Data @Scale Seattle- Muthu Annamalai
summary
ZippyDB is a distributed key‑value store built on RocksDB that Facebook ships after an 8–9 month build and now runs on thousands of servers serving hundreds of billions of requests per day. It uses Paxos for per‑shard primary/secondary replication, persists Paxos messages in a data shuttle replication log, and applies ordered updates to the store; supports TTL via compaction filters, server‑side read‑modify‑write hooks, counters, queues and a high‑throughput lock service. Followers provide hot backups and read scaling; system is production‑proven.
AI-written summary · about
chapters (8)
- 0:00Start
- 0:39Why?
- 3:40What?
- 8:39Verticals
- 14:14Reads
- 15:58Followers
- 16:40Asynchronous replication
- 17:42Bird's Eye View
related talks
57mRocksDB
01h 01mPanel: Facebook's Distributed Database Evolution
12mSpanDB: A Fast, Cost-Effective LSM-tree Based KV Store on Hybrid Storage
52mRocksDB Performance and Reliability Practics
29mCharacterizing, Modeling, and Benchmarking RocksDB Key-Value Workloads at Facebook
11mEvolution of Development Priorities in Key-value Stores Serving Large-scale Applications
19mA zone-aware cache system for distributed databases
51mMyRocks deep dive and production deployment at Facebook