Redis

Build fast, scalable applications with Redis as a cache, message broker, session store, or real-time data engine.

Setup

Docker (quickstart)

# Redis 7 with persistence
docker run -d --name redis -p 6379:6379 \
  -v redis-data:/data \
  redis:7-alpine redis-server --appendonly yes --requirepass "your-password"

Connection

"""redis_client.py — Redis connection with connection pooling."""
import redis

# Single connection
r = redis.Redis(host="localhost", port=6379, password="your-password", db=0,
                decode_responses=True)  # Auto-decode bytes to strings

# Connection pool (recommended for production)
pool = redis.ConnectionPool(
    host="localhost", port=6379, password="your-password",
    max_connections=20,  # Match your app's concurrency
    decode_responses=True,
)
r = redis.Redis(connection_pool=pool)

# Verify connection
r.ping()  # Returns True

// redis-client.js — Node.js connection with ioredis
import Redis from 'ioredis';
const redis = new Redis({
  host: 'localhost', port: 6379, password: 'your-password',
  maxRetriesPerRequest: 3,
  retryStrategy: (times) => Math.min(times * 50, 2000),
});

Caching Patterns

Cache-Aside (most common)

"""cache_aside.py — Cache-aside pattern with automatic expiration."""
import json

def get_user(user_id: int) -> dict:
    """Fetch user from cache, falling back to database.

    Args:
        user_id: The user's ID.

    Returns:
        User dict from cache or database.
    """
    cache_key = f"user:{user_id}"
    cached = r.get(cache_key)
    if cached:
        return json.loads(cached)

    # Cache miss — fetch from database
    user = db.query("SELECT * FROM users WHERE id = %s", user_id)
    r.setex(cache_key, 3600, json.dumps(user))  # Cache for 1 hour
    return user

def update_user(user_id: int, data: dict):
    """Update user in database and invalidate cache.

    Args:
        user_id: The user's ID.
        data: Fields to update.
    """
    db.execute("UPDATE users SET ... WHERE id = %s", user_id)
    r.delete(f"user:{user_id}")  # Invalidate — next read repopulates

Write-Through Cache

"""write_through.py — Write-through: update cache and DB together."""

def save_product(product_id: str, data: dict):
    """Save product to both database and cache atomically.

    Args:
        product_id: Product identifier.
        data: Product data dict.
    """
    # Write to DB first (source of truth)
    db.execute("INSERT INTO products ... ON CONFLICT UPDATE ...", data)

    # Then update cache
    r.setex(f"product:{product_id}", 7200, json.dumps(data))  # 2h TTL

Session Storage

"""session_store.py — HTTP session storage in Redis."""
import secrets, json

SESSION_TTL = 86400  # 24 hours

def create_session(user_id: int, metadata: dict = None) -> str:
    """Create a new session and return the session token.

    Args:
        user_id: Authenticated user's ID.
        metadata: Optional session metadata (IP, user-agent, etc.).
    """
    token = secrets.token_urlsafe(32)
    session_data = {"user_id": user_id, "created_at": time.time(), **(metadata or {})}
    r.setex(f"session:{token}", SESSION_TTL, json.dumps(session_data))
    # Track active sessions per user for "log out everywhere"
    r.sadd(f"user_sessions:{user_id}", token)
    return token

def get_session(token: str) -> dict | None:
    """Validate and return session data, extending TTL on access.

    Args:
        token: Session token from cookie/header.
    """
    data = r.get(f"session:{token}")
    if not data:
        return None
    r.expire(f"session:{token}", SESSION_TTL)  # Sliding expiration
    return json.loads(data)

def destroy_all_sessions(user_id: int):
    """Invalidate all sessions for a user (password change, security breach).

    Args:
        user_id: The user whose sessions to destroy.
    """
    tokens = r.smembers(f"user_sessions:{user_id}")
    if tokens:
        r.delete(*[f"session:{t}" for t in tokens])
    r.delete(f"user_sessions:{user_id}")

Rate Limiting

Sliding Window

"""rate_limiter.py — Sliding window rate limiter using sorted sets."""
import time

def is_rate_limited(key: str, limit: int, window_seconds: int) -> bool:
    """Check if a key has exceeded its rate limit.

    Args:
        key: Identifier (e.g., IP address, API key, user ID).
        limit: Maximum requests allowed in the window.
        window_seconds: Window size in seconds.

    Returns:
        True if rate limited, False if request is allowed.
    """
    now = time.time()
    window_start = now - window_seconds
    pipe = r.pipeline()
    rk = f"ratelimit:{key}"

    pipe.zremrangebyscore(rk, 0, window_start)  # Remove expired entries
    pipe.zadd(rk, {f"{now}": now})              # Add current request
    pipe.zcard(rk)                               # Count requests in window
    pipe.expire(rk, window_seconds)              # Auto-cleanup

    results = pipe.execute()
    count = results[2]
    return count > limit

For smoother rate limiting, consider a token bucket implementation using a Lua script that tracks tokens and refill timestamps in a Redis hash.

Pub/Sub

"""pubsub.py — Real-time messaging with Redis pub/sub."""
import threading

def publish_event(channel: str, event: dict):
    """Publish an event to a channel.

    Args:
        channel: Channel name (e.g., "notifications:user:123").
        event: Event data dict — serialized to JSON.
    """
    r.publish(channel, json.dumps(event))

def subscribe_to_events(pattern: str, callback):
    """Subscribe to channels matching a pattern.

    Args:
        pattern: Glob pattern (e.g., "notifications:*").
        callback: Function called with (channel, data) for each message.
    """
    ps = r.pubsub()
    ps.psubscribe(pattern)

    def listener():
        for msg in ps.listen():
            if msg["type"] == "pmessage":
                callback(msg["channel"], json.loads(msg["data"]))

    thread = threading.Thread(target=listener, daemon=True)
    thread.start()
    return ps  # Return for cleanup: ps.punsubscribe()

Streams (persistent messaging)

Unlike pub/sub, streams persist messages and support consumer groups. Use XADD to add events, XGROUP CREATE to create consumer groups, and XREADGROUP/XACK to consume and acknowledge messages. Set maxlen on XADD to cap stream memory.

Distributed Locking

"""distributed_lock.py — Distributed lock using Redis (Redlock pattern)."""

def acquire_lock(name: str, timeout: int = 10) -> str | None:
    """Acquire a distributed lock.

    Args:
        name: Lock name (e.g., "process:invoice:12345").
        timeout: Lock expiration in seconds (prevents deadlocks).

    Returns:
        Lock token if acquired, None if already held.
    """
    token = secrets.token_urlsafe(16)
    acquired = r.set(f"lock:{name}", token, nx=True, ex=timeout)
    return token if acquired else None

# Lua script ensures atomic check-and-delete (only owner can release)
RELEASE_SCRIPT = """
if redis.call('GET', KEYS[1]) == ARGV[1] then
    return redis.call('DEL', KEYS[1])
end
return 0
"""

def release_lock(name: str, token: str) -> bool:
    """Release a lock (only if we own it).

    Args:
        name: Lock name.
        token: Token returned by acquire_lock.
    """
    return r.eval(RELEASE_SCRIPT, 1, f"lock:{name}", token) == 1

Leaderboards

Use sorted sets (ZADD, ZREVRANGE, ZREVRANK) for real-time leaderboards. ZADD sets scores, ZREVRANGE returns top N entries, and ZREVRANK gets a member's rank.

Production Configuration

Key settings: maxmemory 2gb, maxmemory-policy allkeys-lru, appendonly yes, appendfsync everysec. Always set a maxmemory limit to prevent out-of-memory crashes.

Guidelines

• Use SETEX/SET ... EX with TTLs for all cache keys -- keys without expiration leak memory
• Pipeline multiple commands when doing batch operations -- reduces round trips
• Use Lua scripts for atomic multi-step operations (check-and-set, compare-and-delete)
• Prefer Streams over pub/sub when message persistence matters -- pub/sub drops messages if no subscriber is listening
• Key naming convention: use colons as separators (user:123:profile, cache:product:456)
• Monitor memory usage with INFO memory -- Redis is in-memory, running out kills the process
• Use SCAN instead of KEYS * in production -- KEYS blocks the server on large datasets
• Connection pooling is essential -- creating a new connection per request adds 1-2ms latency
• Redis is single-threaded for commands -- one slow Lua script blocks everything