Redis Patterns

Quick Guide: Use Redis as an in-memory data store for caching, session management, rate limiting, pub/sub messaging, and job queues. Use ioredis (v5.x) as the primary client for its superior TypeScript support, Cluster/Sentinel integration, auto-pipelining, and Lua scripting. Use node-redis (v5.x) only when you need Redis Stack modules (JSON, Search, TimeSeries). Always set maxRetriesPerRequest: null for BullMQ workers, use separate connections for Pub/Sub subscribers, and define Lua scripts via defineCommand for atomic multi-step operations.

<critical_requirements>

CRITICAL: Before Using This Skill

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST use a SEPARATE Redis connection for Pub/Sub subscribers -- a subscribed connection enters a special mode and cannot execute other commands)

(You MUST set maxRetriesPerRequest: null on any ioredis connection passed to BullMQ -- BullMQ requires infinite retries and will throw if this is not set)

(You MUST use Lua scripts (defineCommand or eval) for any operation requiring atomicity across multiple Redis commands -- separate commands are NOT atomic even in a pipeline)

(You MUST handle the error event on every Redis client instance -- unhandled errors crash the Node.js process)

</critical_requirements>

Examples

• Core Patterns -- ioredis/node-redis connection, error handling, reconnection, cluster, sentinel, pipelining, transactions
• Caching Patterns -- Cache-aside, write-through, invalidation, stampede prevention, multi-key pipeline
• Data Structures -- Strings, hashes, lists, sets, sorted sets with typed helpers
• Sessions -- Express connect-redis (node-redis required for v9+), Hono manual middleware
• Pub/Sub -- Publish/subscribe, event broadcasting, pattern subscriptions
• Rate Limiting -- Sliding window (Lua), token bucket (Lua), middleware integration
• Queues & Locks -- BullMQ job queues, Redis Streams with consumer groups, distributed locks

Additional resources:

• reference.md -- Command cheat sheet, connection options, anti-patterns, production checklist

Auto-detection: Redis, ioredis, node-redis, createClient, RedisStore, BullMQ, Queue, Worker, pub/sub, MULTI, EXEC, pipeline, Lua script, defineCommand, xadd, xread, cache-aside, rate limit, session store, connect-redis, Redis.Cluster, Sentinel

When to use:

• Caching database queries or API responses (cache-aside, write-through)
• Session storage for Express/Hono/Fastify applications
• Distributed rate limiting (sliding window, token bucket)
• Real-time messaging with Pub/Sub
• Background job processing with BullMQ queues
• Leaderboards, counters, and real-time analytics with sorted sets
• Distributed locks and atomic operations with Lua scripts

Key patterns covered:

• ioredis connection setup, configuration, and error handling
• Data structures (strings, hashes, lists, sets, sorted sets, streams)
• Cache-aside and write-through caching with TTL management
• Session storage with connect-redis
• Rate limiting with Lua scripts (sliding window, token bucket)
• Pub/Sub messaging with separate connections
• Redis Streams for persistent message queues
• BullMQ for job queues with retries and scheduling
• Pipelining and transactions (MULTI/EXEC)
• Lua scripting for atomic operations
• Cluster mode and Sentinel for high availability

When NOT to use:

• Primary database for relational data (use your relational database)
• Document storage with complex queries (use a document database)
• Large binary file storage (use S3/object storage)
• Data that must survive total memory loss without persistence configured

Philosophy

Redis is an in-memory data store used as a cache, message broker, and streaming engine. The core principle: use Redis for fast, ephemeral, or real-time data -- not as a primary database.

Core principles:

1. Cache, don't store -- Redis complements your primary database. Cache frequently accessed data, but always have a source of truth elsewhere.
2. Atomic operations -- Use Lua scripts or MULTI/EXEC for operations spanning multiple keys. Individual Redis commands are atomic, but sequences are not.
3. Separate concerns -- Use different Redis databases (or key prefixes) for caching, sessions, and queues. Use separate connections for Pub/Sub.
4. Set TTLs on everything -- Memory is finite. Every cached key should expire. Use EX (seconds) or PX (milliseconds) on SET commands.
5. Fail gracefully -- Redis is a cache, not a database. If Redis is down, the application should degrade gracefully (bypass cache, use database directly).

Core Patterns

Pattern 1: ioredis Connection Setup

Configure ioredis with proper error handling and reconnection strategy. See examples/core.md for full examples including node-redis and cluster configuration.

// ✅ Good Example - Proper ioredis setup with error handling
import Redis from "ioredis";

const RETRY_DELAY_BASE_MS = 50;
const RETRY_DELAY_MAX_MS = 2000;

function createRedisClient(): Redis {
  const url = process.env.REDIS_URL;
  if (!url) {
    throw new Error("REDIS_URL environment variable is required");
  }

  const client = new Redis(url, {
    maxRetriesPerRequest: 3,
    retryStrategy(times) {
      return Math.min(times * RETRY_DELAY_BASE_MS, RETRY_DELAY_MAX_MS);
    },
    lazyConnect: true,
  });

  client.on("error", (err) => {
    console.error("Redis connection error:", err.message);
  });

  return client;
}

export { createRedisClient };

Why good: Environment variable validation, named constants for retry delays, lazyConnect prevents connection before ready, error event handler prevents process crash

// ❌ Bad Example - No error handling, hardcoded config
import Redis from "ioredis";
const redis = new Redis("redis://localhost:6379");
// No error handler, no retry strategy, hardcoded URL

Why bad: Missing error event handler crashes Node.js process on connection failure, hardcoded URL prevents environment-specific configuration

Pattern 2: Cache-Aside

Check cache first, fall back to database on miss, populate cache. See examples/caching.md for write-through, invalidation, stampede prevention, and multi-key cache patterns.

// ✅ Good Example - Generic cache-aside helper
const CACHE_TTL_SECONDS = 300;

async function cacheAside<T>(
  redis: Redis,
  key: string,
  fetcher: () => Promise<T>,
  ttlSeconds: number = CACHE_TTL_SECONDS,
): Promise<T> {
  const cached = await redis.get(key);
  if (cached !== null) {
    return JSON.parse(cached) as T;
  }

  const data = await fetcher();

  // Fire-and-forget cache write
  redis.set(key, JSON.stringify(data), "EX", ttlSeconds).catch((err) => {
    console.error(`Cache write failed for ${key}:`, err.message);
  });

  return data;
}

Why good: Generic cacheAside<T> works with any data type, fire-and-forget cache write prevents cache failure from blocking response, configurable TTL

Pattern 3: Pipelining and Transactions

Batch commands for network efficiency (pipeline) or atomicity (MULTI/EXEC). See examples/core.md for full examples.

// ✅ Pipeline - batch for network efficiency (NOT atomic)
const pipeline = redis.pipeline();
pipeline.hset(key, { name: user.name, email: user.email });
pipeline.expire(key, TTL_SECONDS);
await pipeline.exec();

// ✅ Transaction - atomic execution with optimistic locking
await redis.watch(fromKey);
const results = await redis
  .multi()
  .decrby(fromKey, amount)
  .incrby(toKey, amount)
  .exec();
if (!results) {
  /* WATCH detected change, retry */
}

Why good: Pipeline reduces round-trips, MULTI/EXEC provides atomicity, WATCH enables optimistic locking

// ❌ Bad Example - Non-atomic balance transfer
await redis.decrby("balance:user1", 100);
await redis.incrby("balance:user2", 100);
// Crash between commands causes data loss

Why bad: Two separate commands are not atomic, crash between them causes data inconsistency

Pattern 4: Lua Scripting for Atomicity

Lua scripts execute atomically on the Redis server. See examples/rate-limiting.md for complete Lua-based rate limiters.

// ✅ Define custom atomic command
redis.defineCommand("rateLimit", {
  numberOfKeys: 1,
  lua: `
    local key = KEYS[1]
    local limit = tonumber(ARGV[1])
    local window = tonumber(ARGV[2])
    local now = tonumber(ARGV[3])
    redis.call('ZREMRANGEBYSCORE', key, 0, now - window)
    local count = redis.call('ZCARD', key)
    if count < limit then
      redis.call('ZADD', key, now, now .. '-' .. math.random(1000000))
      redis.call('EXPIRE', key, window)
      return 1
    end
    return 0
  `,
});

Why good: defineCommand uses EVALSHA internally for performance, Lua script is atomic -- no race conditions

Pattern 5: Pub/Sub Messaging

Requires separate connections for subscribing and publishing. See examples/pub-sub.md for event broadcasting system.

// ✅ Good Example - Separate connections
const publisher = new Redis(url);
const subscriber = new Redis(url); // MUST be separate

await subscriber.subscribe("notifications");
subscriber.on("message", (channel, message) => {
  handleNotification(JSON.parse(message));
});

await publisher.publish("notifications", JSON.stringify(data));

Why good: Separate connections for pub and sub (required by Redis protocol), error handlers on both

// ❌ Bad Example - Same connection
const redis = new Redis();
await redis.subscribe("channel");
await redis.set("key", "value"); // THROWS: connection is in subscriber mode

Why bad: A subscribed connection cannot execute non-pub/sub commands

Pattern 6: BullMQ Job Queues

Robust job queuing with retries, scheduling, and priorities. See examples/queues.md for complete examples with workers, events, and graceful shutdown.

// ✅ Good Example - BullMQ connection factory
function createBullMQConnection(): Redis {
  return new Redis(process.env.REDIS_URL!, {
    maxRetriesPerRequest: null, // REQUIRED for BullMQ
  });
}

const emailQueue = new Queue<EmailJobData>(QUEUE_NAME, {
  connection: createBullMQConnection(),
  defaultJobOptions: {
    attempts: MAX_RETRY_ATTEMPTS,
    backoff: { type: "exponential", delay: BACKOFF_DELAY_MS },
  },
});

Why good: maxRetriesPerRequest: null is required for BullMQ, separate connection per Queue/Worker

// ❌ Bad Example - Missing required config
const connection = new Redis(); // Missing maxRetriesPerRequest: null
const queue = new Queue("emails", { connection });
// BullMQ will throw: "maxRetriesPerRequest must be null"

Why bad: BullMQ requires infinite retries -- without null, ioredis gives up after a set number of attempts

Pattern 7: Redis Streams

Persistent, ordered message logs with consumer groups. See examples/queues.md for full producer/consumer implementation with pending message recovery.

// Producer
const entryId = await redis.xadd(
  STREAM_KEY,
  "*",
  "orderId",
  id,
  "action",
  action,
);

// Consumer group setup (idempotent)
try {
  await redis.xgroup("CREATE", STREAM_KEY, GROUP_NAME, "0", "MKSTREAM");
} catch (err) {
  if (!(err instanceof Error) || !err.message.includes("BUSYGROUP")) throw err;
}

// Consumer read + acknowledge
const results = await redis.xreadgroup(
  "GROUP",
  GROUP_NAME,
  consumerName,
  "COUNT",
  "10",
  "BLOCK",
  "5000",
  "STREAMS",
  STREAM_KEY,
  ">",
);
await redis.xack(STREAM_KEY, GROUP_NAME, messageId);

Why good: MKSTREAM creates stream if absent, BUSYGROUP handling for idempotent setup, XACK confirms processing

Performance Optimization

• Auto-pipelining -- Enable enableAutoPipelining: true to batch commands issued during the same event loop tick. Does NOT work with WATCH/MULTI or blocking commands. See examples/core.md for setup.
• Manual pipelining -- Use redis.pipeline() for explicit command batching. See examples/core.md.
• Key expiration -- Set TTLs on all cache keys. Use named constants (TTL_SHORT_SECONDS = 60, TTL_MEDIUM_SECONDS = 300, etc.). See examples/core.md.
• SCAN over KEYS -- Never use KEYS in production (blocks Redis). Use redis.scanStream({ match: "pattern", count: 100 }). See examples/core.md.
• UNLINK over DEL -- Use unlink for large keys (non-blocking deletion).

<decision_framework>

Decision Framework

Which Redis Client?

Which Redis client should I use?
├─ Need Redis Stack modules (JSON, Search, TimeSeries)? -> node-redis (v5.x)
├─ Using BullMQ for job queues? -> ioredis (BullMQ requires it)
├─ Need Cluster or Sentinel support? -> ioredis (built-in, battle-tested)
├─ Need auto-pipelining? -> ioredis (enableAutoPipelining option)
└─ General caching/sessions/pub-sub? -> ioredis (recommended default)

Which Caching Strategy?

How should I cache this data?
├─ Read-heavy, tolerates brief staleness? -> Cache-aside with TTL
├─ Needs strong consistency after writes? -> Write-through (update DB + invalidate cache)
├─ Write-heavy, can tolerate brief data loss? -> Write-behind (async cache update)
└─ Data changes rarely? -> Cache-aside with long TTL + manual invalidation

Which Data Structure?

What Redis data structure should I use?
├─ Simple key-value (cache, sessions)? -> Strings (GET/SET)
├─ Object with multiple fields? -> Hashes (HSET/HGET)
├─ Ordered ranking/leaderboard? -> Sorted Sets (ZADD/ZRANGE)
├─ Queue (FIFO/LIFO)? -> Lists (LPUSH/RPOP)
├─ Unique collection (tags, categories)? -> Sets (SADD/SMEMBERS)
├─ Persistent message log with consumers? -> Streams (XADD/XREAD)
└─ Rate limiting (sliding window)? -> Sorted Sets + Lua script

Which Messaging Pattern?

How should I implement real-time messaging?
├─ Fire-and-forget broadcast? -> Pub/Sub (no persistence)
├─ Need message persistence and replay? -> Streams with consumer groups
├─ Need reliable job processing with retries? -> BullMQ (built on Redis)
└─ Need request-reply pattern? -> Pub/Sub with correlation IDs

Atomicity Decision

Do I need atomicity across multiple commands?
├─ YES -> Are the commands on the same key?
│   ├─ YES -> Use a single atomic command (INCR, SETNX, etc.)
│   └─ NO -> Use Lua script (defineCommand)
├─ NO, but I want batching -> Use pipeline (non-atomic, single round-trip)
└─ Need optimistic locking? -> Use WATCH + MULTI/EXEC

</decision_framework>

Integration Guide

Common integration patterns:

• Database caching -- Cache query results using cache-aside pattern; invalidate cache on writes
• HTTP framework sessions -- Session storage middleware, rate limiting middleware
• BullMQ job queues -- Built on Redis (requires ioredis with maxRetriesPerRequest: null)
• WebSocket scaling -- Redis adapter for distributing WebSocket connections across servers

Replaces / Conflicts with:

• In-memory caches -- Redis provides distributed caching across multiple app instances
• Database-backed sessions -- Redis sessions are faster and reduce database load
• Simple message queues -- Redis Streams and BullMQ cover most queue use cases; use dedicated message brokers for complex routing or massive throughput

<red_flags>

RED FLAGS

High Priority Issues:

• Using the same connection for Pub/Sub subscribe and regular commands -- subscribed connections cannot execute non-pub/sub commands
• Missing maxRetriesPerRequest: null on BullMQ connections -- BullMQ throws immediately without this setting
• Using KEYS command in production -- blocks the entire Redis server while scanning all keys
• No error event handler on Redis client -- unhandled errors crash the Node.js process
• Storing large objects (> 1 MB) in Redis -- degrades performance and wastes memory; store a reference and fetch from object storage

Medium Priority Issues:

• Missing TTL on cached keys -- causes unbounded memory growth until Redis runs out of memory
• Using del with many keys instead of unlink -- del blocks Redis; unlink frees memory asynchronously
• Not using pipelining for batch operations -- each command is a separate network round-trip
• Serializing/deserializing complex objects without error handling -- malformed JSON in cache crashes on parse
• Sharing a single Redis connection across BullMQ Queue and Worker -- each needs its own connection

Common Mistakes:

• Assuming pipeline commands are atomic -- pipelines batch for network efficiency but do not provide atomicity (use MULTI/EXEC or Lua)
• Forgetting that hgetall returns an empty object {} for non-existent keys (not null) -- check Object.keys(result).length === 0
• Using MULTI/EXEC without WATCH for conditional updates -- transactions execute unconditionally unless you WATCH keys first
• Not handling null returns from GET -- cache misses return null, not undefined
• Connecting to Redis without TLS in production -- credentials sent in plaintext over the network

Gotchas & Edge Cases:

• Redis HGETALL returns all values as strings -- numbers stored with HSET come back as strings, requiring explicit parsing
• EXPIRE resets when a key is overwritten with SET -- if you SET a key that already has a TTL, the TTL is removed unless you include EX/PX in the SET command
• Pub/Sub messages are fire-and-forget -- if no subscriber is listening when a message is published, it is lost forever (use Streams for persistence)
• Redis Cluster does not support multi-key operations across different hash slots -- use {hash-tag} prefix to force related keys to the same slot
• WATCH is connection-scoped -- concurrent requests sharing a connection will interfere with each other's WATCH state
• ioredis auto-pipelining does not work with WATCH/MULTI or blocking commands (BRPOP, BLPOP, XREAD BLOCK)

</red_flags>

<critical_reminders>

CRITICAL REMINDERS

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST use a SEPARATE Redis connection for Pub/Sub subscribers -- a subscribed connection enters a special mode and cannot execute other commands)

(You MUST set maxRetriesPerRequest: null on any ioredis connection passed to BullMQ -- BullMQ requires infinite retries and will throw if this is not set)

(You MUST use Lua scripts (defineCommand or eval) for any operation requiring atomicity across multiple Redis commands -- separate commands are NOT atomic even in a pipeline)

(You MUST handle the error event on every Redis client instance -- unhandled errors crash the Node.js process)

Failure to follow these rules will cause pub/sub failures, BullMQ connection errors, race conditions, and application crashes.

</critical_reminders>