redis.mdc

---
description: "Redis: data structures, caching patterns, pub/sub"
globs: ["*.ts", "*.py", "*.conf"]
alwaysApply: true
---

# Redis Cursor Rules

You are an expert Redis developer. Follow these rules:

## Data Structures
- Strings for simple key-value and counters (INCR/DECR are atomic)
- Hashes for objects — HSET user:123 name "Alice" email "alice@..."
- Sorted sets for leaderboards, rate limiters, priority queues
- Lists for queues (LPUSH/RPOP) and activity feeds
- Sets for unique collections and intersection/union operations
- Streams for event sourcing and message queues

## Key Design
- Namespace keys with colons: user:123:session, post:456:likes
- Keep keys short but readable — abbreviate namespaces in high-volume scenarios
- Set TTL on every cache key — no key should live forever without reason
- Use SCAN over KEYS — KEYS blocks the single-threaded server

## Caching Patterns
- Cache-aside: app checks cache, falls back to DB, writes to cache
- Write-through for data that must stay consistent
- TTL-based invalidation for most use cases. Event-based for critical consistency
- Serialize as JSON or MessagePack — avoid language-specific formats
- Cache stampede prevention: use locks or stale-while-revalidate

## Performance
- Pipeline multiple commands to reduce round trips
- Lua scripts for atomic multi-step operations (EVAL/EVALSHA)
- Avoid large values (>100KB) — split into smaller keys
- Use OBJECT ENCODING to verify memory-efficient encodings
- Monitor with INFO stats, SLOWLOG, and CLIENT LIST

## Pub/Sub & Streams
- Pub/Sub for fire-and-forget broadcasting — messages are lost if no subscriber
- Streams for persistent messaging with consumer groups
- XACK processed messages. Set MAXLEN to cap stream size
- Use consumer groups for distributed work distribution

## Operations
- Redis Sentinel for high availability, Redis Cluster for horizontal scaling
- RDB snapshots for backups. AOF for durability
- maxmemory-policy: allkeys-lru for caches, noeviction for primary data
- Monitor memory with INFO memory — watch fragmentation ratio