A kubectl plugin to show the differences between the recommendations of a VerticalPodAutoscaler and the actual resources requests of the targeted controller's pod(s).

The plugin is compatible with Kubernetes servers starting from version 1.16, which have the autoscaling.k8s.io/v1 API group installed. See the other prerequisites for using VPAs on your cluster.

Follow these instructions to install krew. Then run the following commands:

$ kubectl krew install vpa-recommendation
$ kubectl vpa-recommendation --help

Head to the releases page to download an existing release for your OS/arch, and add the path to the binary to your $PATH.

⚠️ The plugin requires Go 1.17+ to build.

To build the plugin yourself from the source, use the following commands:

$ git clone [email protected]:wI2L/kubectl-vpa-recommendation.git
$ cd kubectl-vpa-recommendation
$ make build

It creates a single binary file for the host machine platform/architecture in the dist/ folder. Move it somewhere within your $PATH, such as $GOPATH/bin. You can then invoke it directly (kubectl-vpa-recommendation), or via kubectl (kubectl vpa-recommendation).

The columns % CPU DIFF and % MEMORY DIFF represents the percentage of increase/decrease for the request in terms of the recommendation.

For example, if a request value is set to 4 CPU (4000m), and the recommendation is only 1 CPU (1000m), the difference printed is +300%. On the contrary, if the request (125m) is lower than the recommendation (250m), the difference is then -50%. As a rule of thumb, you can think of positive values as over commitment and negative values as under commitment.

The following examples were produced from a brand-new Kubernetes cluster created with k3d. The VerticalPodAutoscaler resources were automatically created by the goldilocks operator.

Apart from the flags defined by the genericclioptions package and some logging flags, the following options are available with the plugin:

• --all-namespaces, -A: List VerticalPodAutoscaler resources in all namespaces
• --critical-threshold: Critical threshold of percentage difference for colored output. Default to 50
• --namespace, -n: If present, the namespace scope for the request
• --no-colors: Do not use colors to highlight increase/decrease percentage values
• --no-headers: Do not print table headers
• --output, -o: Output format. One of: wide | split | split-wide
• --recommendation-type: The type of recommendation to use in comparisons. One of: lower-bound, target, uncapped-target, upper-bound. Default to target
  • see RecommendedContainerResources for more details about the fields represented by each value
• --show-containers, -c: Display containers recommendations for each VerticalPodAutoscaler resource
• --show-kind, -k: Show the resource type for the requested object(s) and their target
• --show-namespace: Show resource namespace as the first column
• --sort-columns: Comma-separated list of column names for sorting the table. Any of: cpu-diff | cpu-rec | cpu-req | mem-diff | mem-rec | mem-req | name | namespace | target. Default to namespace,name
• --sort-order: The sort order of the table columns. Either asc or desc. Default to asc
• --warning-threshold: Warning threshold of percentage difference for colored output. Default to 20

To view the full list of available options, use the following command:

$ kubectl vpa-recommendation --help

• Unlike the official VPA recommender, which is fully generic and handle any kind of "scalable" resources, the plugin recognize only some well-known controllers such as: CronJob, DaemonSet, Deployment, Job, ReplicaSet, ReplicationController, StatefulSet.

This plugin is licensed under the MIT license. See the LICENSE file.