Documentation
Troubleshooting
The things that usually turn out to be missing.
A cluster has stopped answering
A banner names the cluster, how long it has been quiet and what went wrong: timed out, connection refused, certificate not trusted, address did not resolve, or not authorised. The rows already on screen are kept and marked with when they were read, so nothing silently ages into looking current. Reconnect from the banner once the cause is fixed.
No usage numbers in the tables
Current CPU and memory come from metrics-server. If the cluster has none installed, those columns stay empty.
No history on the charts
History needs a Prometheus. If discovery does not find one, point at it with --prometheus namespace/service:port. Without one, Spinoza samples metrics-server while the window is open and labels the chart accordingly.
Rollback or uninstall is refused
Those two call the helm binary. Put helm on PATH, or point --helm at it.
A distroless pod will not open a shell
Spinoza falls back to an ephemeral debug container, which it creates with kubectl. Put kubectl on PATH, or point --kubectl at it.
The node shell is not offered
It is off unless you start with --node-shell, and it is hidden when the apiserver says your identity may not create the pod it would need.
The cluster is a version Spinoza does not claim
Spinoza is built on client-go v0.36, so Kubernetes 1.35 to 1.37 are inside the supported skew. Outside it, discovery may report types the client cannot serialise.