Tracking Millions of Ganks in Near Real Time
summary
League of Legends moves from MySQL blobs to an Riak cluster to serve millions concurrent and terabytes of game data with near‑real‑time reads and writes. The work describes concrete data models: raw immutable game records, match‑history sets with non‑primary indexes, counter aggregation using per‑app in‑memory shards and periodic flushes, and a eventually consistent key/value store conflict strategy that uses deltas plus truncated aggregates to bound object size. System is live in regions but rolled out carefully and even powers a Thunderdome demo streamed at twitch tv slash teemo dies.
AI-written summary · about
chapters (10)
- 0:00Start
- 4:49LEGACY
- 9:08CHALLENGES TO SOLVE
- 10:10ARCHITECTURE
- 12:23HOW RIAK WORKS
- 14:23APP CHALLENGES
- 16:15RAW DATA/STATEMENTS OF TRUTH
- 17:13MATCH HISTORY/1:N INDEX
- 20:17AGGREGATE STATS/SETS OF COUNTERS
- 23:58GLOBAL AGGREGATES/HIGH FREQUENCY SETS OF COUNTERS
related talks
46mScaling With Friends
45mReal Time Analytics with MongoDB
37mDiabolic Database Design
33mCounting things, very fast, on the Internet
46mEvents Analysis with Riak, Perl and Erlang at Booking.com
39mUsing Erlang, Riak and the ORSWOT CRDT at bet365 (...)
48mHyper Speed! When Big Data Blooms
46mScalable Realtime Analytics using Druid