SIG Autoscaling Deep Dive
summary
Alpha Vertical Pod Autoscaler is available as a CRD and uses an admission plugin to override pod requests on creation while recommendations come from metrics-server and VPA stores per-container min/max and target values; the updater applies changes by killing pods one-by-one and currently cannot resize in-place. Cluster Autoscaler is GA since 1.8, supports GCP AWS and self‑hosted, runs every 10s, treats identical machines as node groups (MIG/ASG), makes decisions by importing scheduler predicates and running a simulated scheduler + greedy bin‑packing, and reacts only to unschedulable pods.
AI-written summary · about
chapters (28)
- 0:00Intro
- 0:21What is Autoscaling
- 0:38Horizontal Autoscaling
- 1:18Vertical Autoscaling
- 2:14API
- 2:46Vertical Path Autoscaling
- 6:19The API
- 6:29Update Pot Requests
- 7:13Override Pot Requests
- 7:23Admission Plugins
- 8:20Recommendations
- 10:16Scaling
- 11:05Cluster Autoscaling
- 11:18Scheduling
- 13:54Note Groups
- 14:45Scaler
- 17:13Node Groups
- 18:02Bin Packing
- 18:58Options
- 22:22Misconceptions
- 23:26What does it do
- 24:30If it doesnt work
- 24:51Cluster Status
- 25:57Events
- 26:57FAQ
- 27:19SIG T
- 27:44Questions
- 29:39Expander
