Introducing Zelos - Zookeeper API leveraging Delos
summary
Zelos implements ZooKeeper semantics on top of Delos, uses a native log plus RocksDB in production, and reaches order-of-magnitude more data per ensemble while matching ZooKeeper latency after adding a batching engine. Session semantics come from a session ordering engine, local and global session managers, a request processor and a watch manager. First production cluster lands in eight months, a migration client flips a barrier node to perform a zero downtime migration and 50 workloads (including critical services) switch databases atomically.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:28What is Zookeeper?
- 4:38Design principles
- 6:02Overview of Delos
- 7:41Zookeeper API
- 9:55Zelos Architecture
- 14:45Zelos vs. Zookeeper performance
- 15:27Performance parity
- 19:24Online Migration
related talks
19mVirtualizing Consensus in Delos for Rapid Upgrades and Happy Engineers
24mOxia - A Horizontally Scalable Alternative to Apache Zookeeper
27mMesos HA Getting Rid of Zookeeper
20mA ZooKeeper Layer for FoundationDB
19mA zone-aware cache system for distributed databases
27mService Discovery Challenges at Scale
30mData @Scale Seattle- Muthu Annamalai
01h 03mZenoh: The Edge Computing Data Fabric