Skip to main content
In this guide, you’ll learn how to execute a pod-deletion experiment on Amazon EKS using AWS FIS. We’ll walk through:
  1. Selecting the built-in EKS pod-delete scenario
  2. Creating and configuring an experiment template
  3. Launching the experiment
  4. Monitoring pod replacement and service continuity
This approach helps you validate your Kubernetes Deployment or ReplicaSet automatically replaces pods without user-facing disruption.

​
Prerequisites

Before you begin, ensure:
  • You have an existing Amazon EKS cluster named pet-site.
  • The IAM role you’ll assign to FIS (e.g., EKS-FIS-Role) has permissions to delete pods and write logs.
  • CloudWatch Logs group (e.g., FISExperiments) is either created or will be created by FIS.
IAM PermissionPurpose
fis:CreateExperimentTemplateCreate FIS experiment templates
fis:StartExperimentLaunch experiments
eks:DeletePodsDelete pods in the target EKS cluster
logs:CreateLogGroupCreate CloudWatch Logs groups
logs:PutLogEventsPublish experiment logs
Make sure your IAM role trusts the fis.amazonaws.com service principal. For more details, see AWS FIS IAM permissions.

​
1. Navigate to Fault Injection Service

  1. Open the AWS Management Console.
  2. Go to AWS Resilience Hub → Fault Injection Service.
  3. Click Experiment templates in the sidebar.
The image shows the AWS Resilience Hub - Fault Injection Service webpage, which offers tools for improving resiliency and performance through controlled experiments. It includes options for creating experiment templates and provides pricing information.

​
2. Select the EKS Pod Delete Scenario

  1. Under Scenario library, search for EKS Stress: Pod Delete.
  2. Click Create experiment template.
FIS immediately populates the template’s name, description, actions, and targets based on the built-in scenario.
The image shows the AWS Fault Injection Service (FIS) Scenario Library interface, displaying options for "EKS Stress: Network Latency" and "EKS Stress: Pod Delete" scenarios.
The image shows an AWS console interface for creating an experiment template, specifically for deleting EKS pods based on cluster and application labels. It includes fields for description, name, and actions/targets.

​
3. Configure the Target

By default, the scenario target is set to delete pods in your cluster. Verify or update:
  • Resource type: eks:pod
  • Cluster: pet-site
  • Label selector: app=pet-site
Ensure these match your Deployment labels so FIS deletes only the intended pods.
The image shows an AWS console interface for editing a target in an EKS (Elastic Kubernetes Service) environment, with options to specify resource type, actions, and parameters.
Review the auto-generated actions and targets:
The image shows an AWS console interface for creating an experiment template, specifically for an EKS pod delete scenario. It includes sections for actions and targets, with "EksPodDelete" and "EksPodDeleteTarget" specified.

​
4. Select IAM Role and Logging

  1. Under Service role, choose your FIS execution role (EKS-FIS-Role).
  2. For Experiment logging, select the CloudWatch Logs group (FISExperiments) or create one on the fly.
The image shows a screenshot of the AWS console, specifically the section for creating an experiment template in AWS Fault Injection Simulator (FIS). It includes options for service access and IAM role selection.
The image shows an AWS CloudWatch interface where a user is selecting a log group from a list, with "FISExperiments" highlighted. The "Choose" button is visible at the bottom right.

​
5. Define the Hypothesis

Hypothesis: When FIS deletes a pet-site pod, the Kubernetes control plane immediately spins up a replacement pod. Service latency and availability remain within SLAs.
Validate this by monitoring your Deployment or ReplicaSet’s desired vs. actual pod counts.
The image shows an AWS Fault Injection Simulator (FIS) interface, specifically an experiment template for deleting one or more EKS pods based on cluster and application label.

​
6. Start and Monitor the Experiment

  1. Click Start experiment.
  2. Watch the experiment lifecycle transition from Initiating → Running → Completed.
  3. Use the FIS dashboard and CloudWatch metrics (e.g., kube_deployment_status_replicas_available) to confirm a new pod appears within seconds.
The image shows an AWS Fault Injection Simulator (FIS) dashboard displaying details of a running experiment, including its ID, state, creation time, and associated resources.
Run FIS experiments only in non-production or controlled environments. Always validate on staging clusters before production.

​
AWS CLI Shortcut

After creating the template, you can start the experiment via CLI: 
aws fis start-experiment \
  --experiment-template-id <your-template-id>

Watch Video