MSFTCOST07-BP03 Change your scale strategy for Windows Containers on Kubernetes using Karpenter
Karpenter is a Kubernetes cluster autoscaler that dynamically provisions EC2 instances based on your workload demands, automatically launching right-sized instances in response to pending pods and continuously evaluating the cluster to optimize costs by consolidating workloads onto more efficient instance types. The tool proactively replaces outdated nodes with newer ones to maintain security compliance and supports diverse compute requirements by selecting from a broad range of instance types and purchasing options, including both On-Demand and Spot instances.
Desired outcome: Expect to achieve improved resource utilization, reduced operational overhead, and optimized cloud costs. EKS clusters will dynamically scale to meet application demands, maintain up-to-date and secure infrastructure, and efficiently manage diverse workloads without manual intervention, ultimately leading to a more responsive, cost-effective, and easily managed Kubernetes environment on AWS.
Common anti-patterns:
-
Teams often configure Karpenter with unnecessarily specific instance type constraints or narrow capacity requirements, limiting its ability to efficiently provision nodes and potentially increasing costs by forcing the use of suboptimal instance types.
-
Organizations frequently deploy Karpenter without properly configuring Pod Disruption Budgets (PDBs), leading to unexpected application downtime during node consolidation or replacement operations, as Karpenter may terminate nodes without ensuring proper workload migration.
Benefits of establishing this best practice:
-
By allowing Karpenter to intelligently select from a broad range of instance types and automatically consolidate workloads, organizations can significantly reduce their AWS compute costs while maintaining optimal performance for their applications.
-
Teams spend less time on manual cluster management and capacity planning, as Karpenter automates node provisioning, scaling, and replacement activities, enabling engineers to focus on higher-value development tasks.
-
With Karpenter's automated node replacement feature, clusters maintain better security hygiene through regular updates and patches, reducing the risk of vulnerabilities while ensuring compliance with security standards without manual intervention.
Level of risk exposed if this best practice is not established: Medium
Implementation guidance
To implement Karpenter effectively, define flexible provisioner configurations that accommodate both Linux and Windows workloads, ensuring appropriate instance types are available for each OS. Set up distinct provisioners with OS-specific requirements, configure Pod Disruption Budgets for critical applications, and establish proper taints and tolerations to ensure workloads land on compatible nodes. Regularly monitor cluster behavior and costs to optimize your configuration.
Implementation steps
-
Install and configure Karpenter in your EKS cluster, ensuring proper IAM permissions and VPC settings
-
Create flexible provisioner configurations for both Linux and Windows workloads, specifying appropriate instance types and purchasing options
-
Set up Pod Disruption Budgets for critical applications to maintain availability during node consolidation
-
Configure monitoring and alerting to track Karpenter's performance and cluster resource utilization
-
Regularly review and adjust Karpenter settings based on observed cluster behavior and cost metrics
Resources
Related documents:
