RocksDB
summary
RocksDB is Facebook’s fork of LevelDB optimized for SSDs and production scale, shipping multi-threaded compaction and per-file WAL + manifest atomic updates. Deployments report 2x smaller on disk versus InnoDB and dramatic write reduction (examples show ~100 MB/s down to ~15 MB/s), plus Parse runs MongoDB on RocksDB in production with huge compression (reportedly 5 TB → 285 GB) and lower P99 latencies via document-level locking. Ongoing work targets memtable concurrency, tiered storage, tombstone handling and adaptive ingest shaping.
AI-written summary · about
chapters (24)
- 0:00Intro
- 2:50Overview
- 5:49Experimentation (2011-2013)
- 7:23Explosion (2013-2015)
- 9:17New Challenges (2015-)
- 9:47MongoRocks
- 11:44MyRocks
- 13:29Log Structured Merge Trees
- 15:49Flush full memtable
- 16:07Compaction
- 17:07Reads
- 21:20RocksDB Files - MANIFEST
- 25:12RocksDB Files - Table files
- 28:16Backups
- 31:52Work in progress
- 33:48Web scale MySQL at FB
- 39:25Better has many dimensions
- 40:49Storage efficiency
- 41:59Tiered Storage • Performance with efficiency • Write, read MB/second- disk
- 45:14Multi-threaded memtable
- 46:11Manage ingest
- 48:47Adaptive algorithm
- 49:38Database RAM
- 51:30X is faster than RocksDB
related talks
52mRocksDB Performance and Reliability Practics
51mMyRocks deep dive and production deployment at Facebook
47mLessons Learned from Our Main Database Migrations at Facebook
12mMyRocks at Facebook and a Roadmap
11mEvolution of Development Priorities in Key-value Stores Serving Large-scale Applications
45mMyRocks Table Engine for MySQL
12mSpanDB: A Fast, Cost-Effective LSM-tree Based KV Store on Hybrid Storage
50mInnoDB to MyRocks Migration in Main MySQL Database