Learning Redis - Sets, Sorted Sets & HyperLogLog

#Introduction

After Lists and Hashes in episode 4, now we cover three structures related to unique collections: Sets, Sorted Sets, and HyperLogLog.

Sets give you set operations like intersection and union — perfect for tags and membership detection. Sorted Sets add a score for ranking — the foundation of leaderboards and priority queues. Finally, HyperLogLog is a magic trick for counting millions of unique visitors with almost no growth in memory. Let's dissect them one by one.

#Sets: Unordered Unique Collections

#Basic Operations

A Set holds unique elements without order. The same element cannot be added twice:

SADD, SREM, and SMEMBERS

redis-cli SADD tags:post:1 "redis" "database" "cache"
redis-cli SREM tags:post:1 "cache"
redis-cli SMEMBERS tags:post:1

SADD tags:post:1 "redis" "database" "cache" adds three tags, SREM removes one, SMEMBERS lists all. Membership and count:

Check membership and count

redis-cli SISMEMBER tags:post:1 "redis"
redis-cli SCARD tags:post:1

SISMEMBER returns 1 if the element exists — an O(1) operation that is extremely cheap for membership checks. SCARD counts the number of elements.

#Set Operations: SUNION, SINTER, SDIFF

This is the power of Sets that no other data type has:

Union, intersection, and difference

redis-cli SINTER tags:post:1 tags:post:2
redis-cli SUNION tags:post:1 tags:post:2
redis-cli SDIFF tags:post:1 tags:post:2

SINTER gives the intersection (tags that appear in both posts), SUNION the union, SDIFF the difference. Real-world use cases: content recommendations ("users who liked A also liked B"), anomaly detection, and "people you may know" features.

#Sorted Sets: Ranking with a Score

#ZADD and Reading Ranks

A Sorted Set is like a Set, but every element has a numeric score that determines ordering:

ZADD and reading ranks

redis-cli ZADD leaderboard 100 "player1"
redis-cli ZADD leaderboard 250 "player2"
redis-cli ZRANGE leaderboard 0 -1 WITHSCORES
redis-cli ZREVRANGE leaderboard 0 -1

ZADD leaderboard 100 "player1" adds a player with a score of 100. ZRANGE sorts ascending, ZREVRANGE descending — the standard leaderboard pattern. Equal scores are ordered lexicographically by member.

#ZSCORE, ZRANK, and ZINCRBY

Score, rank, and increment

redis-cli ZSCORE leaderboard "player2"
redis-cli ZREVRANK leaderboard "player2"
redis-cli ZINCRBY leaderboard 50 "player2"

ZSCORE shows a member's score, ZREVRANK its rank position (0 = highest), and ZINCRBY increments the score atomically — exactly what's needed to update real-time game scores.

#ZRANGEBYSCORE: Ranges Based on Score

Query by score range

redis-cli ZRANGEBYSCORE leaderboard 100 200

ZRANGEBYSCORE leaderboard 100 200 lists all members with a score between 100 and 200. This pattern becomes the basis for the sliding window rate limiter (episode 11) and score-based priority queues.

#HyperLogLog: Counting Unique with Constant Memory

#The Probabilistic Concept

Counting millions of unique visitors exactly requires a large Set and lots of memory. HyperLogLog uses probabilistic estimation: about 0.81% error, but memory stays at ~12KB no matter how many elements there are. Redis doesn't store the elements themselves; it only exploits hash distribution properties to estimate cardinality.

PFADD, PFCOUNT, and PFMERGE

redis-cli PFADD visits:2026-08-03 "user-1" "user-2" "user-1"
redis-cli PFCOUNT visits:2026-08-03
redis-cli PFADD visits:2026-08-04 "user-2" "user-3"
redis-cli PFMERGE visits:week1 visits:2026-08-03 visits:2026-08-04

PFADD visits:2026-08-03 "user-1" "user-2" "user-1" adds visitors (duplicates are ignored), PFCOUNT estimates the unique count. PFMERGE merges several HLLs — for example counting unique weekly visitors from daily data.

Info

The 0.81% error tolerance is almost always sufficient for analytics dashboards and reach estimation. If you need exact counts for financial or audit purposes, use a Set or Sorted Set — at the price of much larger memory.

#Main Use Cases

Sets: article tags, online user lists, whitelists/blacklists, set operations for recommendations.
Sorted Sets: game leaderboards, top products, priority queues, task scheduling (score = timestamp).
HyperLogLog: daily/weekly unique visitors, event analytics, large-scale deduplication.

Structure	Properties	Typical Use Case
Set	Unique, unordered, set operations	Tags, online users
Sorted Set	Unique, ordered by score	Leaderboards, rate limiting
HyperLogLog	Unique estimation, ~12KB memory	Unique visitor analytics

#Summary

Episode 5 equipped you with Sets with set operations, Sorted Sets for score-based ranking, and HyperLogLog for cardinality estimation with constant memory: SADD/SINTER, ZADD/ZREVRANK/ZINCRBY, and PFADD/PFCOUNT/PFMERGE.

Key takeaways:

Sets guarantee uniqueness; SISMEMBER is an O(1) membership check.
SINTER/SUNION/SDIFF provide set operations for recommendations and filtering.
Sorted Sets order elements by score — the heart of leaderboards.
ZINCRBY updates scores atomically; ZRANGEBYSCORE unlocks the rate limiter pattern.
HyperLogLog estimates millions of uniques with only ~12KB of memory.
Choose exact (Set/Sorted Set) when you need accuracy, HLL when you need memory savings.

In the next episode, episode 6, we cover Streams — a log-based structure for event streaming and message brokering, similar to Apache Kafka but built into Redis. You'll learn XADD, consumer groups, acknowledgments, and pending message management. This is favorite material for many backend engineers!