Systems @Scale Summer Live Q&A Session #1
summary
Polaris acts as a client-only detector that flags cache state divergence while a per-host consistency tracing library embedded in every cache traces state transitions end to end and pinpoints causes like failovers or dropped invalidations. Unit tests catch known scenarios but teams add chaos-style fault injection and replayable distributed tests to find unknown ones. TTL versus invalidation tradeoffs depend on update frequency and read patterns. A new Delos/Zelos stack using RocksDB and out-of-order execution outperforms ZooKeeper across workloads and is slated to replace it, with the Delos platform planned for open source.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:24Caching bugs and consistency
- 2:30Polaris detects cache issues
- 4:14Data motion vs data at rest
- 6:01Kafka trade-offs: self vs cloud
- 8:49Open sourcing Delos and Zelos
- 9:54Cache consistency: invalidation vs TTL
- 12:03Re-architecting Kafka for cloud
- 14:30Event-driven future
- 17:45Zelos horizontal scalability
- 19:27Zelos vs ZooKeeper performance
- 21:15Debugging cache invalidation
- 23:03Closing
related talks
34mTwo Hard Things: Naming Things & Cache Invalidation, Valkey Edition
26mWeek 1 Live Q&A
52mWeek 3 Live Q&A
17mDeveloper Experience Q&A
37mWeek 2 Live Q&A
50mPRAGMATIC DEVELOPMENT AT STACK OVERFLOW
19mTech Tales: How Data DX enables business success
27mEngineering Productivity Starts with Predictable Behaviour