Documentation
¶
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type DepProvider ¶
type DepProvider struct {
// contains filtered or unexported fields
}
DepProvider is a dependency provider.
The instances it returns are either
- old implementations backed by k8sdeps code,
- new implementations backed by kyaml code.
History:
kubectl depends on k8s.io code, and at the time of writing, so does kustomize. Code that imports k8s.io/api* cannot be imported back into k8s.io/*, yet kustomize appears inside k8s.io/kubectl.
To allow kustomize to appear inside kubectl, yet still be developed outside kubectl, the kustomize code was divided into the following packages
api/ k8sdeps/ (and internal/ks8deps/) ifc/ krusty/ everythingElse/
with the following rules:
Only k8sdeps/ may import k8s.io/api*.
Only krusty/ (and its internals) may import k8sdeps/. I.e., ifc/ and everythingElse/ must not import k8sdeps/ or k8s.io/api*.
Code in krusty/ may use code in k8sdeps/ to create objects then inject said objects into everythingElse/ behind dependency neutral interfaces.
The idea was to periodically copy, not import, the large k8sdeps/ tree (plus a snippet from krusty/kustomizer.go) into the kubectl codebase via a large PR, and have kubectl depend on the rest via normal importing.
Over 2019, however, kubectl underwent large changes including a switch to Go modules, and a concerted attempt to extract kubectl from the k8s repo. This made large kustomize integration PRs too intrusive to review.
In 2020, kubectl is based on Go modules, and almost entirely extracted from the k8s.io repositories, and further the kyaml library has a appeared as a viable replacement to k8s.io/api* KRM manipulation code.
The new plan is to eliminate k8sdeps/ entirely, along with its k8s.io/api* dependence, allowing kustomize code to be imported into kubectl via normal Go module imports. Then the kustomize API code can then move into the github.com/kubernetes-sigs/cli-utils repo. The kustomize CLI in github.com/kubernetes-sigs/kustomize and the kubectl CLI can then both depend on the kustomize API.
So, all code that depends on k8sdeps must go behind interfaces, and kustomize must be factored to choose the implementation.
That problem has been reduced to three interfaces, each having two implementations. (1) is k8sdeps-based, (2) is kyaml-based.
ifc.Kunstructured
1) api/k8sdeps/kunstruct.UnstructAdapter
This adapts structs in k8s.io/apimachinery/pkg/apis/meta/v1/unstructured to ifc.Kunstructured.
2) api/wrappy.WNode
This adapts sigs.k8s.io/kustomize/kyaml/yaml.RNode to ifc.Unstructured.
At time of writing, implementation started. Further reducing the size of ifc.Kunstructed would really reduce the work (e.g. drop Vars, drop ReplacementTranformer).
resource.ConflictDetector
1) api/internal/k8sdeps/conflict.conflictDetectorJson api/internal/k8sdeps/conflict.conflictDetectorSm
Uses k8s.io/apimachinery/pkg/util/strategicpatch, apimachinery/pkg/util/mergepatch, etc. to merge resource.Resource instances.
2) api/internal/conflict.smPatchMergeOnlyDetector
At time of writing, this doesn't report conflicts, but it does know how to merge patches. Conflict reporting isn't vital to kustomize function. It's rare that a person would configure one transformer with many patches, much less so many that it became hard to spot conflicts. In the case of an undetected conflict, the last patch applied wins, likely what the user wants anyway. Regardless, the effect of this is plainly visible and usable in the output, even if a conflict happened but wasn't reported as an error.
ifc.Validator
1) api/k8sdeps/validator.KustValidator
Uses k8s.io/apimachinery/pkg/api/validation and friends to validate strings.
2) api/internal/validate.FieldValidator
See TODO inside the validator for status. At time of writing, this is a do-nothing validator as it's not critical to kustomize function.
Proposed plan:
[ ] Ship kustomize with the ability to switch from 1 to 2 via
an --enable_kyaml flag.
[ ] Make --enable_kyaml true by default.
[ ] When 2 is not noticeably more buggy than 1, delete 1.
I.e. delete k8sdeps/, transitively deleting all k8s.io/api* deps.
This DepProvider should be left in place to retain these
comments, but it will have only one choice.
[ ] The way is now clear to reintegrate into kubectl.
This should be done ASAP; the last step is cleanup.
[ ] With only one impl of Kunstructure remaining, that interface
and WNode can be deleted, along with this DepProvider.
The other two interfaces could be dropped too.
When the above is done, kustomize will use yaml.RNode and/or KRM Config Functions directly and exclusively. If you're reading this, plan not done.
func NewDefaultDepProvider ¶
func NewDefaultDepProvider() *DepProvider
func NewDepProvider ¶
func NewDepProvider(useKyaml bool) *DepProvider
func (*DepProvider) GetConflictDetectorFactory ¶
func (dp *DepProvider) GetConflictDetectorFactory() resource.ConflictDetectorFactory
func (*DepProvider) GetFieldValidator ¶
func (dp *DepProvider) GetFieldValidator() ifc.Validator
func (*DepProvider) GetKunstructuredFactory ¶
func (dp *DepProvider) GetKunstructuredFactory() ifc.KunstructuredFactory
func (*DepProvider) GetResourceFactory ¶
func (dp *DepProvider) GetResourceFactory() *resource.Factory
Source Files