Token bucket rate limiter with Redis and redis-py

Implement a token bucket rate limiter with Redis and redis-py

This guide shows you how to implement a distributed token bucket rate limiter using Redis and Lua scripts for atomic operations.

Overview

Rate limiting is a critical technique for controlling the rate at which operations are performed. Common use cases include:

Limiting API requests per user or IP address
Preventing abuse and protecting against denial-of-service attacks
Ensuring fair resource allocation across multiple clients
Throttling background jobs or batch operations

The token bucket algorithm is a popular rate limiting approach that allows bursts of traffic while maintaining an average rate limit over time.

How it works

The token bucket algorithm works like a bucket that holds tokens:

Initialization: The bucket starts with a maximum capacity of tokens
Refill: Tokens are added to the bucket at a constant rate (for example, 1 token per second)
Consumption: Each request consumes one token from the bucket
Decision: If tokens are available, the request is allowed; otherwise, it's denied
Capacity limit: The bucket never exceeds its maximum capacity

This approach allows for burst traffic (using accumulated tokens) while enforcing an average rate limit over time.

Why use Redis?

Redis is ideal for distributed rate limiting because:

Atomic operations: Lua scripts execute atomically, preventing race conditions
Shared state: Multiple application servers can share the same rate limit counters
High performance: In-memory operations provide microsecond latency
Automatic expiration: Keys can be set to expire automatically (though not used in this implementation)

The Lua script

The core of this implementation is a Lua script that runs atomically on the Redis server. This ensures that checking and updating the token bucket happens in a single operation, preventing race conditions in distributed environments.

Here's how the script works:

local key = KEYS[1]
local capacity = tonumber(ARGV[1])
local refill_rate = tonumber(ARGV[2])
local refill_interval = tonumber(ARGV[3])
local now = tonumber(ARGV[4])

-- Get current state or initialize
local bucket = redis.call('HMGET', key, 'tokens', 'last_refill')
local tokens = tonumber(bucket[1])
local last_refill = tonumber(bucket[2])

-- Initialize if this is the first request
if tokens == nil then
    tokens = capacity
    last_refill = now
end

-- Calculate token refill
local time_passed = now - last_refill
local refills = math.floor(time_passed / refill_interval)

if refills > 0 then
    tokens = math.min(capacity, tokens + (refills * refill_rate))
    last_refill = last_refill + (refills * refill_interval)
end

-- Try to consume a token
local allowed = 0
if tokens >= 1 then
    tokens = tokens - 1
    allowed = 1
end

-- Update state
redis.call('HMSET', key, 'tokens', tokens, 'last_refill', last_refill)

-- Return result: allowed (1 or 0) and remaining tokens
return {allowed, tokens}

Script breakdown

State retrieval: Uses HMGET to fetch the current token count and last refill time from a hash
Initialization: On first use, sets tokens to full capacity
Token refill calculation: Computes how many tokens should be added based on elapsed time
Capacity enforcement: Uses math.min() to ensure tokens never exceed capacity
Token consumption: Decrements the token count if available
State update: Uses HMSET to save the new state
Return value: Returns both the decision (allowed/denied) and remaining tokens

Why atomicity matters

Without atomic execution, race conditions could occur:

Double spending: Two requests could read the same token count and both succeed when only one should
Lost updates: Concurrent updates could overwrite each other's changes
Inconsistent state: Token count and refill time could become desynchronized

Using EVAL or EVALSHA ensures the entire operation executes atomically, making it safe for distributed systems.

Using the Python module

The TokenBucket class provides a simple interface for rate limiting (source):

import redis
from token_bucket import TokenBucket

# Create a Redis connection
r = redis.Redis(host='localhost', port=6379, decode_responses=True)

# Create a rate limiter: 10 requests per second
limiter = TokenBucket(
    redis_client=r,
    capacity=10,        # Maximum burst size
    refill_rate=1,      # Add 1 token per interval
    refill_interval=1.0 # Every 1 second
)

# Check if a request should be allowed
allowed, remaining = limiter.allow('user:123')

if allowed:
    print(f"Request allowed. {remaining} tokens remaining.")
    # Process the request
else:
    print("Request denied. Rate limit exceeded.")
    # Return 429 Too Many Requests

Configuration parameters

capacity: Maximum number of tokens in the bucket (controls burst size)
refill_rate: Number of tokens added per refill interval
refill_interval: Time in seconds between refills

For example:

capacity=10, refill_rate=1, refill_interval=1.0 allows 10 requests per second with bursts up to 10
capacity=100, refill_rate=10, refill_interval=1.0 allows 10 requests per second with bursts up to 100
capacity=60, refill_rate=1, refill_interval=60.0 allows 1 request per minute with bursts up to 60

Rate limit keys

The key parameter identifies what you're rate limiting. Common patterns:

Per user: user:{user_id} - Limit each user independently
Per IP address: ip:{ip_address} - Limit by client IP
Per API endpoint: api:{endpoint}:{user_id} - Different limits per endpoint
Global: global:api - Single limit shared across all requests

Running the demo

Get the source files

The demo consists of two Python files. Download them from the redis-py source folder on GitHub, or grab them with curl:

mkdir rate-limiter-demo && cd rate-limiter-demo
BASE=https://raw.githubusercontent.com/redis/docs/main/content/develop/use-cases/rate-limiter/redis-py
curl -O $BASE/token_bucket.py
curl -O $BASE/demo_server.py

Start the demo server

A demonstration web server is included to show the rate limiter in action (source):

# Install dependencies
pip install redis

# Run the demo server
python demo_server.py

The demo provides an interactive web interface where you can:

Submit requests and see them allowed or denied in real-time
View the current token count
Adjust rate limit parameters dynamically
Test different rate limiting scenarios

The demo assumes Redis is running on localhost:6379 but you can easily change the host and port in the demo_server.py script. Visit http://localhost:8080 in your browser to try it out.

Response headers

It's common to include rate limit information in HTTP response headers:

allowed, remaining = limiter.allow(f'user:{user_id}')

# Add standard rate limit headers
response.headers['X-RateLimit-Limit'] = str(limiter.capacity)
response.headers['X-RateLimit-Remaining'] = str(int(remaining))
response.headers['X-RateLimit-Reset'] = str(int(time.time() + limiter.refill_interval))

if not allowed:
    response.status_code = 429  # Too Many Requests
    response.headers['Retry-After'] = str(int(limiter.refill_interval))

Alternative rate limiting algorithms

The token bucket algorithm above handles most use cases but Redis supports other rate limiter patterns that might fit your requirements better. The table below lists four other algorithms alongside token bucket and summarizes their features:

Algorithm	Memory	Accuracy	Burst behavior	Best for
Token bucket	1 key (hash)	Exact	Controlled bursts	APIs with bursty traffic
Fixed window counter	1 key (string)	Approximate	2x burst at boundaries	Simple API limits
Sliding window log	O(n) entries	Exact	No bursts	High-value APIs, audit trails
Sliding window counter	2 keys (string)	Near-exact	Smoothed boundaries	General-purpose APIs
Leaky bucket (policing)	1 key (hash)	Exact	No bursts	Strict no-burst enforcement

The sections below give example implementations of these other algorithms. All of them use redis.call('TIME') inside the Lua script to derive the current timestamp from the Redis server clock. This eliminates clock drift when the limiter runs across multiple application servers.

Fixed window counter

Counts requests within discrete, non-overlapping time intervals. Simplest algorithm — one key per window, one EVAL round trip.

import redis

SCRIPT = """
local key    = KEYS[1]
local limit  = tonumber(ARGV[1])
local window = tonumber(ARGV[2])

local count = redis.call('INCR', key)
if count == 1 then
    redis.call('EXPIRE', key, window)
end

local ttl = redis.call('PTTL', key)

if count > limit then
    return {0, ttl}
end
return {1, ttl}
"""


def is_allowed(client: redis.Redis, key: str, limit: int, window_seconds: int) -> dict:
    script = client.register_script(SCRIPT)
    allowed, ttl = script(keys=[key], args=[limit, window_seconds], client=client)
    return {"allowed": bool(allowed), "retry_after_ms": ttl if not allowed else 0}

Trade-off: A client can make 2x requests by sending limit requests at the end of one window and limit requests at the start of the next.

Sliding window log

Records the exact timestamp of every request in a sorted set. Provides a true rolling window with no boundary bursts.

import uuid
import redis

SCRIPT = """
local key    = KEYS[1]
local limit  = tonumber(ARGV[1])
local window = tonumber(ARGV[2])
local member = ARGV[3]

local t      = redis.call('TIME')
local now    = tonumber(t[1]) + tonumber(t[2]) / 1e6
local cutoff = now - window

redis.call('ZREMRANGEBYSCORE', key, '-inf', cutoff)

local count = redis.call('ZCARD', key)

if count < limit then
    redis.call('ZADD', key, now, member)
    redis.call('EXPIRE', key, window * 2)
    return {1, 0}
end

local oldest = redis.call('ZRANGE', key, 0, 0, 'WITHSCORES')
local retry_after_ms = 0
if oldest[2] then
    retry_after_ms = math.floor((tonumber(oldest[2]) + window - now) * 1000)
end

return {0, retry_after_ms}
"""


def is_allowed(client: redis.Redis, key: str, limit: int, window_seconds: int) -> dict:
    script = client.register_script(SCRIPT)
    member = str(uuid.uuid4())
    allowed, retry_after_ms = script(
        keys=[key], args=[limit, window_seconds, member], client=client
    )
    return {"allowed": bool(allowed), "retry_after_ms": int(retry_after_ms)}

Trade-off: Memory grows O(n) with request volume. Not ideal for high-volume, high-cardinality rate limiting.

Sliding window counter

Blends two fixed-window counters using a weighted average to approximate a true sliding window. Near-exact accuracy with the same low memory footprint as a fixed window. The two keys use hash tags so they map to the same slot in Redis Cluster.

import redis

SCRIPT = """
local base   = KEYS[1]
local limit  = tonumber(ARGV[1])
local window = tonumber(ARGV[2])

local t   = redis.call('TIME')
local now = tonumber(t[1]) + tonumber(t[2]) / 1e6

local window_num = math.floor(now / window)
local elapsed     = (now % window) / window

local curr_key = base .. ':' .. window_num
local prev_key = base .. ':' .. (window_num - 1)

local prev = tonumber(redis.call('GET', prev_key) or 0)
local curr = tonumber(redis.call('GET', curr_key) or 0)

local estimate = prev * (1 - elapsed) + curr

if estimate >= limit then
    return {0, 0}
end

local new_count = redis.call('INCR', curr_key)
if new_count == 1 then
    redis.call('EXPIRE', curr_key, window * 2)
end

return {1, 0}
"""


def is_allowed(client: redis.Redis, key: str, limit: int, window_seconds: int) -> dict:
    script = client.register_script(SCRIPT)
    allowed, _ = script(
        keys=["{" + key + "}"],
        args=[limit, window_seconds],
        client=client,
    )
    return {"allowed": bool(allowed)}

Trade-off: The weighted estimate may let slightly more or fewer requests through than the exact limit. Negligible for most apps.

Leaky bucket (policing)

A virtual bucket fills with incoming requests and drains at a fixed rate. If the bucket is full, requests are rejected immediately. This is the policing variant — requests are allowed or denied instantly with no delay.

import redis

SCRIPT = """
local key       = KEYS[1]
local capacity  = tonumber(ARGV[1])
local leak_rate = tonumber(ARGV[2])

local t   = redis.call('TIME')
local now = tonumber(t[1]) + tonumber(t[2]) / 1e6

local data      = redis.call('HGETALL', key)
local level     = 0
local last_leak = now

if #data > 0 then
    for i = 1, #data, 2 do
        if data[i] == 'level' then
            level = tonumber(data[i+1])
        elseif data[i] == 'last_leak' then
            last_leak = tonumber(data[i+1])
        end
    end
end

local elapsed = now - last_leak
level = math.max(0, level - elapsed * leak_rate)

if level + 1 > capacity then
    return {0, math.floor((level + 1 - capacity) / leak_rate * 1000)}
end

level = level + 1
local ttl = math.ceil(capacity / leak_rate) + 1

redis.call('HSET', key, 'level', level, 'last_leak', now)
redis.call('EXPIRE', key, ttl)

return {1, 0}
"""


def is_allowed(client: redis.Redis, key: str, capacity: int, leak_rate: float) -> dict:
    script = client.register_script(SCRIPT)
    allowed, retry_after_ms = script(
        keys=[key], args=[capacity, leak_rate], client=client
    )
    return {"allowed": bool(allowed), "retry_after_ms": int(retry_after_ms)}

Trade-off: Overflow traffic is rejected immediately. Clients must handle 429 Too Many Requests and retry with backoff.

Learn more

EVAL command - Execute Lua scripts
EVALSHA command - Execute cached Lua scripts
Lua scripting - Introduction to Redis Lua scripting
HMGET command - Get multiple hash fields
HMSET command - Set multiple hash fields
Transactions - Alternative to Lua scripts for atomicity
redis-rate-limiting-python - community-maintained GitHub repo (contributed by Yashwin Reddy) with examples of the alternative rate limiting algorithms.