Why should I use a Sorted Set for a live leaderboard in Redis?
Question
I'm designing a live leaderboard (the top 100 highest-scoring players) for a gaming platform; millions of players' scores change in real time. I want to keep the data in Redis. Instead of storing each player's score in a plain String key (`user:123:score`) and pulling all users to sort them every time, what are the architectural advantages and the time-complexity (O(log N)) analysis of using Redis's Sorted Set (ZSET) data type?
Answer
Short answer
Storing scores as plain user:123:score strings and sorting in the app is O(N log N) per read over millions of keys — and you re-fetch everything each time. The right tool is a Sorted Set (ZSET).
Context
The real point: the nature of a leaderboard is “sort and give me the top N.” If you try to do that work at read time, scale crushes you; you need to move the sorting to write time. And while you’re putting Redis in the middle of your hot path, it’s worth looking at the cache-side traps too — I covered stampedes in a separate answer.
Why
- The skip-list sorts as you write. The skip-list behind a ZSET keeps everything ordered as you write. So the cost is spread to write time, not read time; reads stay cheap and almost independent of player count.
- Read cost becomes independent of player count.
ZRANGE leaderboard 0 99 REV WITHSCORESreturns the top 100 players already sorted in O(log N + 100); even with millions of players this read is near-constant cost. - The string approach re-fetches everything for every question. Even a single question like “what rank am I?” means pulling and counting everyone; with a ZSET it’s one command.
What to do
- Update scores with
ZADD.ZADD leaderboard <score> <user>updates a member’s score in O(log N). When a player’s score changes it’s a single command; you don’t touch the whole table. The moment the score changes the set stays internally sorted. - Read the top 100 with
ZRANGE ... REV.ZRANGE leaderboard 0 99 REV WITHSCORESis all you need (ZREVRANGEis deprecated as of Redis 6.2.0); you do no sorting on the app side. - Get one player’s rank with
ZREVRANK. “What rank am I?” is answered byZREVRANK leaderboard <user>in O(log N). - Window large sets. For huge sets, use periodic snapshots and time-windowed boards (separate daily/weekly keys). Watch ties: equal scores order lexicographically; if needed, encode a tiebreaker (e.g. a timestamp) into the score to make the order deterministic.
Bottom line: a ZSET is purpose-built for live leaderboards; use it instead of strings + app-side sorting. The moment you move sorting out of read time and into write time, the leaderboard runs instant and cheap even with millions of players.
Compiled from questions I have run into in the field or been asked.
Comments
Sign in with your GitHub account to join the discussion. Comments are stored in GitHub Discussions.