feat(ci): add Ubuntu 24.04 hydrophone canaries

[Rico Lin (ricolin)]

Depends-On: #1084

Summary

• parameterize Hydrophone guest image URLs with an image_prefix variable
• add Ubuntu 24.04 Hydrophone canary jobs for the newest generated Kubernetes version, currently v1.36.1, for both Calico and Cilium
• update DevStack and user/developer image examples to consume capo-image-elements Ubuntu 24.04 artifacts with os_distro=ubuntu

Rationale

vexxhost/capo-image-elements already publishes ubuntu-24.04-v<version>.qcow2 artifacts. This keeps the existing Ubuntu 22.04 Hydrophone matrix intact while adding a narrow Magnum-specific Noble validation path before widening the CI matrix.

Validation

• uv run --no-project --script hack/bump/kubernetes.py --self-test
• bash -n hack/bump/kubernetes.sh hack/run-integration-tests.sh
• YAML parse check for zuul.d/jobs.yaml, zuul.d/project.yaml, and zuul.d/hydrophone-jobs.yaml
• git diff --check
• uvx --python 3.12 pre-commit run --all-files
• dry run of hack/bump/kubernetes.py --image-prefix ubuntu-24.04 against temporary output files

[Rico Lin (ricolin)]

Ubuntu 24.04 validation report

Validated PR head 7b495eda8666349a2c78ed4b856ac60a3db93f36 with Ubuntu 24.04 and Kubernetes v1.36.1.

Passed:

• Baseline Ubuntu 24.04 cluster create reached CREATE_COMPLETE.
• Nodes came up Ready on Ubuntu 24.04.4 LTS with Kubernetes v1.36.1.
• Fresh create/delete validation against the PR head passed:
  • Fresh cluster reached CREATE_COMPLETE.
  • Delete completed and cleaned up the associated compute/load-balancer resources.
• Upgrade path: Ubuntu 22.04.5 LTS + Kubernetes v1.35.5 to Ubuntu 24.04.4 LTS + Kubernetes v1.36.1 passed:
  • Magnum cluster status reached UPDATE_COMPLETE.
  • Cluster template and nodegroup image moved to the Ubuntu 24.04 target.
  • Control-plane and worker machines were replaced.
  • Final nodes were all Ready on Ubuntu 24.04.4 LTS with Kubernetes v1.36.1.
  • No non-running pods after convergence.
• Upgrade path: Ubuntu 22.04.5 LTS + Kubernetes v1.36.1 to Ubuntu 24.04.4 LTS + Kubernetes v1.36.1 passed:
  • Magnum cluster status reached UPDATE_COMPLETE.
  • Control-plane and worker machines were replaced.
  • Final nodes were all Ready on Ubuntu 24.04.4 LTS with Kubernetes v1.36.1.
  • No non-running pods after convergence.
• Worker resize through the supported Magnum path passed on Ubuntu 24.04:
  • openstack coe cluster resize --nodegroup default-worker ... 2
  • Result: nodegroup stayed UPDATE_COMPLETE, second worker became Ready on Ubuntu 24.04.4 LTS.
• Worker resize down through the same supported path passed:
  • Explicit node removal returned the worker nodegroup to 1 worker.
  • Result: UPDATE_COMPLETE, remaining worker Ready, no non-running pods.
• Magnum API/conductor restart passed while Ubuntu 24.04 clusters existed:
  • API and conductor rolled successfully.
  • PR overlay still imported from the expected path after restart.
  • Cluster and nodegroup reads still returned UPDATE_COMPLETE.
• Workload networking checks passed:
  • Pod scheduling across workers.
  • CoreDNS service discovery.
  • ClusterIP Service routing.
  • Direct pod-to-pod traffic across workers.
• Service type=LoadBalancer passed with the default Octavia provider path:
  • Service reached EnsuredLoadBalancer and received a load-balancer address.
  • No non-running pods after the test.
• Cinder CSI checks passed:
  • PVC provisioned and bound with the default Cinder storage class.
  • Pod mounted the volume and successfully wrote/read data.
  • Cross-worker reattach passed: data written from a pod on one worker was read from a pod on another worker after pod deletion/recreate.
  • Test PVC/volume cleanup completed.
• Cilium smoke test passed with Ubuntu 24.04:
  • Cilium cluster reached CREATE_COMPLETE.
  • Cilium daemonset and operator were ready.
  • Nodes were Ready on Ubuntu 24.04.4 LTS with Kubernetes v1.36.1.
  • In-cluster Service connectivity succeeded.
  • No non-running pods after convergence.
• Temporary upgrade/Cilium test clusters and templates were cleaned up after validation.

Observed non-blocking issues:

• Generic openstack coe cluster update ... replace node_count=3 is not implemented in the mCAPI driver path and raises NotImplementedError from update_cluster(). The supported coe cluster resize --nodegroup ... path works and was used for successful update/resize validation.
• Service type=LoadBalancer did not complete when the cluster cloud config requested Octavia provider ovn but that provider was unavailable. Retesting with the default provider path passed, so this still looks like a provider/template config mismatch rather than an Ubuntu 24.04 regression.

Final state after validation: the retained Ubuntu 24.04 validation cluster is healthy at 1 control-plane + 2 workers, with all nodes Ready and no non-running pods from the test suite.

[Mohammed Naser (mnaser)]

lets just switch everything to 24.04

in theory, we are not testing if the images work, we already do that in capo-image-elements by deploying a cluster with them, we're only using the images as an artifact to validate.

[Mohammed Naser (mnaser)]

This should be a pinned release, we can use hack/bump to update it with time, otherwise, it will break with time.

[Mohammed Naser (mnaser)]

Same issue here