Abstract

A method is disclosed for refusing the rollout of an autoscaled set of guaranteed-quality-of-service workload instances, such as self-hosted continuous-integration runners, onto a node pool that cannot hold them. Before the rollout, a preflight derives the capacity the rollout will demand from the same declarative values file that the rollout applies: it reads the set's maximum instance count and the per-instance CPU and memory from the set's requests block, normalizes both into integer common units, and multiplies them, optionally scaling the product by a configured headroom percentage. It then sums the allocatable CPU and memory reported by every node matching a configured pool selector, and refuses when no node matches, including when the cluster cannot be read, or when either summed dimension falls below the derived demand. An existing change-window predicate is consulted first, invoked rather than re-implemented. The apply step runs the preflight, changes nothing on any non-zero result, and otherwise invokes the release tool with the identical values-file path, so the sizing rule cannot drift from the specification it guards. A verify step takes the last instance of a listing of the set, ordered by creation time on the primary label-selected path, and passes only when the quality-of-service class the orchestrator reports for it is the guaranteed class, reporting without failing when no instance is listed. Outcomes map onto distinct exit classes for pass, refusal, usage or parse error, and, in the preflight, unavailable tooling.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS