Core Concepts
Cluster Map
The Cluster Map draws your cluster's resources and how they relate, grouped by namespace, in Grid, Flow and Traffic layouts you can read and filter.
The Cluster Map draws the active cluster’s resources and how they relate, grouped by namespace. Open it with Cluster Map in the top bar, next to Dashboard. It reads the live API connection, so it is not available on a metrics-only cluster.
Layout modes
- Grid — Compact grid of resources within namespace regions. Best for overview.
- Flow (the default) — Horizontal dependency chain, left to right: Ingress / Gateway → HTTPRoute → Service → Deployment, StatefulSet, DaemonSet, CronJob → ReplicaSet, Job → Pod, then the ConfigMaps, Secrets, HPAs, volumes and nodes they use. Best for understanding how requests reach a workload.
- Traffic — Only what is actually talking: pods that appear in observed
Hubble flows, the Services in front of them, the Ingresses, Gateways and
HTTPRoutes that let traffic in, and destinations outside the cluster (by
FQDN or IP) in an
(external)region. Edges carry the observed rate and the verdict (dropped traffic stands out), aggregated over a Flow window of 1, 2, 5 or 10 minutes. It needs the agent with Hubble enabled; without flow data the canvas stays empty and explains why.
Reading and filtering the map
Namespace regions are laid out in a grid whose column count grows with the number of visible regions (up to 4), with infrastructure namespaces and external endpoints placed last. Clicking a resource opens its details in a side panel. The configuration panel on the left holds:
- Namespaces — Collapse all folds every region into a compact super-node for a zoomed-out read; click one to drill in. Expand all brings the resources back.
- Workloads — Grouped (the default) shows the healthy replicas of a workload as one node with a count (click it to open the group); One per pod draws every pod. Unhealthy pods, StatefulSet pods and pods behind a headless Service are never folded. History hidden / shown toggles retired ReplicaSets and completed Jobs (hidden by default); a rollout in progress and a failed Job always show.
- Edges — one toggle per group: Ownership, Service, Config, Storage, Autoscale and Traffic. The Traffic group overlays observed flows on the Grid and Flow layouts too.
- View — Animated / Static, and Reset for nodes you dragged.
- Resource Types — show or hide each kind; the namespace list under it narrows the regions.
The layout, the flow window, the grouping and the nodes you drag are remembered per browser, with drag positions kept separately for each layout. The map refreshes every minute and on live changes.
Relationship Detection
| Relationship | Detection Method | Edge Type |
|---|---|---|
| Deployment → ReplicaSet → Pod, StatefulSet / DaemonSet / Job → Pod, CronJob → Job | ownerReferences chain | owns |
| Service → Pods | Label selector matching | selects |
| Ingress → Service | spec.rules.http.paths.backend | routes |
| Gateway → HTTPRoute | HTTPRoute spec.parentRefs | routes |
| HTTPRoute → Service | spec.rules.backendRefs | routes |
| HPA → its scale target | spec.scaleTargetRef | hpa |
| Pod → PVC | spec.volumes.persistentVolumeClaim | uses |
| PVC → PV | spec.volumeName | bound |
| Pod → ConfigMap / Secret | volumes (projected included) | mounts |
| Pod → ConfigMap / Secret | envFrom and env valueFrom | envFrom |
| Pod → Secret | spec.imagePullSecrets | imagePull |
| Caller → Service → Pod, Pod → external endpoint | observed Hubble flows | traffic |