andrijapanicsb commented on PR #14223: URL: https://github.com/apache/cloudstack/pull/14223#issuecomment-6566984791
## Functional and regression testing on the final commit `a8a500f` Lab: 2 management servers + 2 KVM hosts (AlmaLinux 9.1, libvirt 8.5.0 / QEMU 7.0.0), Kubernetes 1.33.1, CloudManaged cluster 1 control + 2 workers (3 after scale-out), isolated network. Packages built from `a8a500f`, identical management jar on both servers, new classes present. Hosts have 3 cores, so offerings went up to 3 vCPU and RAM moved in 256–512 MiB steps. | Area | Tests | Result | |---|---|---| | Live increases: worker CPU-only, RAM-only, combined; control CPU-only (API) and RAM-only (UI); both node types in one call; cluster-wide offering; scale-out + offering change in one call | 13 | PASS | | Jobs executed on either management server (node access and reconciliation from MGMT_B) | – | PASS | | Same committed offering again, same-capacity offering change, stopped-cluster offering change | – | PASS: no node reads beyond one `get node`, no markers, no kubelet restart | | Node preparation only after placement: control VM live-migrated for capacity, then node read, then the single preparation patch | – | PASS | | Failure paths: control plane unreachable, node SSH unreachable, placement/capacity failure (no pending marker left), placement failure with earlier pending work (kept byte-identical), guest already above target, kubelet failure after resize (pending + CKS cordon kept, second worker untouched, retry finishes), scale timeout, operator cordon with failure, stale ownership marker, foreign marker, committed-offering retry with a mixed-offering node, VM stopped outside CKS, concurrent scale, parameter validation | 15 | PASS | | Regression: scale-out, scale-in (`size` and `nodeids`), migrate after resize, stop/start, deleted offering in totals, plain (non-CKS) VM live/stopped scale, UI totals with mixed offerings | 7 | PASS | | Autoscaling toggle | 1 | Pre-existing issue, same code on main → #14552 | In every live resize: no VM reboot, no pod restart or UID change, DNS/HTTP service checks passed, the API server answered every readiness poll during control-node resizes, cluster metadata was committed only after all nodes succeeded. Not tested: VPC tier / Netris and shared (direct-access) networks (node access changed in 306cc2e), multi-control HA, dedicated etcd, hypervisor rejection after dispatch (covered by unit tests), Kubernetes version upgrade after resizes. Found during testing, pre-existing and independent of this PR: - #14552: scale requests without an offering (cluster-autoscaler, `size`, enabling autoscaling) resize every node to an outdated cluster-wide offering - #14553: a single CKS node can be scaled from the Instances view, leaving the cluster inconsistent The Trillian smoke run on `a8a500f` is still to be done. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
