Kubernetes Cost Prediction Action

Predict the cost of Kubernetes manifests (specs) in CI! Make cost decisions before merging changes.

This is a GitHub Action, powered by Kubecost, to make cost predictions for K8s workloads before they are applied to your cluster. It does not require you to have Kubecost installed, but will have highly-accurate cost and usage information for your environment if you do.

In action:

Usage

Add this Action as a step in one of your Actions workflows and point it at a single YAML file or a directory containing at least one YAML file. Non-YAML files will be ignored. The YAML files will be interpreted as Kubernetes manifests and a cost prediction will be run on supported types of Kubernetes objects.

If you aren't familiar with GitHub Actions, check out GitHub's quickstart documentation.

Simple

Below is an excerpt from a workflow written with this Action. This is the easiest way to add Kubernetes cost prediction to your CI. If you want a premade workflow file to riff on, check out the "Advanced" example below.

- name: Run prediction
  id: prediction
  uses: kubecost/cost-prediction-action@v0.1.1
  with:
    # Set this to the path containing your YAML specs. It can be a single
    # YAML file or a directory. The Action will recursively search if this
    # is a directory and process all .yaml/.yml files it finds.
    path: ./repo

# Write/update a comment with the prediction results.
- name: Update PR with prediction results
  uses: edumserrano/find-create-or-update-comment@v1
  with:
    issue-number: ${{ github.event.pull_request.number }}
    body-includes: '<!-- kubecost-prediction-results -->'
    comment-author: 'github-actions[bot]'
    edit-mode: replace
    body: |
      <!-- kubecost-prediction-results -->
      
      ## Kubecost's total cost prediction for K8s YAML Manifests in this PR

      \```
      ${{ steps.prediction.outputs.PREDICTION_TABLE }}
      \```

Advanced (full workflow)

This is a full Actions workflow file, with commented-out sections and explanations highlighting advanced features and some complex use-cases. You can copy-paste this into a file in your .github/workflows folder and start tuning it to use as a live Action on your repo.

name: Predict K8s spec cost
on: [pull_request]

jobs:
  predict-cost:
    runs-on: ubuntu-latest
    steps:
      # Check out the current repo to ./repo
      - uses: actions/checkout@v2
        with:
          path: ./repo
          
      # If using the API support, you need to make sure the Action runner has
      # network access to your instance of Kubecost. This is infra dependent;
      # the following example works with GKE (make sure to set up the necessary
      # secrets).
      # https://docs.github.com/en/actions/guides/deploying-to-google-kubernetes-engine
      # - name: Setup gcloud
      #   uses: google-github-actions/setup-gcloud@v0.2.0
      #   with:
      #     service_account_key: ${{ secrets.GCP_SA_KEY_B64 }}
      #     project_id: ${{ secrets.GKE_PROJECT_ID }}
      # 
      # Get GKE credentials so kubectl has access to the cluster
      # - name: Get GKE credentials
      #   uses: google-github-actions/get-gke-credentials@v0.2.1
      #   with:
      #     cluster_name: ${{ secrets.GKE_CLUSTER }}
      #     location: ${{ secrets.GKE_ZONE }}
      #     credentials: ${{ secrets.GCP_SA_KEY_B64 }}
      #     project_id: ${{ secrets.GKE_PROJECT_ID }}
      # 
      # - name: Forward the kubecost service
      #   run: |
      #     kubectl port-forward --namespace kubecost service/kubecost-cost-analyzer 9090 &
      #     sleep 5
      
      # If you use Helm, you should template the chart and then run the Predict
      # Action targeting the result. Here's an example of how to do that.
      # 
      # - name: Install helm
      #   run: |
      #     curl https://raw.githubusercontent.com/helm/helm/master/scripts/get-helm-3 | bash
      # 
      # - name: Helm template
      #   run: |
      #     helm template RELEASENAME ./repo >> ./templated.yaml

      - name: Run prediction
        id: prediction
        uses: kubecost/cost-prediction-action@v0.1.1
        with:
          log_level: "info"
          # Set this to the path containing your YAML specs. It can be a single
          # YAML file or a directory. The Action will recursively search if this
          # is a directory and process all .yaml/.yml files it finds.
          # 
          # If you use Helm, you probably want to run "helm template", output
          # to a path like ./templated.yaml, and set "path: ./templated.yaml".
          path: ./repo
          # Set this to either:
          # - localhost:9090/model if port forwarding OR
          # - The URL of your Kubecost instance if the runner has direct network
          #   access, e.g. "https://kubecost.example.com:9090/model"
          #
          # If unset, the Action will use Kubecost's default pricing to make a
          # prediction and it will be unable to make
          #
          # kubecost_api_path: "http://localhost:9090/model"

      # Write/update a comment with the prediction results.
      - name: Update PR with prediction results
        uses: edumserrano/find-create-or-update-comment@v1
        with:
          issue-number: ${{ github.event.pull_request.number }}
          body-includes: '<!-- kubecost-prediction-results -->'
          comment-author: 'github-actions[bot]'
          edit-mode: replace
          body: |
            <!-- kubecost-prediction-results -->
            
            ## Kubecost's total cost prediction for K8s YAML Manifests in this PR

            \```
            ${{ steps.prediction.outputs.PREDICTION_TABLE }}
            \```

      # Alternatively, you can just output the prediction in the Action log.
      # - name: output raw yaml prediction
      #   run: |
      #     echo "${{ steps.prediction.outputs.PREDICTION_TABLE }}"

Inputs/Outputs

Action inputs

Name	Description	Required?	Default
`path`	The path of a file or directory that contains K8s YAML manifests to predict the cost impact of	Yes
`kubecost_api_path`	URL of your Kubecost API. If provided, cost predictions will be a diff based on cost data tracked by your Kubecost instance. If not provied, cost predictions will be a total cost based on Kubecost's default pricing.	No
`log_level`	The log level to run the Action with. Set to `debug` for more granularity or `warn` or `error` for less granularity.	No	`info`

Action outputs

Name	Description
`PREDICTION_TABLE`	An ASCII-formatted table of the cost prediction. Best rendered in monospace.

Limitations

The Action currently only supports predicting .yml/.yaml specs. If you have specs in other formats, you will have to put them into YAML before running prediction logic. E.g. for Helm, use helm template. More support planned, please open an issue describing your use case if it is not yet supported.

The Action supports a limited set of Kubernetes object types. We are working to expand the set of supported types.

The Action does not yet support prediction on only changed files.

The Action does not provide predictions for objects/specs without container resource requests.

Development

Source code for the container is mostly closed. Kubecost engineers, visit cmd/costpredictionaction in KCM for more information about development, testing, and releasing.