Summary

# ===== Data types — pick by access pattern =====
# string  GET, SET, INCR           counters, cache, locks, JSON blobs
# list    LPUSH, RPOP, LRANGE       queues, activity feed
# hash    HSET, HGETALL             per-entity bag (session)
# set     SADD, SISMEMBER, SUNION    unique members, tags
# zset    ZADD, ZRANGEBYSCORE        leaderboards, schedules, time series
# stream  XADD, XREADGROUP           durable queues with consumer groups
# hll     PFADD, PFCOUNT             approx unique counts
# geo     GEOADD, GEOSEARCH          proximity queries
# bitmap  SETBIT                     boolean flags per id

# ===== TTL discipline =====
# EVERY user-facing key should have an explicit TTL
SET key value EX 60
EXPIRE key 60
PERSIST key   # remove TTL
TTL key

# ===== Patterns by intent =====
# CACHE (read-through)
#   GET key -> on miss, fill from DB, SET key val EX 300

# RATE LIMIT (fixed window)
#   INCR rl:ip:bucket + EXPIRE 60 -> reject when > N

# IDEMPOTENCY KEY
#   SET idem:$key 1 NX EX 86400 -> reject duplicate

# DISTRIBUTED LOCK
#   SET lock:res token NX PX 5000 -> release via Lua CAS

# JOB QUEUE
#   Streams + Consumer Groups (XADD + XREADGROUP + XACK)

# SESSION STORE
#   HSET sess:$id field val + EXPIRE $id 1209600

# LEADERBOARD
#   ZADD lb:weekly score user; ZREVRANGE lb:weekly 0 9

# ===== Persistence =====
# RDB   snapshot at intervals (save 900 1)        fast restore
# AOF   append every write                         durable but slower
# Both  recommended in production

# ===== Replication + HA =====
# Primary + N replicas, async by default
# Redis Sentinel for HA failover
# Redis Cluster for shards across nodes (hash slot routing)

# ===== Security =====
# - TLS for all connections
# - ACLs per app (+@read +@write -flushall)
# - rename-command CONFIG ''; rename-command FLUSHALL ''
# - Bind to private IPs only
# - require_pass strong-secret

# ===== Memory hygiene =====
maxmemory 4gb
maxmemory-policy allkeys-lru
MEMORY USAGE key       # inspect single key memory
redis-cli --bigkeys     # find the largest keys
redis-cli --hotkeys     # find the most-accessed

# ===== Observability =====
INFO memory / clients / stats / replication
SLOWLOG GET 10
LATENCY HISTORY event
CLIENT LIST
# NEVER MONITOR for long in prod (serialises traffic)

# ===== Prometheus exporter =====
# docker run -d -p 9121:9121 oliver006/redis_exporter --redis.addr=redis://prod.redis
# Alerts that pay rent:
# - hit rate < 90%
# - evicted_keys > 0
# - replication_lag_seconds > 5
# - blocked_clients > 0
# - memory used > 90% of maxmemory

# ===== Decision matrix =====
# - In-memory cache              Redis OR Memcached
# - Durable jobs                  Streams (NOT pubsub)
# - Pub/sub for real-time         Streams or PUBSUB (fire and forget)
# - Distributed lock              SET NX PX + Lua release
# - Leaderboard / ranking         zset
# - Approximate cardinality       HyperLogLog
# - Proximity search              geo (or PostGIS / specialised)

# ===== Pitfalls =====
# - 'KEYS *' to enumerate keys (O(N) blocking) -> use SCAN
# - PUBSUB for durable delivery (no replay)
# - Storing 1MB blobs in strings (bytes bite)
# - Forgetting TTL -> memory grows unbounded
# - One client for many tenants without keyspace prefixes
# - Production Redis on the public internet
# - No maxmemory + eviction policy -> OOM crashes

# ===== Reach for these libraries =====
# Node:    ioredis, BullMQ (queues), Upstash (serverless)
# Python:  redis-py, arq (queues), celery (heavier)
# Go:      github.com/redis/go-redis
# Ruby:    redis-rb + Sidekiq (queues)

# Next: RediSearch, Redis JSON, time-series, vector search with HNSW.