Deep Dive into Autoscaling
Talkat KubeCon + CloudNativeCon North America 2019 · on Wednesday, 2019-11-20· at 11:50· Exhibit Hall AB - San Diego Convention Center, Ground Level·
summary
Explains HPA, VPA and Cluster Autoscaler internals and how they interact to cut cloud bills: recommend price expander by default, describe expanders (least-wastage, most-pods, priority) and show why CA picks two N1-standard-4s vs one N1-standard-8; CA emits Kubernetes events for every scale decision. Gives tunable flags and defaults: scale down unneeded time = 10m, default utilization ≈50%, max empty bulk delete = 10, plus API/QPS and pod disruption checks, and notes CA targets serving workloads but can be tuned for batch.
AI-written summary · about
chapters (28)
- 0:00Intro
- 0:19Overview
- 1:38What does VPA do
- 2:17What does HP do
- 3:01What does CA do
- 4:22What does Cluster Scaler do
- 6:23Why did we scale out N1 standard
- 6:41Expanders
- 7:31Least wastage expanded
- 8:21Price expanded
- 9:19More expanded
- 9:52Priority expanded
- 10:27Pick the right expander
- 10:58Scale down empty nodes
- 14:25Cost vs availability
- 19:07Unready
- 21:45How to use Autoscaling
- 22:53Questions
- 24:08How can we use these methods
- 25:24Can we use Autoscaling in an onprem setup
- 26:53Can we use Autoscaling while balancing availability
- 27:59Is there any work being done to use utilization
- 29:30Does the cluster autoscaler take into account taints
- 30:00Does the cluster autoscaler take into account erosion
- 30:33observability best practices
- 31:26considerations
- 31:58Cluster Autoscaling
- 35:32Can Autoscaling create a new load pool
related talks
34mIntroduction to Autoscaling
37mDeep Dive: Autoscaling SIG
36mIntro: Autoscaling SIG
35mSIG Autoscaling Intro
17mScaling Made Simple: Implementing Autoscaling for Optimal K8s Performance
24mSIG-Autoscaling: Introduction + HPA/VPA Updates
29mKubernetes Autoscaling: Beyond One Size Fits All
34mDeep Dive: SIG Autoscaling