Skip to main content
A cluster’s Overview says in one screen whether it is connected, what needs you and how healthy it is. Its Health page breaks the health down into scores and five checks, each with a button to the page that explains it. This guide takes you through both, in about five minutes, and says where to go when a check is not healthy.

Before you start

  • A cluster connected to Ankra. Both pages read what the cluster agent reports. While the agent is offline they show the last readings and say so.
  • Permission to read the cluster. Every built-in role can read clusters (Roles and access).

1. Open the cluster’s Overview

Go to Infrastructure → Clusters → <your cluster> (/next/clusters/<cluster-id>). Clusters is in the main menu under Infrastructure, in the Platform group. Overview is the first item of the cluster menu. Under the cluster’s name, the header gives its environment, Kubernetes version and when it was created, and a badge for its connection: The header also holds Settings, Clone cluster (Clone a cluster) and Reconcile cluster, which applies each Stack’s saved definition again: changes made directly on the cluster to what a Stack deploys can be replaced by the saved definition, nothing is changed in Git, and nothing outside the Stacks is touched. Check: the badge reads Connected. If it reads Offline, go to When health is unknown.

2. Read the Overview from the top

When pods crash-loop, fail or cannot pull their image, add-ons are degraded or nodes are not ready, Reported health also shows Ask Ankra why. It sends Ankra AI the question “What is causing … on this cluster?” with what is wrong filled in. The line under Reported health says how fresh the readings are: “Read live from the cluster.”, “Served from the platform cache, N seconds old.” or, past five minutes, “Served from the platform cache, N minutes old: stale.” Check: no Needs you section means the latest operation did not fail and no resource’s latest change failed.

3. Open the Health page

Click View health, or go to Infrastructure → Clusters → <your cluster> → Health (/next/clusters/<cluster-id>/health). The page is headed Cluster health. Refresh health reads everything again.

The scores

Reported health carries a badge and four scores: A cluster with no nodes, pods or add-ons scores 100% on that part. While the agent is offline, Ankra reports no score.

The health checks

Each check reports what the cluster last observed and has a button to the page that explains it. A check has no pass or fail of its own: read the counts. The page says it plainly: these are cluster observations, and running pods and a connected repository do not prove application availability. Under the checks, Runtime charts what each node group used over the last hour: CPU used, Memory used, Network traffic and Disk used.

4. Go from a check that is not healthy

When the cause is not obvious, click Ask Ankra why on the Overview, or ask Ankra AI about the cluster.

5. When health is unknown

When the readings are not recent, the Health page shows Current health is unknown, withholds the scores, and renames Health checks to Last reported checks. The alert says why: If the page cannot read the cluster at all, it says “Cluster health could not be read.” and “Previous readings cannot confirm current health.” The Overview says “The cluster overview could not be read.” Click Try again.

From the CLI and MCP

The CLI reads the same signals without the portal:
Over MCP, get_cluster_status returns the cluster’s health overview and get_cluster_insights the platform’s own AI health diagnoses (MCP tools).

Verify

The cluster is in good shape when:
  • the Overview’s badge reads Connected and it shows no Needs you section,
  • the Health page’s badge reads Healthy, with no Current health is unknown alert,
  • Nodes shows 0 unavailable, Pods 0 failed and 0 crash-looping, Add-ons 0 degraded, and Cluster agent “Reported online”.

Where to go next

  • Cluster metrics - usage over time, per node and namespace
  • Operations - read a failed step and retry it
  • Alerts - be told when a cluster turns unhealthy instead of checking
  • Home - what needs you across the organisation