Kubernetes

The Kubernetes view divides the cost of your container infrastructure across your clusters. It also finds the workloads that have too many resources, and the workloads with a risk of throttling. Open the view from Lenses > Kubernetes in the left sidebar.

Use Kubernetes when you need to:

• See how the CPU, memory, storage, and network costs divide across your clusters.
• Find the idle spend of infrastructure and workloads that have too many resources.
• Get rightsizing recommendations with the exact CPU and memory values to apply.
• Reduce cost without a new risk of throttling or out-of-memory (OOM) errors.

The view has two tabs: Expenses and Rightsizing.

Expenses

The Expenses tab shows how the Kubernetes costs divide across your clusters. It also shows which part of the spend the workloads use, and which part is idle.

Summary

At the top of the Expenses tab, four summary cards give a quick view of your Kubernetes spend:

• Total Spend gives the total cost of the selected date range. An efficiency bar divides the total into a used part and an idle part.
• CPU Spend gives the cost of the CPU requests of all workloads.
• Memory Spend gives the cost of the memory requests of all workloads.
• Efficiency gives the percentage of the total spend that active workloads use. A low efficiency value means that much of the cost is idle.

Group By

Use the Group By control to change how the chart and the table aggregate the costs. You can group by these dimensions:

• Namespace (default)
• Cluster
• Node
• Deployment
• Pod
• DaemonSet
• ReplicaSet

Change the grouping to move from a broad cluster view to a workload view or a node view of the cost.

Cost Breakdown Table

The table groups the rows by cluster. For each resource in the selected group-by dimension, it gives a cost breakdown.

To download the current view for an external analysis, click Export CSV above the table.

Rightsizing

The Rightsizing tab analyzes the recent resource usage of your workloads. It then recommends new CPU requests and memory requests. For a workload with too many resources, the recommendation reduces the cost. For a workload with too few resources, the recommendation removes the risk of throttling or of OOM events.

Summary

Four summary cards show the state of your workloads:

• Workloads Scanned gives the number of workloads that DigiUsher analyzed in the selected interval.
• To Shrink gives the number of workloads with at least one resource that is too large. You can reduce the CPU requests or the memory requests without a risk.
• Under-provisioned gives the number of workloads with a risk of CPU throttling or out-of-memory errors.
• Right-sized gives the number of workloads that need no action.

Analysis Interval

The interval selector sets the period of usage metrics that DigiUsher averages for the recommendations. A short interval shows the recent behavior. A long interval removes the effect of temporary spikes.

You can select these intervals: 1H, 3H, 6H, 1D, 1W, and 1M.

If the traffic of a workload changes much, select a long interval (1W or 1M). The recommendation then does not come from a quiet period only. If you investigate a recent change, select a short interval.

Recommendations Table

Each row of the table is one workload. The row gives the current resource requests and the recommended values.

A Decrease row is a savings opportunity. The workload uses much less than it requests, and you can reduce the requests without an effect on the performance.

An Increase row is a reliability risk. The average usage is at or near the current limit, and this increases the risk of CPU throttling or of a memory OOM kill. Correct these workloads before you optimize the cost.

The table sorts by action priority: first Decrease, then Increase, then OK. The recommendations that need action are therefore at the top.

Group By

The Rightsizing tab has the same grouping dimensions as the Expenses tab: Namespace, Cluster, Node, Deployment (default), Pod, DaemonSet, and ReplicaSet.