Rate limiting with sliding window in Redis

Introduction

Rate limiting helps control how often a user can do something, like sending messages or making requests. The sliding window method lets us count actions in a moving time frame, so limits are fair and smooth.

To limit how many times a user can send a message in the last minute.

To control API calls so a user doesn't overload the server in a short time.

To prevent too many login attempts within a few minutes to stop attacks.

To manage how often a user can post comments on a website in the last hour.

Syntax

Redis

redis-cli

// Use a sorted set to store timestamps of actions
// Key: user ID or action identifier
// Command to add current timestamp:
ZADD key current_timestamp current_timestamp

// Remove timestamps older than window size (e.g., 60 seconds):
ZREMRANGEBYSCORE key 0 (current_timestamp - window_size)

// Count actions in the window:
ZCARD key

// Check if count exceeds limit
// If not, allow action; else, deny

Sorted sets store timestamps as both score and member for easy range queries.

Use current time in seconds or milliseconds depending on precision needed.

Examples

Add a timestamp, remove entries older than 60 seconds ago, then count current actions.

Redis

ZADD user:123 1687000000 1687000000
ZREMRANGEBYSCORE user:123 0 1686999940
ZCARD user:123

Sliding window moves forward; old timestamps are removed to keep count accurate.

Redis

ZADD user:123 1687000050 1687000050
ZREMRANGEBYSCORE user:123 0 1686999990
ZCARD user:123

Check current count before allowing new action.

Redis

ZCARD user:123
// If result < limit, allow action
// Else, deny

Sample Program

This example shows adding timestamps for user:42, cleaning old entries outside the 60-second window, and counting actions to enforce rate limit.

Redis

127.0.0.1:6379> ZADD user:42 1687000100 1687000100
(integer) 1
127.0.0.1:6379> ZADD user:42 1687000120 1687000120
(integer) 1
127.0.0.1:6379> ZREMRANGEBYSCORE user:42 0 1687000040
(integer) 0
127.0.0.1:6379> ZCARD user:42
(integer) 2

// Now add a new action at timestamp 1687000150
127.0.0.1:6379> ZADD user:42 1687000150 1687000150
(integer) 1
// Remove timestamps older than 60 seconds from 1687000150 (i.e., before 1687000090)
127.0.0.1:6379> ZREMRANGEBYSCORE user:42 0 1687000090
(integer) 1
// Count current actions
127.0.0.1:6379> ZCARD user:42
(integer) 2

Important Notes

Time complexity for ZADD, ZREMRANGEBYSCORE, and ZCARD is O(log(N)) where N is the number of elements in the sorted set.

Space complexity depends on how many timestamps are stored per user; old timestamps are removed to save space.

Common mistake: forgetting to remove old timestamps, which causes the count to grow and limits to be inaccurate.

Use sliding window rate limiting when you want smooth limits over time, unlike fixed windows that reset abruptly.

Summary

Sliding window rate limiting counts actions in a moving time frame for fairness.

Redis sorted sets store timestamps and allow efficient adding, removing, and counting.

Always remove old timestamps to keep counts accurate and storage small.

Practice

1. What is the main purpose of using a sliding window in rate limiting with Redis?

easy

A. To count actions within a moving time frame for fair limits

B. To store user passwords securely

C. To permanently block users after one action

D. To sort data alphabetically

Solution

1. Step 1: Understand sliding window concept

   Sliding window rate limiting counts actions in a recent time frame that moves forward as time passes.

2. Step 2: Identify purpose in Redis context

   Redis uses sorted sets to track timestamps of actions to enforce limits fairly over time.

3. Final Answer:

   To count actions within a moving time frame for fair limits -> Option A

4. Quick Check:

   Sliding window = moving time frame count [OK]

Hint: Sliding window means counting recent actions, not all time [OK]

Common Mistakes:

• Thinking sliding window blocks permanently
• Confusing with password storage
• Assuming it sorts data alphabetically

2. Which Redis command is essential to add a timestamp to a sorted set for sliding window rate limiting?

easy

A. ZADD

B. GET

C. HSET

D. LPUSH

Solution

1. Step 1: Identify command to add to sorted set

   ZADD adds members with scores (timestamps) to a sorted set in Redis.

2. Step 2: Confirm other commands' purpose

   GET retrieves string values, HSET sets hash fields, LPUSH adds to lists, none add to sorted sets.

3. Final Answer:

   ZADD -> Option A

4. Quick Check:

   ZADD adds to sorted set [OK]

Hint: Use ZADD to add timestamps to sorted sets [OK]

Common Mistakes:

• Using GET which only reads strings
• Confusing HSET (hashes) with sorted sets
• Using LPUSH which is for lists

3. Given this Redis Lua script snippet for sliding window rate limiting:

local now = tonumber(redis.call('TIME')[1])
redis.call('ZREMRANGEBYSCORE', key, 0, now - window)
local count = redis.call('ZCARD', key)
if count < limit then
  redis.call('ZADD', key, now, now)
  return 1
else
  return 0
end

What will be the return value if the current count is equal to the limit?

medium

A. 1

B. nil

C. 0

D. An error

Solution

1. Step 1: Analyze the condition count < limit

   If count is equal to limit, the condition count < limit is false.

2. Step 2: Determine return value when condition is false

   When false, the script returns 0, indicating rate limit reached.

3. Final Answer:

   0 -> Option C

4. Quick Check:

   count == limit returns 0 [OK]

Hint: If count equals limit, return 0 (limit reached) [OK]

Common Mistakes:

• Assuming count == limit returns 1
• Expecting an error instead of 0
• Thinking nil is returned

4. You have this Redis Lua script for sliding window rate limiting:

local now = tonumber(redis.call('TIME')[1])
redis.call('ZREMRANGEBYSCORE', key, 0, now - window)
local count = redis.call('ZCARD', key)
if count <= limit then
  redis.call('ZADD', key, now, now)
  return 1
else
  return 0
end

What is the bug in this script?

medium

A. ZREMRANGEBYSCORE is used incorrectly

B. Using <= allows one extra request beyond limit

C. ZADD should use a string member, not number

D. TIME command returns a string, not number

Solution

1. Step 1: Check condition count <= limit

   This condition allows adding a request when count equals limit, exceeding limit by one.

2. Step 2: Verify other commands usage

   ZREMRANGEBYSCORE usage is correct to remove old timestamps; ZADD accepts numbers as members; TIME returns array with string seconds converted to number.

3. Final Answer:

   Using <= allows one extra request beyond limit -> Option B

4. Quick Check:

   count <= limit allows limit+1 requests [OK]

Hint: Use < not <= to enforce strict limit [OK]

Common Mistakes:

• Thinking ZREMRANGEBYSCORE is wrong here
• Believing ZADD can't use numbers as members
• Confusing TIME return type

5. You want to implement a sliding window rate limiter in Redis that allows 100 requests per 60 seconds. Which sequence of Redis commands best enforces this limit efficiently?

hard

A. HSET to store count, GET to read count, INCR to increment count

B. LPUSH to add timestamp, LLEN to count, LTRIM to remove old timestamps

C. SETNX to set a flag, EXPIRE to reset after 60 seconds

D. ZREMRANGEBYSCORE to remove old timestamps, ZCARD to count, ZADD to add current timestamp

Solution

1. Step 1: Understand sliding window enforcement

   Use sorted sets to store timestamps, remove old ones, count current requests, and add new timestamp.

2. Step 2: Evaluate each option's commands

   ZREMRANGEBYSCORE to remove old timestamps, ZCARD to count, ZADD to add current timestamp uses ZREMRANGEBYSCORE, ZCARD, ZADD correctly for sliding window. LPUSH to add timestamp, LLEN to count, LTRIM to remove old timestamps uses lists which are less efficient for range removals. HSET to store count, GET to read count, INCR to increment count uses hashes and counters, not suitable for sliding window. SETNX to set a flag, EXPIRE to reset after 60 seconds uses simple flags, not counting requests.

3. Final Answer:

   ZREMRANGEBYSCORE, ZCARD, ZADD sequence -> Option D

4. Quick Check:

   Sorted sets + remove old + count + add = correct [OK]

Hint: Use sorted sets with ZREMRANGEBYSCORE, ZCARD, ZADD [OK]

Common Mistakes:

• Using lists which are inefficient for sliding windows
• Using counters without timestamps
• Using flags without counting requests