cmd/operator-sdk: add --crd-version flag to relevant commands #2684
Conversation
|
I think it can close: #2579 |
joelanford
force-pushed
the
crd-versions
branch
from
5b33d64 to
c249e7b
Compare
openshift-ci-robot
added
the
needs-rebase
joelanford
force-pushed
the
crd-versions
branch
from
c249e7b to
50c3e0e
Compare
openshift-ci-robot
removed
the
needs-rebase
| err error | ||
| ) | ||
| switch g.crdVersion { | ||
| case "v1beta1": |
There was a problem hiding this comment.
Since we're going to switch to v1.CustomResourceDefinition by default in the future, we could set crd := apiextv1.CustomResourceDefinition{} above and return a apiextv1.CustomResourceDefinition from newCRDForResource(), then convert v1 -> v1beta. We'd also have to do some conversion if the file exists on disk and is v1beta1.
There was a problem hiding this comment.
Yeah, the reason I didn't do that was because of the possibility of a v1beta1 CRD on disk. v1 moved several top level fields from v1beta1 into the versions items, so it is not possible to unmarshal the v1beta1 YAML into a v1 struct.
However, the v1beta1 struct is a superset of the v1 struct, so it is possible to unmarshal v1 YAML into a v1beta1 struct and then convert back.
There was a problem hiding this comment.
Is it possible to use the generic apiextensions struct for both?
There was a problem hiding this comment.
If I properly understood the suggestion here, IHMO is better we keep each version implementation since in the future we will just remove the v1beta1 and keep all generating v1 only.
There was a problem hiding this comment.
Is it possible to use the generic apiextensions struct for both?
I can't get that to work either. For some reason (still trying to figure out why) everything gets marshaled without having the correct YAML layout (e.g. in the output there's actually an objectMeta key containing the GVK info, instead of that being inlined at the root).
There was a problem hiding this comment.
I’ve had trouble marshaling the generic CRD struct before. The current approach is fine then.
There was a problem hiding this comment.
👍
Once we bump to v1.19, we'll have to basically revert this PR and change from v1beta1 to v1. This function will basically go back to how it looked before, except with the v1 types.
joelanford
force-pushed
the
crd-versions
branch
from
50c3e0e to
34031f5
Compare
joelanford
force-pushed
the
crd-versions
branch
from
34031f5 to
9db5a51
Compare
Description of the change:
Add CRD version flag to relevant subcommands that enables users to generate CRDs with either
v1beta1orv1.v1beta1is still the default until OLM supports operator bundles withv1CRDs.Motivation for the change:
CRD
v1beta1is deprecated and will be removed in Kubernetes 1.19, so we need to give users time to begin migrating.Closes #2579