eBay Search On K8s
Talkat KubeCon + CloudNativeCon North America 2019 · on Tuesday, 2019-11-19· at 11:50· Room 6C - San Diego Convention Center, Upper Level·
summary
eBay runs its Cassini search on Kubernetes serving ~1.4 billion active listings and ~300K QPS by modeling the query layer as a resizable grid: columns = shards, rows = replicas, deployed with a matrix deployment operator and populated by a file distribution operator that coordinates a daemonset data agent. A mutating webhook shares local PVCs between agent and query pods. Performance tuning (kernel, CPU pinning, disabling irqbalance, p‑state/c‑state, switching to IPVS networking) brings latency near bare‑metal at ~80% CPU utilization.
AI-written summary · about
chapters (11)
- 0:00Start
- 2:51Architecture
- 8:43The Why?
- 13:01Search Grid Deployment
- 18:01Data sharing between Pods
- 19:43Out of the Box Performance
- 20:13What moved the curve?
- 21:56Performance Optimizations
- 22:31Lessons Learned
- 26:23Future Work
- 29:28Conclusion
related talks
32mRunning eBay’s High-Performance Workloads with Kubernetes
35meBay Geo-Distributed Database on Kubernetes [A]
38mModel and Operate Datacenter by Kubernetes at eBay
28mBuilding and Managing Kubernetes with Kubernetes
28mBuilding and Managing Kubernetes with Kubernetes
32mScaling Shopify's Search: Enhancing Elasticsearch Resilience With Kubernetes and KE...
27mContinuous Integration at Scale on Kubernetes [B]
36mOperator Framework to Manage Production Grade Stateful Workloads