Tsara real-time leaderboard ga miliyoyin 'yan wasa

Question

IT Interview Practice · Accepted Answer

**Yi amfani da Redis sorted set (ZSET) da aka key da score: rank da top-N su zama O(log N) built-in operations, don haka ba ka taɓa sort na miliyoyin rows a kowane request ba.** ZSET yana ajiye members da aka tsara da score ta atomatik; wannan daidai shi ne leaderboard.

## Gine-gine

```text
Game svc ──ZADD score──▶ Redis ZSET (sorted by score) ──ZREVRANGE 0 99──▶ Top-100 API
                              │
                              ├─ ZREVRANK user ─▶ "your rank" (around-me: rank±5)
                              └─ AOF/RDB + async ─▶ Postgres (durable source of truth)
```

## Yadda yake aiki

- **ZADD** yana upsert score na ɗan wasa a O(log N). Redis yana kula da oda ta **skip list**, don haka ba a buƙatar re-sort.
- **Top-N**: `ZREVRANGE board 0 99 WITHSCORES` yana dawo da top 100 a O(log N + 100).
- **Rank ɗinka**: `ZREVRANK board user` yana ba da matsayin 0-based; **around-me** shi ne `ZREVRANGE rank-5 rank+5` — bangaren 'yan wasa kusa da kai.
- **Durability**: Redis in-memory ne. Kunna **AOF** (append-only file) don replay, kuma ka mirror scores zuwa **Postgres** cikin asynchronous a matsayin ainihin system of record, domin asarar Redis ta zama mai iya dawowa.

## Sharding don miliyoyin masu amfani

ZSET ɗaya yana riƙe da tens of miliyoyin members da kyau (memory-bound, kusan 100MB ga entries miliyan 1). Sama da haka, ko don global scale, **shard da score range** (0–1k, 1k–10k…) domin kowane shard ya zama contiguous band — global top-N yana karanta top shard kawai. Guji sharding da user hash: sake haɗa ranks a kan random shards mai tsada ne.

## Ra'ayin hira

Mai yin hirar yana son ganin ka **nemi data structure daidai** maimakon `ORDER BY score LIMIT 100` a SQL (wanda ke re-sort a kowane hit kuma yana mutuwa a ƙarƙashin write load). Amsoshi masu kyau suna ambaton ZSET; amsoshi masu kyau sosai suna rufe **around-me queries, ties, persistence, da shard strategy**. Tarkon gargajiya: mantawa cewa Redis volatile ne kuma ɗaukar shi a matsayin source of truth — restart yana share board sai dai idan AOF/RDB ko DB mirror ya kasance.

## Sulhu & kurakurai

- **Ties.** Scores masu daidai ana tsara su da member lexicographically, wanda yayi kama da random. Encode tiebreaker cikin score (misali `score * 1e13 - timestamp`) domin masu nasara na farko su yi rank sama.
- **Hot key.** Global ZSET ɗaya key ne a shard ɗaya — write hotspot. Buffer/batch `ZADD`s, ko yi amfani da time-bucketed boards (kullum/mako) don baza load.
- **Consistency.** Reads na iya bin async DB mirror da daƙiƙu — yayi kyau ga leaderboards, ba ga payouts ba; yi sulhu daga Postgres kafin ba da kyaututtuka.
- **Memory cost.** Miliyoyin members suna kashe ainihin RAM; expire seasonal boards da TTLs.
- **Lokacin da ba za a yi ba:** leaderboard na 'yan dubbai da ake refresh kowane minti yayi kyau a matsayin cached SQL query — Redis ZSET yana haske lokacin da writes suke constant kuma rank dole ya zama nan take.

Gine-gine

Yadda yake aiki

Sharding don miliyoyin masu amfani

Ra'ayin hira

Sulhu & kurakurai

Tambayoyi masu alaƙa