Implementing Vertical Pod Autoscaling

1. Installing VPA Components

Example: Install VPA using Helm

helm repo add vpa https://charts.fairwinds.com/stable
helm install vpa vpa/vpa --namespace vpa-system --create-namespace

Component	Description	Installation Method	Status
VPA Admission Controller	Validates and modifies pod specs	Helm Chart	NEW
VPA Recommender	Collects resource usage data and makes recommendations	Helm Chart	NEW
VPA Updater	Applies recommended resource limits to pods	Helm Chart	NEW

Note: Ensure the namespace vpa-system exists before installing.

Warning: VPA requires cluster-admin permissions to function properly.

2. Creating VPA Resource

Example: Basic VPA Configuration

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-app
  updatePolicy:
    updateMode: "Auto"

Field	Description	Values
apiVersion	API version of the VPA resource	`autoscaling.k8s.io/v1`
kind	Kind of the resource	`VerticalPodAutoscaler`
metadata.name	Name of the VPA resource	String
spec.targetRef	Reference to the target resource (Deployment, StatefulSet, etc.)	`apiVersion`, `kind`, `name`
spec.updatePolicy.updateMode	How VPA updates the resource	`Off`, `Initial`, `Auto`, `Recreate`

Note: The VPA resource must be in the same namespace as the target resource.

Warning: Ensure that the target resource has sufficient permissions to allow VPA to make changes.

3. Setting Update Mode

Mode	Description	Behavior	Notes
Off	Vertical Pod Autoscaler (VPA) does not make any changes to pod resource requests.	No updates are made to pods.	Useful for testing or when you want to observe VPA behavior without applying changes.
Initial	VPA sets resource requests only on newly created pods.	Existing pods are not modified; new pods get updated resource requests.	Helps in avoiding disruption to running workloads while allowing new pods to benefit from optimized resources.
Recreate	VPA updates resource requests and recreates pods to apply new settings.	Pods are terminated and recreated with updated resource requests.	Useful when you want to ensure all pods are running with optimized resources, but may cause downtime.
Auto	VPA updates resource requests and attempts to update pods without recreating them.	Pods may be updated in-place if possible, otherwise recreated.	Best for minimizing disruption while still optimizing resource usage.

Example: Setting Update Mode in VPA Spec

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-deployment
  updatePolicy:
    updateMode: "Recreate"

4. Configuring Resource Policy

Policy	Description	Values	Notes
resourcePolicy	Defines how the VPA should handle resource requests and limits	`minChangePercentage`, `maxChangePercentage`	Controls how aggressively VPA can change resources
minChangePercentage	Minimum percentage change in resource allocation	`10`, `25`	Prevents small, frequent adjustments
maxChangePercentage	Maximum percentage change in resource allocation	`50`, `100`	Prevents drastic changes that may destabilize pods

Example: VPA Resource Policy Configuration

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-policy-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-app
  resourcePolicy:
    containerPolicies:
    - containerName: "app-container"
      minChangePercentage: 20
      maxChangePercentage: 50

Note: The resourcePolicy field is optional and defaults to no restrictions on changes. It's recommended to set minChangePercentage to prevent excessive updates.

Warning: Setting maxChangePercentage too high can lead to resource starvation or over-allocation. Ensure that changes align with your application's stability requirements.

5. Setting Min Allowed Resources

Resource	Minimum Allowed	Description
cpu	`100m`	Minimum CPU allocation for the pod
memory	`128Mi`	Minimum memory allocation for the pod

Example: VPA Configuration with Min Allowed Resources

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: example-deployment
  resourcePolicy:
    containerPolicies:
    - containerName: "*"
      minAllowed:
        cpu: 100m
        memory: 128Mi

6. Setting Max Allowed Resources

Resource	Property	Value Type	Description
maxAllowed	`cpu`	`string`	Maximum CPU allocation allowed for a pod (e.g., `2`, `2000m`)
maxAllowed	`memory`	`string`	Maximum memory allocation allowed for a pod (e.g., `4Gi`, `4096Mi`)

Example: Configuring Max Allowed Resources in VPA

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-deployment
  resourcePolicy:
    containerPolicies:
    - containerName: "app"
      maxAllowed:
        cpu: "2"
        memory: "4Gi"

Note: Setting maxAllowed prevents VPA from requesting more resources than specified, ensuring resource limits are respected.

7. Viewing VPA Recommendations

Resource	Recommendation Type	Values	Notes
vpa.name	`string`	The name of the VPA resource	Used to identify the VPA in Kubernetes API
vpa.namespace	`string`	The namespace where VPA is defined	Used to scope the VPA to a specific namespace
vpa.spec.targetRef	`object`	Reference to the target resource (Deployment, StatefulSet, etc.)	Points to the workload that VPA will manage
vpa.status.conditions	`array`	List of conditions indicating VPA status	Includes `Ready`, `Initializing`, and `Recommending`
vpa.status.recommendations	`object`	Recommended resource requests and limits	Includes `container`, `requests`, and `limits`

Example: Viewing VPA Recommendations via kubectl

kubectl get vpa my-vpa -n my-namespace -o yaml

Example: Viewing VPA Status

kubectl describe vpa my-vpa -n my-namespace

Note: VPA recommendations are updated every few minutes based on historical usage patterns.

Warning: VPA does not recommend limits for containers; it only recommends requests.

8. Understanding VPA Components

Component	Description	Role
VPA Controller	Core component that manages VPA objects and updates Pod resource requests	Manages VPA lifecycle, calculates recommendations, and applies changes to Pods
VPA Recommender	Component that analyzes historical resource usage and generates recommendations	Collects metrics from Prometheus or Heapster, analyzes usage patterns, and provides resource suggestions
VPA Updater	Component that modifies Pod resource requests based on VPA recommendations	Applies updated resource limits and requests to existing Pods in the cluster

Example: VPA Controller Configuration

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-app
  updatePolicy:
    updateMode: "Auto"
  resourcePolicy:
    containerPolicies:
    - containerName: "app"
      minAllowed:
        cpu: "100m"
        memory: "128Mi"
      maxAllowed:
        cpu: "2"
        memory: "1Gi"

Note: VPA components work together to provide intelligent resource scaling based on historical usage patterns.

9. Using VPA with HPA

Vertical Pod Autoscaler (VPA) vs Horizontal Pod Autoscaler (HPA)

VPA adjusts resource requests/limits of existing pods, while HPA scales the number of pods
VPA focuses on pod resource allocation; HPA focuses on pod count
Both can be used together for comprehensive autoscaling strategies

Feature	VPA	HPA
Resource Adjustment	Modifies CPU/Memory requests/limits	Adjusts pod replica count
Trigger Condition	Based on resource usage (CPU/Mem)	Based on metrics like CPU, memory, or custom
Pod Impact	Updates existing pods	Creates/destroys pods
Configuration	Uses `v1` API, `VerticalPodAutoscaler` kind	Uses `v2` API, `HorizontalPodAutoscaler` kind

Example: VPA Configuration

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: vpa-example
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-app
  updatePolicy:
    updateMode: "Auto"

Example: HPA Configuration

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: hpa-example
spec:
  scaleTargetRef:
    apiVersion: "apps/v1"
    kind: Deployment
    name: my-app
  minReplicas: 2
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 70

Best Practices for Using VPA with HPA

Deploy both VPA and HPA in the same namespace for better visibility
Use VPA to optimize resource allocation and HPA to scale based on demand
Monitor both autoscalers for potential conflicts or over-optimization

10. Deleting VPA Resources

Resource Type	Description	Command
VerticalPodAutoscaler	Represents the VPA resource to be deleted	`kubectl delete vpa <vpa-name>`

Example: Deleting a VPA Resource

kubectl delete vpa my-app-vpa

Note: Ensure that the VPA resource is not in use by any pods before deletion to avoid unexpected behavior.