Skip to content

Commit aa36d3f

Browse files
fix(kubernetes)!: remove the talos block and other leftovers of the worker pool split (#5191)
## What this PR does Cleanup after the Kubernetes app was split into `Kubernetes` (control plane and addons) and `KubernetesNodes` (one release per worker pool): what the split left behind in values, docs, templates and tests. ### Kubernetes The `Kubernetes` application has no `talos` field any more. Worker pools are `KubernetesNodes` resources and take their Talos release, image, installer and registry mirrors from their own `talos` and `osImage`, so the cluster's copies did nothing. The last one, `talos.version`, fed a duplicate of the Talos to Kubernetes support-matrix check. The pool chart already runs that check against the Talos release each pool actually boots, including one picked through an `osImage` override; the cluster's copy compared the Kubernetes version against a release no node runs. A stored spec that still carries `talos.*` keeps rendering, because the values schema does not close the object and the API does not validate `spec` against it, and a unit test pins that. The pool's check now has render coverage. Its own suite pins the pairings it must accept, and a bats test renders the refusals through `talos.version`, `osImage.factory.version` and `osImage.builtin.version` with schema validation skipped, since the schema offers no Kubernetes version the matrix refuses. I fixed the README and field descriptions that described behavior the chart no longer has, or never had: a dedicated etcd cluster, a `blockSize.matchVolume` setting that does not exist, a default `md0` pool, the host default, the substrate mismatch behavior, and rollout phase names. The README's Breaking Changes section becomes a pointer to the per-release upgrade notes on cozystack.io. Unused helpers and stale comments are removed; the chart renders the same bytes. ### KubernetesNodes Five field descriptions and the `talos` block told users to keep them in sync with parent fields that do not exist. NOTES warned about colliding with a parent-managed `md0` pool, although the parent renders no pools and `md0` is the normal name of a pool. The installer image was credited to an upgrade Job that never runs one. The values descriptions, README intro and NOTES now say what the chart renders today, including which fields have no effect on `substrate: proxmox`: `storageClass`, `diskSize`, `instanceType`, `talos.imageFactoryURL` (beyond its format check) and `talos.imageFactoryCA`. The render error for a pool whose MachineDeployment belongs to another release told users to remove the pool from the parent's `nodeGroups`, which cannot clear the conflict. It now names a remedy that works. Split-era history is gone from Helm and test comments, along with two `storageClass` fallbacks that can only return the value they fall back from, a constant variable, and two `.helmignore` rules that match nothing. The tenant pod and service CIDRs are literals in both charts, and nothing checked that they agree. A new bats test renders both charts and fails when they differ. No rendered manifest of the pool chart changes. I rendered it for the snapshot values, GPU, kubelet-override, factory/CA, storageClass variants and three Proxmox variants on main and on every commit, and the output is byte-identical. The snapshot and TalosConfigTemplate hash tests pass unchanged, so no worker rolls. YAML comments inside the hashed machine config and machine template are left as they are for that reason. ### The rule `docs/agents/overview.md` gains a rule so this does not come back with the next change: app READMEs and field descriptions describe current behavior, a change users must act on goes to the website's upgrade page, and a description does not ask the reader to keep it in sync with another chart. ```release-note [kubernetes] The Kubernetes application no longer has a `talos` field. Talos settings for workers, and the Talos to Kubernetes compatibility check, live on each KubernetesNodes pool. Existing clusters that still store `talos` values keep working; the values are ignored. KubernetesNodes field descriptions, README and install notes now describe the worker pool as it works today, including which fields have no effect on Proxmox pools. ``` <!-- This is an auto-generated comment: release notes by coderabbit.ai --> <!-- coderabbit-ai-generated:v1 --> ## Summary by CodeRabbit * **Configuration** * Worker OS image, installer, and registry-mirror settings are configured per worker pool; these settings are no longer available at the cluster level. * Kubernetes and Talos version compatibility is checked for each worker pool. A worker Kubernetes version cannot be ahead of its parent cluster. * Worker pools support KubeVirt and Proxmox, with pool-specific sizing and disk settings. * Rendering is refused when a pool uses a different substrate from its cluster or conflicts with an existing worker pool owned by another release. * The default cluster hostname now includes the `kubernetes-` prefix. * Existing cluster configurations that still include Talos settings receive a warning; those settings are no longer used. * **Documentation** * Updated guidance for cluster and worker-pool settings, storage, upgrades, and version compatibility. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
2 parents 4f7eb63 + 0fed7c5 commit aa36d3f

53 files changed

Lines changed: 686 additions & 713 deletions

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎api/apps/v1alpha1/kubernetes/types.go‎

Lines changed: 3 additions & 24 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

‎api/apps/v1alpha1/kubernetes/zz_generated.deepcopy.go‎

Lines changed: 0 additions & 17 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

0 commit comments

Comments
 (0)