> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rafftechnologies.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Kubernetes troubleshooting

> The issues people actually hit, and how to resolve them

<sub>Updated August 24, 2026</sub>

## My PVC is stuck in Pending

The cluster has no storage nodes, so there is no default storage class. Storage nodes can only be added at cluster creation today — recreate the cluster with 2–3 storage nodes, or use [Object Storage](/products/store/object-storage) for S3-style needs.

## Scale-down says a node "could not be emptied"

A PodDisruptionBudget blocked eviction — the pool's status message names the pod. That's the platform protecting your availability: fix the PDB (e.g. `maxUnavailable: 0` with a single replica can never be drained) or scale the workload up first, then retry.

## My app went down briefly during scale-down

Workloads with **one replica** restart when their node is drained — there's no second copy to serve meanwhile. Run 2+ replicas with a PodDisruptionBudget (`minAvailable: 1`) for anything that must stay up; the drain then waits for the moved pod to be ready.

## A node shows Cordoned

Someone (or a scale-down in progress) marked it unschedulable. `kubectl uncordon <node>` reverses a manual cordon; a scale-down uncordons automatically if it aborts.

## The cluster shows "Delete failed"

Teardown hit an infrastructure error and stopped **without removing anything**, so nothing is half-deleted or wrongly billed. Click **Retry delete** — if it persists, contact support with the cluster ID.

## Creating a cluster returns 402 from the API

On a subscription account the API charges the saved payment method automatically. A `402` means the charge could not be set up — usually no active card, or a card that can't be auto-charged. Fix the payment method under Billing → Payment methods and retry, or create the cluster from the dashboard.

## kubectl says connection refused / certificate error

Re-download the kubeconfig from the cluster page (it changes if the cluster was recreated), and check your firewall allows the API — port 6443 is open through the platform baseline, but a custom security group must allow it.

## An installed app ignores my custom values

Values are passed to the Helm chart as-is; each chart has its own value names (Grafana uses `replicas`, Bitnami charts use `replicaCount`). Check the chart's documentation.