wantal

Use Case	How VPA Helps
Stateful databases	Scales CPU and memory without restarting pods, preserving connections and avoiding failovers
Batch and ML workloads	Adjusts resources dynamically as job phases shift between CPU-intensive and memory-intensive stages
JIT-compiled applications	Provides extra CPU during startup (via CPU Startup Boost) and scales back once the application is warm
Cost optimization	Eliminates over-provisioning by continuously right-sizing pods based on observed consumption patterns
Game servers and real-time services	Adapts to fluctuating demand (e.g., shifting player counts) without service interruption

wantal

Tool	Purpose	Installation
Minikube	Provisions a local, single-node Kubernetes 1.35 cluster	Download or choco install minikube
kubectl	Kubernetes CLI, configured to communicate with the Minikube cluster	Install guide or choco upgrade kubernetes-cli
Git	Required to clone the official autoscaler repository	Pre-installed on most systems

wantal

Feature	Status in 1.35	What It Means
In-Place Pod Resize	Stable (GA)	CPU and memory are mutable on running Pods; CGroups are patched live
VPA InPlaceOrRecreate mode	Beta	VPA can resize Pods in-place before falling back to eviction
Memory limit decrease	Now permitted	Previously prohibited; Kubelet now allows it with best-effort OOM prevention
Prioritized resizes	New	Deferred resizes are retried based on PriorityClass, QoS class, and age
Resize observability	New	Kubelet metrics and Pod events for tracking in-place resize operations