Intro + DeepDive: SIG Scalability
Talkat KubeCon + CloudNativeCon North America 2019 · on Thursday, 2019-11-21· at 16:25· Room 7AB - San Diego Convention Center, Upper Level·
summary
Defines a practical scalability envelope using SLIs/SLOs (roughly 99% of API calls within 1s and 99% of pod startups within 5s), tests against concrete limits (5,000 nodes, ~30 pods/node), and runs density and load scenarios with cluster-loader and kube‑mark (hollow nodes) to simulate large clusters — kube‑mark cuts node VMs by ~10x. Uses perf dashboard, Go profiles and Prometheus snapshots to detect regressions; introduces leases and endpoint slices to reduce API load and improve real-world scalability.
AI-written summary · about
chapters (18)
- 0:00Intro
- 2:00What is Kubernetes Scalability?
- 3:06What is Scalability?
- 4:22K8s Scalability Envelope
- 7:59Scalability - how to define it?
- 9:34Scalability - SLO Coverage
- 10:43SLI/SLO Principles
- 15:10Scalability SLI / SLOS
- 18:51Scalability SLO Framework
- 19:40Scalability Limits
- 21:07Feedback Loop
- 27:04Types of Scalability Tests
- 32:06Perf-test Infrastructure #4
- 33:31ClusterLoader2 API
- 36:48ClusterLoader2 #2
- 41:33Kubemark's Hollow Node
- 48:41Scalability Regressions
- 52:43Scalability Improvements
related talks
29mIntro + DeepDive: SIG Scalability
43mIntro + Deep Dive: SIG Scalability
34mDeep Dive: Scalability SIG
39mKubernetes Scalability Definition Evolution
48mScale and Performance Testing of Kubernetes: Answers for Specific Applications
25mAn SLO-Driven Approach to Enhance Kubernetes Cluster Reliability
35mDeep Dive: Kubernetes (Instrumentation) SIG
20mSIG-Autoscaling Deep Dive and Q+A