Skip to main content
A cluster’s Metrics page charts its resource usage and workload signals over a window you choose. The charts work without any setup: the Ankra agent collects node, namespace and workload metrics from the Kubernetes API and each node’s kubelet, and request counts and response times from a Traefik ingress controller’s own metrics, and sends them to Ankra. A Prometheus you connect adds the charts those built-in metrics cannot draw, such as network drops, CPU throttling and GPU telemetry.

Before you start

  • A cluster connected to Ankra with its agent online. The agent sends the built-in metrics. Its connection is on Settings → Agent and on the cluster’s Health page.
  • Permission to read the cluster. The page needs clusters.read. Charts read from a connected Prometheus also need kubernetes.read. Every built-in role holds both; see Roles and access.
  • A connected Prometheus, for some charts only. Network drops, Disk operations, CPU throttling and the GPU telemetry charts read only from a Prometheus. GPU telemetry also needs the NVIDIA DCGM exporter in that Prometheus. To connect one, see Prometheus.

Open the Metrics page

Go to Infrastructure → Clusters → <your cluster> → Metrics (/next/clusters/<cluster-id>/metrics). The page is headed Understand cluster usage and opens on the Resources charts for the Last 1 hour.

Choose the window and the charts

The Metrics view section at the top holds every control. The page address keeps the time range, chart group, source and filters, so you can bookmark or share a view. Type a filter and click Apply filters; Clear filters empties both. The charts redraw for the new window, group, source or filter. With Choose an available source, each chart reads Ankra metrics when the cluster is sending them and the built-in metrics can draw that chart. Otherwise it reads the connected Prometheus.

What each group charts

Under each chart, a line names the source that answered and when it was read. A chart with more than one series has a selector to show one series or All series. Operations recorded in the window are marked on the charts. Missing samples stay gaps in the line: a gap is not a zero.

Check where the metrics come from

Below the controls, three rows say what the page can read: Configure a metrics source opens the cluster’s Settings → Integrations page, where you connect a Prometheus.

Size namespaces against their requests

Choose Namespace sizing in Metrics. A table compares what each namespace requests with what it used over the most recent stretch of the selected window’s length: The table covers every namespace. The node and namespace filters apply to the charts, not to this table. A namespace that uses far less than it requests holds capacity other workloads cannot schedule onto.

See the operations in the window

Operations during this window, at the bottom of the page, lists the platform operations recorded in the selected window with their Operation, Time and Status. Click an operation’s name to open it. When nothing ran, the section reads “No operations were recorded in this window.” Use it to tie a change in usage to a deploy or an upgrade. Operations are described on Operations.

From the CLI

The CLI reads live usage and runs PromQL against the connected Prometheus.
ankra cluster top needs metrics-server in the cluster, not Prometheus. ankra cluster metrics query and query-range need a connected Prometheus. Over MCP, the same reads are get_node_usage, get_pod_usage, query_prometheus and query_prometheus_range (MCP tools). Every flag is in the CLI reference.

What can go wrong

A chart reads “This metric is unavailable.” with “Choose Connected Prometheus for this metric.” The chart is one the built-in metrics cannot draw and Source is set to Ankra metrics. Set Source to Connected Prometheus, and connect one first if Connected Prometheus reads Not configured. A chart reads “This metric is unavailable.” with “Check the source and filters. This is not a zero measurement.” The source did not answer. Read the source rows: Connected but failing means Ankra reached the Prometheus and its queries failed; follow what can go wrong with a Prometheus. Click Try again on the chart once the source answers. A chart reads “No samples returned”. The source answered with nothing for this window. Filters match exact names only, so check the spelling of the node or namespace, or clear the filters. For a chart drawn from a connected Prometheus, the exporter behind it must run in the cluster: node-exporter for Network drops, the kubelet’s cAdvisor metrics for Disk operations and CPU throttling, and the DCGM exporter for GPU telemetry. Ankra metrics reads “Not receiving recent metrics”. The agent is not sending metrics: it is not connected, or metrics collection has been turned off in the agent. Check its connection on Settings → Agent or the Cluster agent check on the cluster’s Health page. Agent Helm values describes the metrics_collect.enabled setting. Ankra metrics reads “Not enabled”. Ankra’s built-in metrics are not offered where this cluster is connected, so the charts can read only a connected Prometheus. Connect one; see Prometheus. The source rows read “The source status could not be read. Charts report their own results.” Ankra could not read the cluster’s metrics sources. Each chart still reports whether it could be drawn. Click Refresh metrics to read them again.

Prometheus

Connect a Prometheus, Grafana Cloud, Mimir or VictoriaMetrics.

Cluster overview and health

Read the cluster’s health checks and what to do when one fails.

Alerts

Get notified when a metric crosses a threshold.

Operations

Follow the platform operations that run on the cluster.