Zero-Touch Fault-Tolerance for Cloud-Native Geo-Distributed Databa...
Talkat Data on Kubernetes Day North America 2023 · on Monday, 2023-11-06· at 09:50· W471ab, McCormick Place·
summary
Demo shows a multi-cluster orchestrator (Nova) deploying YugabyteDB across three regions (three clusters per region, replication factor 3) on five GKE clusters using Helm, StatefulSets and a service mesh, plus Google Cloud managed Prometheus alerts. A recovery job auto-provisions an EastPrime standby, redeploys DB pods, and reconfigures universe; read/write throughput (~140 ops/s) stays steady, follower lag drops to ~120 ms, and full recovery completes in about 50s–2 minutes, achieving zero-touch fault-tolerance.
AI-written summary · about
chapters (26)
- 0:00Introduction
- 1:29Problem definition
- 4:44Failure categories
- 5:05YugabyteDB architecture
- 7:13Single-cluster Kubernetes
- 8:38Multi-region challenges
- 10:44Regional failure limits
- 12:27Multicluster orchestrators
- 12:48What is orchestrator
- 14:55Scheduling policies
- 15:56Zero-touch fault tolerance
- 16:39Problem and solution
- 17:01Demo overview
- 18:49Installing orchestrator
- 19:53Istio service mesh
- 20:57YugabyteDB multi-cluster
- 21:41Recovery policies
- 22:45Simulating failure
- 23:27Automated recovery
- 24:53Performance impact
- 25:13Key takeaways
- 26:17Future extensions
- 27:00Acknowledgments
- 27:43Q&A: Recovery time
- 29:47Q&A: Disaster recovery
- 31:52Q&A: Recovery jobs
related talks
32mBuilding the Multi-Cluster Data Layer
26mA Winning Multi Cloud Data Strategy
31mAn Adventure in Multi Cloud with Kubernetes and Distributed Data
32mDay 2 Operations Best Practices
01h 41mMastering Multi-Region Deployments With YugabyteDB
31mToo Far Too Furious: Multi-Region Cloud Applications
50mWeathering the Cloud Storm: Building Resilient Geo-Distributed Apps
25mDiscover the Easiest Path to Distributed DBaaS: Yugabyte Cloud