Rename and revisit _internal.sylva_namespaces computation and tests (!8357) · Merge requests · Sylva-projects / sylva-core

What does this MR do and why?

What

What I'm proposing as actual changes to current behavior, on top of the _internal value rename is:

keep static list part of ._internal.sylva_namespaces_not_labeled_by_flux for only operator-owned namespaces and avoid including K8s initial namespaces like default & kube-system
have the python test exempt known (k8s initial, rancher/fleet ad-hoc, w-c-o) namespaces and thus allow the test to check that all others have the sylva.org/sylva-managed: "true" label.

If you think the default & kube-system need to be kept in ._internal.sylva_namespaces_not_labeled_by_flux I can revert this.

Why

Learnt during this MR's initial state review that ._internal.sylva_namespaces is a small complement to all namespaces (manifests) deployed by Flux (getting sylva.org/sylva-managed: "true" labels directly via commonMetadata.labels), covering only HelmRelease.spec.targetNamespace from units with HelmRelease.spec.install.createNamespace: true and the static list of operator-created (CNI and Rancher only AFAICS) namespaces, not a derivation of "every ns sylva-units deploys".

What it does

Renames the ._internal.sylva_namespaces to ._internal.sylva_namespaces_not_labeled_by_flux to make that contract explicit and adds a doc comment to mention commonMetadata.labels implicit injection by Flux controllers.
Tries to improve this ._internal.sylva_namespaces_not_labeled_by_flux (currently feeding the namespace allow-list consumed by the sylva-namespaces-label Kyverno policy) so it is more accurate by:

2.1 conditioning the namespace generated by Calico CNI (vs Cilium which is deployed in kube-system, not an operator-created ns, but part of the https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/#initial-namespaces), dropping tigera-operator ns from the static list because it seems it's handled by Flux directly:
```
capo-mgmt|default] ☸ capo-mgmt
➜  k get ns tigera-operator --show-managed-fields -o json \
  | jq -r '.metadata.managedFields[]
           | select(.fieldsV1["f:metadata"]["f:labels"]["f:sylva.org/sylva-managed"])
           | .manager'
helm-controller

[capo-mgmt|default] ☸ capo-mgmt 
➜  k get ns calico-system --show-managed-fields -o json \
  | jq -r '.metadata.managedFields[]
           | select(.fieldsV1["f:metadata"]["f:labels"]["f:sylva.org/sylva-managed"])
           | .manager'
background-controller
```
2.2 removing kube-system from the static list due to the fact it's part of https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/#initial-namespaces and I believe the static part of ._internal.sylva_namespaces_not_labeled_by_flux should be for operators-owned ns only. Also, until now it was labeled by Flux but only in some situations, mainly clusters where local-path-provisioner was not specifically disabled, due to local-path-provisioner unit having it as HelmRelease.spec.targetNamespace, with HelmRelease.install.createNamespace: true. Since kube-system is already present on the cluster, I took the decision to drop units.local-path-provisioner.helmrelease_spec.install.createNamespace: true

2.3 similarly to kube-system, the default namespace is also part of https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/#initial-namespaces, so I dropped it from the static list following the same reasoning.

2.4 removing cattle-resources-system as it's created by Flux and labeled via commonMetadata.labels, unit rancher-backup-restore-operator-init

2.5 removing cattle-dashboards as it's created by Flux and labeled via commonMetadata.labels, from monitoring unit's rancher-monitoring helm template, given default chart values:
```
defaultDashboards:
  namespace: cattle-dashboards
  useExistingNamespace: false
```
Note 1: the local namespace is Rancher-generator, so it remains in the static list (dedicated to operators-owned ns):
```
[capo-mgmt|default] ☸ capo-mgmt
➜  kubectl get ns local -o jsonpath='{.metadata.ownerReferences}' | jq .
[
  {
    "apiVersion": "management.cattle.io/v3",
    "kind": "Cluster",
    "name": "local",
    "uid": "1c8fa3c8-daa2-459a-8c43-b316a12e4187"
  }
]
```
Note 2: even though it is an operator-owned ns, we're not considering workload-cluster-operator per-workload cluster namespaces (<team>--<cluster>) because we can't predict their names, since team/cluster definitions can arrive via the workload-teams-repo GitOps path, so that would not be available for s-u at render time.
Refactors test_sylva_namespaces_are_labeled test, which was before expected to fail (due to https://docs.pytest.org/en/stable/how-to/skipping.html#xfail-mark-test-functions-as-expected-to-fail use) due to existing Sylva CI namespaces that can't be expected to be labelled
- K8s initial namespaces: default, kube-system,kube-public,kube-node-lease
- Rancher and Fleet ns: c-, p-, c--p-, cluster-fleet-*
- workload-cluster-operator ns for each workload cluster

We now, post refactoring, have exemptions for the above 3 items, and are able to determine a real miss in our ._internal.sylva_namespaces_not_labeled_by_flux logic if any namespace is found missing sylva.org/sylva-managed: "true" label.

Additionally, we introduce another test_expected_namespaces_carry_label test that is checking the 2 labelling paths (Flux's own commonMetadata.labels and the sylva-namespaces-label Kyverno policy) work, by confirming the sylva.org/sylva-managed: "true" label presence when the ns is either:

managed by Flux - the unit is Kustomization/HelmRelease (ns would have the another label key like kustomize.toolkit.fluxcd.io/name or helm.toolkit.fluxcd.io/name)
listed in _internal.sylva_namespaces_not_labeled_by_flux

Test coverage

CI configuration

Below you can choose test deployment variants to run in this MR's CI.

Click to open to CI configuration

Legend:

Icon	Meaning	Available values
☁️	Infra Provider	`capd`, `capo`, `capm3`
🚀	Bootstrap Provider	`kubeadm` (alias `kadm`), `rke2`, `okd`, `ck8s`
🐧	Node OS	`ubuntu`, `suse`, `na`, `leapmicro`
🛠️	Deployment Options	Deployment option list and description
🎬	Pipeline Scenarios	Available scenario list and description
🟢	Enabled units	Any available units name, by default apply to management and workload cluster. Can be prefixed by `mgmt:` or `wkld:` to be applied only to a specific cluster type
🔴	Disabled units	Any available units name, by default apply to management and workload cluster. Can be prefixed by `mgmt:` or `wkld:` to be applied only to a specific cluster type
🏗️	Target platform	Can be used to select specific deployment environment Available platform list and description
⚡	Pipeline control	`autorun`, `manual` or `blocking`. Can be used to override global config and start a deployment pipeline the required way

Global config for deployment pipelines

autorun pipelines
allow failure on pipelines
record sylvactl events

Notes:

Enabling autorun will make deployment pipelines to be run automatically without human interaction
Disabling allow failure will make deployment pipelines mandatory for pipeline success.
if both autorun and allow failure are disabled, deployment pipelines will need manual triggering but will be blocking the pipeline

Be aware: after configuration change, pipeline is not triggered automatically. Please run it manually (by clicking the run pipeline button in Pipelines tab) or push new code.

Edited Sep 21, 2026 by Bogdan-Adrian Burciu

Rename and revisit _internal.sylva_namespaces computation and tests