Docs / Checks and insights
Checks and insights
Certificates, best practices, upgrade readiness, vulnerabilities, rightsizing, cost, permissions, image updates and connection tests.
Each of these runs across every open context.
Certificates
Certificates lists everything that expires, soonest first:
- TLS Secrets (only the public certificate is read, never the key);
- cert-manager Certificates, with their renewal time and readiness;
- the CA bundles of admission webhooks, API services and CRD conversion webhooks (an expired one breaks the cluster quietly);
- your kubeconfig's own client certificates, cluster CAs and bearer-token expiry, read on this computer.
Anything expired, or due within 14 days, needs action; within 30 days it's a warning. Certificates that cert-manager renews only count within 7 days, when renewal is probably failing. The Health page shows a line when something needs attention.
Best practices
Best practices checks the workloads (Deployments, StatefulSets, DaemonSets, CronJobs, Jobs and standalone Pods) for common problems, by level:
- Danger: privileged containers, sharing the Node's network, processes or IPC.
- Warning: no CPU or memory requests, no memory limit, no readiness probe on a container that serves ports, images without a fixed version, Node directories, broad Linux capabilities, and a disruption budget that blocks every eviction and so hangs Node drains.
- Info: no liveness probe, may run as root or gain privileges, a writable root filesystem, a single replica.
Each finding says what to change. System namespaces are left out unless you ask. New resource runs the same checks on a manifest before you apply it.
Upgrade check
Before upgrading Kubernetes, pick the target version and see what in each context still uses an API it removes (from 1.16 to 1.32), with the version to use instead.
The API server stores objects at current versions, so Klystiq looks where old ones show up:
- Helm release manifests (the release's next upgrade would fail);
- kubectl's last-applied record (the next apply of that file would fail);
- the API version each tool last wrote with, from managed fields (leaving out the cluster's own controllers).
Manifests in Git or CI aren't checked; use Local manifests for those.
Vulnerabilities
Vulnerabilities lists the workloads with the images their Pod templates run, scanned with your own Trivy or Grype, whichever is on your PATH (pick one if you have both). Scan everything shown, one image at a time, or one workload (Scan images in its actions menu).
Each workload shows its known vulnerabilities by severity. Select it to list each image's findings with the installed and fixed versions, and filter by severity, fixable only, or text. The scanner pulls image data with your registry credentials and keeps its own database. Nothing is installed in the cluster, and results are kept until Klystiq quits.
Rightsizing
The Rightsizing tab on the Resources page suggests requests from the last 7 days of each cluster's Prometheus, per container of every Deployment, StatefulSet and DaemonSet:
- CPU from the 95th percentile of the 5-minute rate, plus 15%;
- memory from the peak working set, plus 15%;
- a limit raised where it would be below the new request.
Pods that were replaced during the week count too. Changes under 25% are left out unless you ask. Each suggestion shows the CPU and memory freed or added over all replicas and, with cost turned on, the monthly difference. Review… opens the change as YAML in New resource (dry run and diff first); Review all shown… does the same for a context's whole list; Ask AI asks the assistant for a second opinion. A workload that Helm, Argo CD or Flux manages says so, since its next sync would put the old values back.
Cost
Turn on Settings → Cost and the Resources page shows what each cluster's requested CPU and memory cost per month at your prices (per vCPU-hour and GiB-hour, for every context or some; OpenCost's on-demand list prices to start), how much of it sits idle right now, and a Cost by namespace tab.
When a cluster runs OpenCost or Kubecost, Klystiq also asks its API through the service proxy for the last 7 days per namespace, and shows that, projected to a month, beside your estimate. Estimates cover requests only: no discounts, storage, network or Node overhead.
Permissions
Permissions (in the sidebar, or Show permissions on a ServiceAccount) answers three questions:
- My access: what you may do in a namespace, from the API server's own review, plus an exact "Can I …?" check.
- By subject: a ServiceAccount's, user's or group's effective rules, and the binding and role that grant each, including rules it gets through groups every ServiceAccount belongs to.
- Who can: everyone allowed a verb on a resource, and through what.
Answers come from the RBAC objects you can read. Identity-provider groups and webhook authorizers aren't visible.
Image updates
Image updates (under Overview) lists the images your workloads run, and the newest tag in their registry that's a newer patch, minor or major of the same kind: the same v prefix, number of parts and suffix, so 1.25.3-alpine is compared with 1.27.1-alpine, never with 1.27.1 or latest.
Registries are asked only when you click Check, anonymously or with your Docker credentials (docker login, or credential helpers in ~/.docker/config.json such as ecr-login or gcloud). Credentials go only to the registry they belong to. Images pinned by digest, and tags that aren't versions, are left alone.
Connection tests
Test a connection… in a Pod's actions menu asks whether that Pod can reach a Service, another Pod or any address over TCP:
- Check policies reads the NetworkPolicies as Kubernetes defines them, for the DNS lookup, leaving the Pod and reaching each target Pod, and says Allowed, Blocked or Depends, naming the policy.
- Run the probe tries it for real. An unprivileged ephemeral container in the Pod's network looks the name up with the Pod's search domains and opens a TCP connection, and its output is shown.
Ephemeral containers can't be removed: the probe stays in the Pod's spec, stopped, until the Pod is replaced. Read-only contexts allow the policy check only.