Pod rightsizing

Prev Next

This section describes the Kompass Pod rightsizing optimization solution.

Kubernetes scaling typically focuses on horizontal adjustments—adding or removing pods—but that alone doesn’t address inefficiencies from misconfigured resource requests within pods.

The Kompass Pod rightsizing solution complements horizontal scaling with automated vertical adjustments, ensuring each container has the right CPU (vCPU) and memory allocation (Mi) based on real usage.

Pod rightsizing works frictionlessly with horizontal autoscalers (like HPA, KEDA) that optimize cluster-wide capacity, by fine-tuning resources at the Pod level to reduce waste and improve workload stability.

Pod rightsizing identifies these types of nonoptimized workloads:

  • Under-provisioned workload: memory or CPU are nearing maximum utilization
    Kompass recommends to increase vCPU or Mi requests

  • Over-provisioned workload: memory or CPU are far from maximum utilization
    Kompass recommends to decrease vCPU or Mi requests

The Pod rightsizing solution provides detailed recommendations for rightsizing CPU and memory requests, but its core strength is automated correction: When you apply Pod rightsizing to a workload, Kompass adjusts workload resources in real time to keep workloads optimized without manual effort.

Built-in rollback mechanisms protect workload stability by reverting changes if CPU throttling or out-of-memory (OOM) events occur.

The solution integrates with GitOps tools like ArgoCD, ensuring changes align with CI/CD workflows and are not overridden. Kompass supports common Kubernetes workload types (Deployments, DaemonSets, StatefulSets) and currently operates in AWS environments. For teams running Kubernetes at scale, it reduces manual overhead, enforces consistency, and helps maintain efficient, stable resource usage across clusters.

Pod rightsizing relies on two Kompass components in the cluster:

  • The Kompass Insights agent is installed.
    This agent provides visibility into the workloads running in the cluster.
    It can take up to 24 hours after installing the agent for Kompass to deliver recommendations.

  • The Kompass Pod rightsizing solution is enabled.
    When the solution is enabled, you can apply it to workloads.
    It can take up to 1 hour after enabling the solution before you can apply it to workloads.

For more information, see Deploy Kompass Compute.

Where to go from here