Synopsis
Reads a plane’s nodes and pods and reports load by pool, with whatever recommendations the measurement actually supports. REQUESTS, not utilisation. The scheduler packs against requests, so a pool is full when its requests fill it however idle the CPUs are — and utilisation needs metrics-server, which a plane is not guaranteed to have. The production plane did not. A plane that has never run a build cannot say what a build costs, and this says so rather than doing arithmetic on an idle cluster. That is not a hypothetical: on 2026-09-11 the live plane’s peak CPU across twenty-two hours never reached 5%, because no build had ever run on it.Options
Global options
See global options.SEE ALSO
- kubox plane - Describe and build a management plane from one document