Resource Requests and Limits Under the Hood: The Journey of a Pod Spec
summary
Explains how a Pod spec flows: requests are recorded in etcd, scheduler uses them to pick a node, kubelet hands pod to the CRI runtime which sets cgroups; limits turn into cgroups values and CPU limits map to CFS period/quota (default 100ms so 500m → 50ms, 2000m → 200ms) and memory over-limit triggers the OOM killer. Covers QoS classes (Guaranteed, BestEffort), CPU shares vs request semantics, and VPA modes off/initial/auto and requests are for planning limits are for enforcement.
AI-written summary · about
chapters (23)
- 0:00Intro
- 1:28Doggy Daycare Analogy
- 2:36Here's an ordinary pod spec
- 3:21Requests for planning
- 4:06Update Cluster State
- 5:15Scheduler
- 5:47Create Pod on Node
- 5:59The Journey of a Request
- 6:30Requests passing through
- 7:09Requests summary
- 7:56QoS Class in Kubernetes
- 8:44Now let's talk about limits
- 9:01Limits enforce.
- 9:36Assign Pod to Node
- 10:15Kubelet to Container Runtime
- 10:51Limits summary
- 11:55Limits by level
- 13:08Container Primitives
- 15:54CFS Quota? Period?
- 17:13CRI Runtime vs OCI Runtime
- 18:54Pod Autoscaling
- 19:42Vertical Pod Autoscaler (VPA)
- 22:21Conclusion
related talks
40mLimits, Requests, QoS... Rethinking What You Thought You Knew About Scheduling in Kubernetes
45mLimits, Requests, QoS, PriorityClasses: Demystifying Scheduling in K8s!
47mOptimizing Kubernetes Resource Requests/Limits for Cost-Efficiency and Latency
47mCPU & Memory, Requests & Limits, QoS & CFS: The JVM in the Kubernetes Pod
29mResize Your Pods w/o Disruptions aka How to Have a Cake and Eat a Cake
38mA Tale of Tail Latency: Understanding k18n CPU Requests and Limits for Sust. and Profit
05mResource Roulette: Winning the Kubernetes Allocation Game
36mTaming the Beast: Advanced Resource Management With Kubernetes