Introduction

Prevent your deployment from being down by having at least two pods or an automatic scaling via HorizontalPodAutoscaler. Otherwise, e.g., a single failure in a pod, crashing host, or availability zone being down would cause unavailability of a deployment.

Supported Advice States

Action Needed

The advice automatically discovers whether a Kubernetes deployment has at least two pods configured, e.g., via spec.replicas or HorizontalPodAutoscaler. The discovery is performed by the open-source library kube-score and integrated into this advice.

Validation Needed

Asks you to validate with an automatically created experiment that provided features work as expected when a container of a single pod is restarting.

By default, the experiment contains a TODO step to add validation logic for your service. We highly recommend resolving this step to facilitate the discussion of whether your service works as expected from a user's point of view. You can also tag your Kubernetes Deployment with the label tags.steadybit.com/service-validation to support different validations:

• http: For an HTTP check on a GET endpoint
• k6: For running a k6 load test
• gatling: For running a Gatling load test
• jmeter: For running a JMeter load test

Once you create the experiment, the user is asked to specify the corresponding validation, and Steadybit runs it throughout the experiment.

Implemented

Once you have configured Kubernetes to have at least two pods or the deployment is targeted by a HorizontalPodAutoscaler.

Disabling Advice

Every advice is automatically active as described in extension-kubernetes#configuration. You can also activate advice individually, using for this advice, the identifier com.steadybit.extension_kubernetes.advice.k8s-single-replica.