acquire_lock/release_lockfrom the Distributed Locks section above.
Add caching to a Django/Flask API endpoint:
Use cache-aside with setex and a 5-minute TTL on the response. Key on the request parameters.
Rate-limit an API by user:
Use fixed-window with pipeline(transaction=True) for low-traffic endpoints; use sliding-window Lua for accurate per-user throttling.
Coordinate a background job across workers:
Use acquire_lock with a TTL that exceeds the expected job duration. Always release in a finally block.
Fan-out notifications to multiple subscribers: Use Pub/Sub for fire-and-forget. Switch to Streams if you need guaranteed delivery or replay for late consumers.
| Pattern | When to Use |
|---|---|
| Cache-aside | Read-heavy, tolerate slight staleness |
| Write-through | Strong consistency required |
| Distributed lock | Prevent concurrent access to a resource |
| Sliding window rate limit | Accurate per-user throttling |
| Redis Streams | Durable event queue with consumer groups |
| Pub/Sub | Broadcast with no delivery guarantees needed |
| Sorted Set leaderboard | Ranked scoring, pagination |
| HyperLogLog | Approximate unique count at low memory |
postgres-patterns — relational data patternsbackend-patterns — API and service layer patternsdatabase-migrations — schema versioningdjango-patterns — Django cache framework integrationdatabase-reviewer — full database review workflow