Kubernetes problems new hires hit in the first 90 days
Most Kubernetes issues show up as a pod status. Learning to read kubectl describe, events and previous container logs is the first skill, and the second is knowing which layer, from image to manifest to cluster, caused it.
- CrashLoopBackOff: the real error is usually in kubectl logs --previous, often a missing config value, a bad command or a failing dependency at startup.
- ImagePullBackOff and ErrImagePull: wrong tag, private registry without imagePullSecrets, or a node that cannot reach the registry.
- Pending pods: insufficient CPU or memory requests, taints without tolerations, node selectors or affinity that no node satisfies, and unbound PersistentVolumeClaims.
- OOMKilled containers and CPU throttling caused by limits that do not match real usage.
- Probes: liveness probes that restart a slow-starting app, readiness probes pointing at the wrong path or port.
- RBAC: "forbidden: User cannot list resource" for a service account missing a Role or RoleBinding in that namespace.
Helm, GitOps and networking
Most teams deploy with Helm, Kustomize or both, through Argo CD or Flux. Early tickets include working out which values file wins, why helm upgrade failed with an immutable field error, and why Argo CD shows an app as OutOfSync after a manual kubectl edit. Job Support helps you render templates locally and read the diff before you change anything.
Networking is the other big area: Services that select no pods because of a label typo, an ingress controller returning 404 or 502, TLS certificates from cert-manager stuck in a pending state, and NetworkPolicies quietly blocking traffic.
What a Kubernetes support session looks like
Rahul joins an Infosys platform team running EKS clusters for a US logistics client. During his first on-call week a payments service goes into CrashLoopBackOff after a release. With Job Support open on the Zoom bridge, he checks previous logs, finds the new version expects an environment variable that the Helm values for production never set, and proposes adding it through the chart rather than patching the deployment by hand.
The live transcript helps him follow the incident commander's questions, and he writes the post-incident notes himself.
Using Kubernetes job support responsibly
Clusters run production workloads, so keep control and follow your team's process.
- Never share kubeconfig files, cluster tokens, cloud credentials or Secret values.
- Redact namespaces, hostnames and environment values if your client considers them sensitive.
- Apply changes yourself through the approved GitOps or change process, not ad hoc kubectl edits in production.
- Follow your employer's policy on sharing manifests and charts externally.
Kubernetes job support pricing and how to start
Job Support costs $5 per hour with taxes included. Each hour includes 200 AI credits plus support, and hours and credits never expire, which suits unpredictable on-call weeks. Pay by Stripe or bank transfer. Refunds apply only if the tool does not work.
The free plan includes 5 live Copilot minutes and 100 AI credits, no card needed. Install the Windows or macOS desktop app before your next on-call shift.