Pods

The live pod list, and a pod's drawer for its logs, terminal, graphs, resizing, events and YAML.

Pods (under Workloads in the sidebar) lists every pod in the namespaces you’ve picked, as it changes.

The list

Columns: Name, Namespace, Ready, Status, Owner, Restarts, Node, IP, CPU, Memory and Age. Owner is the pod’s controller from its ownerReferences, as Kind/name linking to that object’s drawer: a Deployment’s pods show the Deployment (read through the ReplicaSet), others show their StatefulSet, DaemonSet, Job, a custom resource such as an Argo Rollout, or the Node for a static pod. A pod nothing controls shows —. Filter it with controller:statefulset or controller:web. CPU and memory show a bar against the pod’s limit, or its request when it has no limit. Filter and sort as on any table. For example, status:crash finds crashing pods, and node:worker-2 finds the pods on one node.

Status is a coloured pill, and the word always stays on it. Green is Running (or Ready), blue is Pending or still starting, purple is Succeeded or Completed, orange is a warning such as Running with containers not ready or Evicted, red is a failure such as Failed, CrashLoopBackOff or ImagePullBackOff, struck-through grey is Terminating, and plain grey is a status Kubebay doesn’t recognise. Other tables use the same colours for their status words.

The ⋮ at the end of a row has View details, Logs, Shell, Edit YAML, Copy name and Delete.

Show pods in the row menu of a Deployment, StatefulSet, DaemonSet, ReplicaSet or Job lists just that workload’s pods. A bar above the table names the selector. Show all pods goes back to the full list.

A pod’s drawer

Click a pod to open its drawer. It always opens on Summary, unless you chose a tab from the row menu.

Summary

The pod’s metadata (including its Pod IP and Host IP, or Pod IPs and Host IPs on a dual-stack cluster), conditions, each container (image, state, ports, environment), resource claims and volumes.

  • Owned by shows the whole chain of owners, each one a link. For example: Deployment → ReplicaSet → pod. An owner that is a custom resource, such as an Argo Rollout, opens like any other owner. When the pod, or a Deployment, StatefulSet or other owner above it, was installed by Helm (it carries the app.kubernetes.io/managed-by: Helm label), the chain ends in Helm release . Click it to open that release’s drawer on the Helm page. The name comes from the meta.helm.sh/release-name annotation, or the app.kubernetes.io/instance label when the annotation isn’t there; the namespace comes from meta.helm.sh/release-namespace. It is read from labels, not from the release records, so a release that was uninstalled, or whose objects were copied from another one, can show a link that finds nothing.
  • Why this pod is not healthy. A pod that is not healthy gets a card at the top of Summary with a short diagnosis, the field it was read from, and up to three next steps that jump to the Logs tab (on the previous container), the Events tab, the YAML tab, or the PVC or node page. It is plain rules, with no AI and no extra requests: it reads the pod’s status and the events Kubebay already streams. It recognises an evicted pod, a node that is not ready, OOMKilled (with the exit code and memory limit), ImagePullBackOff, CreateContainerConfigError, CrashLoopBackOff (with the last exit code), an unbound PersistentVolumeClaim, an unschedulable pod (with the scheduler’s message), a failing probe and a container that exited non-zero. Where it infers rather than reads, it says “likely”. If nothing matches it says no known cause was found, and a healthy pod, or one that is just starting, shows no card.
  • An environment variable from a Secret stays hidden until you click to reveal it.
  • Click a container’s port to forward it to your machine. See Port forwarding.

Logs

Logs stream as the container writes them.

  • Container picks which container to show. It starts on the container the pod’s kubectl.kubernetes.io/default-container annotation names, as kubectl logs does, and otherwise on the first container. The Terminal tab uses the same choice.

  • tail sets how many earlier lines to load.

  • follow keeps the newest line in view.

  • previous shows the run before the last restart, which is where a crash’s cause usually is.

  • Filter lines… narrows what’s shown, and Download saves the lines as a .log file.

  • Lines are coloured by their level: errors red, warnings amber, success green, debug and trace dimmed. Kubebay reads the level from the line itself:

    • a JSON field such as "level":"error", including pino’s numeric levels;
    • level=warn in logfmt;
    • a klog prefix such as E1112;
    • [error] in brackets, or ERROR in capitals;
    • an exception at the start of the line, such as Error:, panic: or Traceback.

    A line with no level that says something failed is an error, and one that says something succeeded is a success. “error” in the middle of a sentence doesn’t count. A line that carries its own colours (ANSI codes) shows them in the theme’s colours rather than as escape codes. The filter matches the text you see, and Download keeps the lines as the container wrote them.

Terminal

A shell in the container: pick the container and the shell (auto, bash, sh or ash). auto tries bash, then sh, then ash. The session ends when you close the drawer.

Debug with ephemeral container is for a container with no shell (a distroless image). It adds a debug container to the running pod, the same call as kubectl debug, and opens a shell in it. Pick the Image (it starts as the Debug image in Settings) and the Target container, whose processes the debug container can see. When the cluster’s name looks like production, you click twice to confirm. The new container is selected in Container and its shell opens at once.

  • An ephemeral container can’t be removed or changed. It stays in the pod’s spec until the pod is replaced, but it stops when you exit its shell.
  • It needs the pod to be Running and your identity to be allowed to update pods/ephemeralcontainers (and pods/exec for the shell). A refusal is shown in the dialog.
  • It is recorded in the Audit Log as debug-container, with the container name, image and target.
  • Needs Kubernetes 1.25 or newer. If the image is slow to pull, the terminal may say the container is still starting; switch tabs and back to retry.

Graphs

CPU and memory per container over the last 15m or 24h, from Prometheus. If Prometheus runs in the cluster, Kubebay finds it by itself and reads it through the cluster’s API server, so the graphs just work. The tab says Auto-detected: monitoring/prometheus-server:80, and Change picks another service or switches to your own URL. Only when nothing is found, and no URL is set, does the tab say so and link to Settings. A URL you set always wins, and if that Prometheus is unreachable Kubebay shows the kubectl port-forward command to reach it.

Size

Change a container’s CPU and memory requests and limits. On Kubernetes 1.33 and later this resizes the running pod in place. Only the fields you fill in are changed, and each container’s resize policy decides whether it restarts. Use suggested fills in a right-sizing suggestion when there is one. Copy as patch gives you the change for kubectl patch.

Vulnerabilities and Signatures

  • Vulnerabilities counts the CVEs Trivy-Operator found in the pod’s images, and Open on the Vulnerabilities page shows them all. See Vulnerabilities.
  • Signatures checks whether the pod’s images are signed. The check contacts each image’s registry, so it runs only when you ask. Choose anonymous access, or the pod’s own pull secrets.

Triage

Asks an AI model what’s wrong with the pod. It sends the pod’s evidence: status, events and recent logs. The tab appears only once you’ve turned triage on for this cluster. See AI triage.

Events and YAML

Events lists the pod’s recent events. YAML is the live object. You can edit it, dry-run the change, and apply it. See Resource details.

Deleting a pod

Delete asks for a plain confirmation that names the pod: click Delete to go ahead or Cancel to keep it. You don’t type the name; pods are recreated by their controller, and every other kind still asks you to type its name. A pod that’s finished, or never scheduled, is deleted at once. If a finalizer holds it, such as a Job’s tracking finalizer, Kubebay offers to force the delete and says exactly what that does: skip the grace period, remove finalizers, or both.