Core Concepts
Resource Views
26 resource list views with search, pagination and live metrics, plus Namespaces, Events, cluster RBAC and Helm releases, and a tabbed detail page for every object.
On this page
Supported resources
KubeBolt has 26 resource list views, grouped in the cluster sidebar the same way as below. Each supports search, namespace filtering, server-side pagination (50 per page) and, where it applies, live CPU and memory from Metrics Server. The sidebar shows a count next to each and marks the ones your identity cannot list (see RBAC).
| Sidebar group | Resources |
|---|---|
| Pinned | Pods, Nodes (plus Insights, Applications and, on KubeBolt Cloud, Autopilot) |
| Workloads | Deployments, StatefulSets, DaemonSets, Jobs, CronJobs |
| Traffic | Services, Ingresses, NetworkPolicies, Cilium Policies (CiliumNetworkPolicies), Cilium Cluster Policies (CiliumClusterwideNetworkPolicies), PodDisruptionBudgets, Gateways, HTTPRoutes, Endpoints (EndpointSlices) |
| Storage | PVCs, PVs, StorageClasses |
| Config | ConfigMaps, Secrets (keys by default — values only through the audited reveal), Service Accounts, HPAs |
| Extensions | Certificates (cert-manager), ArgoCD Apps, VPA (VerticalPodAutoscalers) |
The Cluster group adds three views that aren’t plain lists: Namespaces, Events, and RBAC (ClusterRoles and ClusterRoleBindings), next to Clusters, where clusters are added, renamed and removed (Multi-Cluster). Extension and Cilium views show up with data only when the CRD is installed; a missing CRD is reported as a neutral “Optional integrations not detected” note, not as a permission problem, and a CRD installed later is picked up within two minutes. ReplicaSets have no list of their own: they appear in a Deployment’s history, the Related tab and the Cluster Map.
Namespaces pages 50 at a time, counts every namespace and filters by name. When KubeBolt’s credentials may not list events, Events shows Access Restricted instead of failing, and Kobi is told that seeing no events does not mean there are none.
Applications lists the Helm releases in the cluster: status, chart, values, manifest, notes, the resources a release owns and its revision history. It is read-only: KubeBolt does not install, upgrade or roll back Helm releases. See Helm Applications.
On a metrics-only cluster (an agent that ships metrics but gives KubeBolt no API access) these views are off; the dashboard still works.
Resource detail views
Each resource has a tabbed detail page at /:type/:namespace/:name (_ stands
in for the namespace of cluster-scoped resources). Available tabs vary by type:
| Resource | Tabs beyond Overview, YAML and Events |
|---|---|
| Pods | Containers, Logs, Terminal, Files, Volumes, Related, Monitor |
| Deployments, StatefulSets, DaemonSets | Pods, Logs, Terminal, Related, History, Monitor |
| Jobs | Pods, Logs, Related |
| CronJobs | Jobs, Related |
| Services | Related |
| Nodes | Pods, Monitor |
| NetworkPolicies, Cilium policies, PodDisruptionBudgets | Matched Pods, Related |
| PersistentVolumeClaims | Monitor |
What each tab does:
- Overview — Key fields, status, conditions, labels, annotations
- YAML — Syntax-highlighted definition with copy/download, and edit for editors. Secret
datavalues always come back asREDACTED, and ConfigMap values are redacted too when the key or the value looks like a credential.managedFieldsand thelast-applied-configurationannotation are stripped. An Argo CD Application’s Helm values, parameters and plugin environment are hidden too. Saving puts back what the view hid, a secret store’s inline credential included, so an edit never writesREDACTEDover a real value. - Pods — The pods a workload or node owns. A Deployment’s, StatefulSet’s, DaemonSet’s or Job’s tab pages them 50 at a time, sorted by name, so a workload with hundreds of pods opens right away
- Logs — Pod and workload logs with pod and container selectors and the last 100, 500 or 1000 lines (5000 on a pod). Filters add a time window, the previous container’s logs and a regex. Live tails every 10 s; it is off by default so reading a log is not interrupted
- Terminal — WebSocket-to-SPDY exec bridge with xterm.js; on workloads, pick the pod first
- Files — Exec-based file browser with directory navigation, content viewer and download (Pods only — open a pod from a workload’s Pods tab to browse it). It runs commands in the container, so it needs Editor, like the terminal
- Containers — Container specs, env vars, ports, mounts (Pods only)
- Volumes — Volume mounts and claims (Pods only)
- Related — Parent and child resources from topology edges
- History — Revision history via ReplicaSets (Deployments) or ControllerRevisions (StatefulSets, DaemonSets), with roll back to a revision (editor role)
- Matched Pods — The pods a policy or PDB actually selects
- Events — Events filtered to the specific resource
- Monitor — CPU and memory gauges from Metrics Server, plus history charts (CPU and memory by container, network traffic, filesystem on nodes) when the agent ships metrics; PVCs show space and inode usage from kubelet volume stats. Without the agent the tab keeps the gauges and says how to get history
Actions
The toolbar above the tabs has Describe (the kubectl describe output, hiding the same values as the YAML tab) and,
when Kobi is configured, Ask Kobi, for every role. Writes
need Editor, deletes need Admin, and all of them are audited:
| Resource | Actions |
|---|---|
| Deployments, StatefulSets, DaemonSets | Set image, Restart; under Actions ▾: Set resources, Set env, Scale (not DaemonSets), Pause / Resume rollout (Deployments). Roll back from the History tab |
| Pods | Restart (deletes the pod for its controller to recreate); under Actions ▾: Evict pod (honours PodDisruptionBudgets), Debug (an ephemeral debug container). Port-forward from a container’s ports |
| CronJobs | Trigger now, Suspend / Resume |
| Nodes | Cordon / Uncordon; Drain (Admin) |
| Secrets | Reveal (see below) |
| Every kind | Edit metadata (labels and annotations), Delete (Admin) |
On a cluster reached through the agent, writes, exec, port-forward and the file
browser need the agent in operator mode; in reader mode they come back as
permission errors. See Connecting clusters.
Monitor charts on long ranges
A history chart draws one point per interval, and the interval grows with the range:
| Range | One point every |
|---|---|
| 5m, 15m, 1h | 15 or 30 s |
| 6h | 2 min |
| 24h | 10 min |
| 7d | 1 h |
| 14d | 2 h |
| 30d | 6 h |
From 6h up, the CPU and memory charts of pods, workloads and nodes draw the highest value inside each interval, read once a minute, so a spike between two points still shows. The stats beside those charts change names to match:
max— the range’s true maximum.avg peakandmin peak— the average and the lowest of the intervals’ peaks. They read higher than typical load, on purpose.last— the peak of the last interval, not the value now.
Kobi’s get_workload_metrics reads the same peaks and says when the maximum
happened.
Network charts are not peaked: each point is the rate over the minute before it, so on a long range a burst between two points doesn’t show.
Size a request from typical use (the rightsizing
P95) and a memory limit’s headroom from max.
A pod that runs with hostNetwork shares its node’s network, so it has no
traffic of its own to chart. Its Monitor tab shows Uses the node’s network
with a link to each node’s Monitor tab, and so does a workload whose pods run
that way. Since agent 1.4.8 such a pod sends no pod network at all; its traffic
is counted once, on the node. See Compatibility.
Revealing a Secret value
The YAML view never shows Secret values. When you actually need one — validating
a rotation, say — there is a separate, deliberately narrow path instead of
dropping to kubectl get secret -o yaml | base64 -d:
POST /api/v1/resources/secrets/{namespace}/{name}/reveal
It is designed to be auditable rather than convenient:
POST, notGET. A reveal returns sensitive payload and emits a side effect.GETis cacheable by every layer in between;POSTis not.- Read from the apiserver, not the informer cache. A reveal has to show the post-rotation value, not a cached one, so it costs one apiserver round-trip.
- A written reason is required, between 10 and 500 characters. It goes into the audit record, not into the response.
- Editor or above, enforced at the route. For namespaces matching the
production pattern (
^(prod|production|prd)([-_].+)?$by default, editable in Administration → System → General or viaKUBEBOLT_PROD_NAMESPACE_PATTERN) the handler escalates the requirement to Admin, and a denied attempt is audited too. - Two audit records per reveal — one on the general action log, next to
restarts and scales, and one on a dedicated secret-reveal channel carrying the
reason, the requested keys, the actor, and the outcome
(
success·denied_prod_namespace_requires_admin·not_found·apiserver_error). - Neither record ever contains a value, a hash of a value, or anything derived from one. Values exist only in the response body and are never persisted server-side.
Request body: {"keys": ["…"], "reason": "…"}. Omitting keys reveals every
key and audits the breadth as <all>. Values that are not printable UTF-8 come
back as a sha256 plus a byte count rather than as garbled text.