VPA User's Guide

VPA Resource Deployment

A VPA resource has to be deployed for each Deployment, StatefulSet and DaemonSet. Below is a generic template which can be adapted:

apiVersion: autoscaling.k8s.io/v1beta2
kind: VerticalPodAutoscaler
metadata:
  name: <name_of_vpa_resource>
  namespace: <namespace>
spec:
  targetRef:
    apiVersion: "apps/v1"
    kind:       <Deployment|Statefulset|Daemonset>
    name:      <name_of_target_deployment>
  updatePolicy:
    updateMode: "Auto"
#### Optional section
  resourcePolicy:
    containerPolicies:
      - containerName: "*"
        minAllowed:
          cpu: 300m
          memory: 5Gi
        maxAllower:
          cpu: 4 
          memory: 20Gi
####

A more detailed documentation is available here: https://docs.openshift.com/container-platform/4.5/nodes/pods/nodes-pods-vertical-autoscaler.html#nodes-pods-vertical-autoscaler-using-about_nodes-pods-vertical-autoscaler

Verify VPA

Check the recommendations

• Get a list of VPA resources

  kubectl -n <namespace> get vpa
  

• Get detailes about a specific VPA resource

  kubectl -n <namespace> get vpa <vpa_name> -o yaml
  

Check VPA logs

VPA itself is deployed in the kube-system namespace.

Check evictions

Using kubectl

kubectl -n kube-system logs vpa-updater-XXXXXX

Using grafana => The vpa_caascad_pod_evictions metric can be used to see when pods have been evicted

Check karma

Check karma to see if VPA did not cause any problems. Things to look for are memory and cpu warning.