redis-patterns

name: redis-patterns description: Redis distributed-cache, rate-limiting, distributed-lock, pub/sub and Streams patterns for production backends. Use when adding a distributed cache (cache-aside/write-through, invalidation, TTL, stampede), rate limiting/throttling, cross-worker coordination, or a durable event queue. Out-of-process — not an in-memory app cache.

Redis Patterns

Quick reference for Redis in production backends: distributed cache, rate limiting, distributed locks, and messaging (Pub/Sub vs Streams).

How It Works

Redis is an in-memory data-structure store supporting strings, hashes, lists, sets, sorted sets, streams, and more. Individual Redis commands are atomic on a single instance; multi-step workflows need 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.

Lua scripts below run server-side via EVAL/register_script — that is Redis-native scripting, not client-side eval of untrusted input. It is the standard way to make a multi-step operation atomic.

When to Activate

• Adding a distributed cache shared across processes/instances
• Implementing rate limiting or throttling
• Building distributed locks or cross-worker coordination
• Setting up session or token storage
• Using Pub/Sub or Redis Streams for messaging
• Configuring Redis in production (pooling, eviction, clustering, HA)

Data Structure Cheat Sheet

Distributed Cache Patterns

Cache-Aside (Lazy Loading)

import redis
import json

r = redis.Redis(host='localhost', port=6379, decode_responses=True)

def get_product(product_id: int):
    cache_key = f"product:{product_id}"
    cached = r.get(cache_key)

    if cached:
        return json.loads(cached)

    product = db.query("SELECT * FROM products WHERE id = %s", product_id)
    r.setex(cache_key, 3600, json.dumps(product))  # TTL: 1 hour
    return product

Write-Through Cache

def update_product(product_id: int, data: dict):
    # Write to DB first
    db.execute("UPDATE products SET ... WHERE id = %s", product_id)

    # Immediately update cache
    cache_key = f"product:{product_id}"
    r.setex(cache_key, 3600, json.dumps(data))

Cache Invalidation

# Tag-based invalidation — group related keys under a set
def cache_product(product_id: int, category_id: int, data: dict):
    key = f"product:{product_id}"
    tag = f"tag:category:{category_id}"
    pipe = r.pipeline(transaction=True)
    pipe.setex(key, 3600, json.dumps(data))
    pipe.sadd(tag, key)
    pipe.expire(tag, 3600)
    pipe.execute()

def invalidate_category(category_id: int):
    tag = f"tag:category:{category_id}"
    keys = r.smembers(tag)
    if keys:
        r.delete(*keys)
    r.delete(tag)

Cache Miss Stampede Prevention

Thundering herd on cold start — many requests miss the same key at once and all hit the origin. Gate the recompute behind a lock.

import threading

_locks: dict[str, threading.Lock] = {}
_locks_mutex = threading.Lock()

def get_with_lock(key: str, fetch_fn, ttl: int = 300):
    cached = r.get(key)
    if cached:
        return json.loads(cached)

    with _locks_mutex:
        if key not in _locks:
            _locks[key] = threading.Lock()
        lock = _locks[key]
    with lock:
        cached = r.get(key)  # Re-check after acquiring lock
        if cached:
            return json.loads(cached)
        value = fetch_fn()
        r.setex(key, ttl, json.dumps(value))
        return value

For multi-process deployments, replace the in-process lock with acquire_lock/release_lock from the Distributed Locks section below. Alternatively, use probabilistic early expiry (refresh slightly before TTL with a random jitter) to spread recomputation.

Session / Token Storage

import time
import uuid

def create_session(user_id: int, ttl: int = 86400) -> str:
    session_id = str(uuid.uuid4())
    key = f"session:{session_id}"
    pipe = r.pipeline(transaction=True)
    pipe.hset(key, mapping={
        "user_id": user_id,
        "created_at": int(time.time()),
    })
    pipe.expire(key, ttl)
    pipe.execute()
    return session_id

def get_session(session_id: str) -> dict | None:
    data = r.hgetall(f"session:{session_id}")
    return data if data else None

def delete_session(session_id: str):
    r.delete(f"session:{session_id}")

TTL Strategy

Always set a TTL. Keys without a TTL accumulate indefinitely and cause memory pressure.

Eviction Policies

Set via redis.conf: maxmemory-policy allkeys-lru

Rate Limiting / Throttling

Fixed Window (Simple)

def is_rate_limited(user_id: int, limit: int = 100, window: int = 60) -> bool:
    key = f"ratelimit:{user_id}:{int(time.time()) // window}"
    pipe = r.pipeline(transaction=True)
    pipe.incr(key)
    pipe.expire(key, window)
    count, _ = pipe.execute()
    return count > limit

Sliding Window (Lua — Atomic)

-- sliding_window.lua
local key = KEYS[1]
local now = tonumber(ARGV[1])
local window = tonumber(ARGV[2])
local limit = tonumber(ARGV[3])

redis.call('ZREMRANGEBYSCORE', key, 0, now - window)
local count = redis.call('ZCARD', key)

if count < limit then
    -- Use unique member (now + sequence) to avoid collisions within the same millisecond
    local seq_key = key .. ':seq'
    local seq = redis.call('INCR', seq_key)
    redis.call('EXPIRE', seq_key, math.ceil(window / 1000))
    redis.call('ZADD', key, now, now .. '-' .. seq)
    redis.call('EXPIRE', key, math.ceil(window / 1000))
    return 1
end
return 0

sliding_window = r.register_script(open('sliding_window.lua').read())

def allow_request(user_id: int) -> bool:
    key = f"ratelimit:sliding:{user_id}"
    now = int(time.time() * 1000)
    return bool(sliding_window(keys=[key], args=[now, 60000, 100]))

Fixed-window is cheap but allows bursts at window edges (2× limit across a boundary). Sliding-window via sorted set is accurate per-user throttling. Surface limits to clients with X-RateLimit-* headers and Retry-After on 429.

Distributed Locks

Single Node — SET NX PX

import uuid

def acquire_lock(resource: str, ttl_ms: int = 5000) -> str | None:
    lock_key = f"lock:{resource}"
    token = str(uuid.uuid4())
    acquired = r.set(lock_key, token, px=ttl_ms, nx=True)
    return token if acquired else None

def release_lock(resource: str, token: str) -> bool:
    # Compare-and-delete in Lua so we never release a lock we no longer own
    release_script = """
    if redis.call('get', KEYS[1]) == ARGV[1] then
        return redis.call('del', KEYS[1])
    else
        return 0
    end
    """
    result = r.eval(release_script, 1, f"lock:{resource}", token)
    return bool(result)

# Usage
token = acquire_lock("order:payment:123")
if token:
    try:
        process_payment()
    finally:
        release_lock("order:payment:123", token)

Always set a TTL on the lock (no orphaned locks if the holder crashes) and release with compare-and-delete (never delete a lock you no longer own). For multi-node setups use the redlock-py library implementing the full Redlock algorithm.

Messaging: Pub/Sub vs Streams

Pub/Sub (Fire-and-Forget)

# Publisher
def publish_event(channel: str, payload: dict):
    r.publish(channel, json.dumps(payload))

# Subscriber (blocking — run in separate thread/process)
def subscribe_events(channel: str):
    pubsub = r.pubsub()
    pubsub.subscribe(channel)
    for message in pubsub.listen():
        if message['type'] == 'message':
            handle(json.loads(message['data']))

Redis Streams (Durable Queue, Consumer Groups)

# Producer
def emit(stream: str, event: dict):
    r.xadd(stream, event, maxlen=10000)  # Cap stream length

# Consumer group — guarantees at-least-once delivery
try:
    r.xgroup_create('events:orders', 'processor', id='0', mkstream=True)
except Exception:
    pass  # Group already exists

def consume(stream: str, group: str, consumer: str):
    while True:
        messages = r.xreadgroup(group, consumer, {stream: '>'}, count=10, block=2000)
        for _, entries in (messages or []):
            for msg_id, data in entries:
                process(data)             # Make this idempotent — delivery is at-least-once
                r.xack(stream, group, msg_id)

Prefer Streams over Pub/Sub when you need delivery guarantees, consumer groups, or replay. Pub/Sub drops messages for any subscriber that is offline. With Streams, periodically XAUTOCLAIM/XCLAIM pending entries from dead consumers and route entries that exceed a delivery-count threshold to a dead-letter stream.

Key Design

# Pattern: resource:id:field
user:123:profile
order:456:status

# Pattern: namespace:resource:id
myapp:session:abc123
myapp:ratelimit:user:123

# Pattern: resource:date (time-bound keys)
stats:pageviews:2024-01-01

Connection Management

Connection Pooling

from redis import ConnectionPool, Redis

pool = ConnectionPool(
    host='localhost',
    port=6379,
    db=0,
    max_connections=20,
    decode_responses=True,
    socket_connect_timeout=2,
    socket_timeout=2,
)

r = Redis(connection_pool=pool)

Cluster Mode

from redis.cluster import RedisCluster

r = RedisCluster(
    startup_nodes=[{"host": "redis-1", "port": 6379}],
    decode_responses=True,
    skip_full_coverage_check=True,
)

Sentinel (High Availability)

from redis.sentinel import Sentinel

sentinel = Sentinel(
    [('sentinel-1', 26379), ('sentinel-2', 26379)],
    socket_timeout=0.5,
)
master = sentinel.master_for('mymaster', decode_responses=True)
replica = sentinel.slave_for('mymaster', decode_responses=True)

Anti-Patterns

Quick Reference