Skip to main content
The Kubernetes Clusters page shows what your clusters cost, how well their workloads use what they request, and how much rightsizing would save. It works for any cluster running the CostGraph Operator.

​
Clusters

The clusters list opens with month-to-date figures for all your clusters:
  • Total clusters cost, with the month’s forecast and the cluster that contributes most.
  • The cost split into CPU, memory, and storage.
  • Savings available / mo, split into Rightsizing (downsize recommendations) and Idle cleanup (recommendations to remove a workload).
Each row shows the cluster, its provider, its node count, its cost this month and last month, the change, the potential cost impact, and its health. Filter by status, provider, region, Kubernetes version, or operator version.

​
Cluster overview

Select a cluster to open it. The header shows its vCPUs, memory, Kubernetes version, node count, and its cost this month against last month. The tabs are Overview, Nodes, GPUs, Persistent volumes, and Workloads. Overview shows:
  • Cluster details and Nodes & workloads: on-demand and spot nodes, and scheduled and unscheduled pods.
  • Cost & efficiency: total cost for the period you pick, the potential cost impact, workload efficiency, and the spending trend.
  • CPU and memory usage over time.
  • Namespaces: each namespace’s pods, CPU and memory used, cost, and number of recommendations, for the last 7 days, the last 30 days, or month to date.

​
Workload efficiency

Workload efficiency is the share of your current spend that you would still spend if every container were rightsized. At 100%, workloads request what they use. At 60%, rightsizing would cut 40% of the cost. CostGraph shows a dash when no container in the cluster sets requests.

​
Nodes

The Nodes tab lists every node with its location, instance type and pricing model, CPU and memory, pod count, month-to-date and projected cost, and recommendation. Filter by status, provider, region, availability zone, instance type, architecture, operating system, or usage type. Select a node to see its cost per hour, its metrics, and the containers running on it. Its Recommendations tab compares the node’s instance type with a recommended one.

​
Workloads

The Workloads tab shows the current cost of your workloads, the monthly savings from rightsizing, and how many workloads have recommendations. Each row shows the workload, its namespace and kind, its containers, its current cost per hour, its most significant recommended action, and its urgency. Filter by workload, namespace, resource, action, or urgency. Select a workload to see its cost summary, and each container’s current request, recommended request, and cost. Open a container to see:
  • Usage: CPU or memory usage over 7 or 30 days, at the percentile you pick, with the recommendation and estimated savings.
  • Metrics: the container’s CPU, memory, disk, and GPU over time.
  • Details: cost, recommended cost, savings, urgency, and its requests, limits, and usage.
A recommendation raises or lowers a container’s request, limit, or both, or suggests removing a workload that does no work. When CostGraph can’t make one, it says why, such as not enough data, missing metrics, or an unstable workload. Negative savings mean a workload is under-provisioned and needs more than it requests. See How CostGraph works for how recommendations are sized.

​
Persistent volumes

The Persistent volumes tab shows how many volumes you have, their total provisioned size, their cost, and the potential cost impact. Each row shows the volume’s provisioner, phase, claims, size, and P99 used space. It also shows a cheaper disk type where one exists, the recommended action, and its urgency. Select a volume to see its storage class, access mode, zone, price, and the claims bound to it. It also shows how much of its IO and cost each container accounts for.

​
GPUs

The GPUs tab lists the accelerators on the cluster’s nodes. See GPUs for how CostGraph measures GPU use.

​
Idle cost

Idle cost is the part of a node’s bill that no container accounts for. Read it through the OpenCost-compatible API, which returns it as its own allocation or spreads it across your workloads.

​
Metrics time range

Metrics charts on clusters, nodes, and containers cover the last hour by default. You can switch to the last 6 hours, 24 hours, 7 days, or 30 days.

​
API

Everything on these pages is available in the CostGraph API under /api/v1/tenant/clusters: clusters, namespaces, nodes, workloads, containers, persistent volumes, and their metrics. Graph AI and the MCP server read the same data.

​
Next steps

Insights

Over-provisioned workloads, idle volumes, and other findings across your clusters.

Recommendations

Every source of savings recommendations in CostGraph.