MyRocks: best practice at Alibaba - MySQL Storage Engine
summary
Alibaba tests MyRocks (RocksDB) against AliSQL/InnoDB and deploys it to over a hundred production databases, cutting storage to 1/7 or typically 1/3–1/8 of InnoDB and saving millions, while inserts run about 30–40% slower, primary-key reads match InnoDB and range scans are ~50% slower. Known limits include no online DDL, single-threaded compaction, heavy commits and occasional OOM on large DDLs. Improvements add CJK charset support, thread pool, slow-SQL filtering, PK optimizations and table-level parallel replication; roadmap targets hybrid hot-InnoDB / cold-MyRocks migration.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:13Agenda
- 1:53MyRocks DB
- 2:42Issues at Alibaba
- 5:55Performance
- 8:30MyRocks without a cycle
- 11:15Benefits of MyRocks
- 12:07No limitation
- 16:15Improvements
- 17:26Modifications
- 21:18Take away
- 22:59Whats next
- 25:53Outro
related talks
12mMyRocks at Facebook and a Roadmap
45mMyRocks Table Engine for MySQL
52mRocksDB Performance and Reliability Practics
50mInnoDB to MyRocks Migration in Main MySQL Database
47mLessons Learned from Our Main Database Migrations at Facebook
57mRocksDB
51mMyRocks deep dive and production deployment at Facebook
39mLessons Learned from Our Main Database Migrations