Support field selectors for CRDs #53345

ncdc · 2017-10-02T19:49:50Z

Signed-off-by: Andy Goldstein [email protected]

What this PR does / why we need it: allow field selectors to be used with custom resources

Which issue this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close that issue when PR gets merged): fixes #51046, fixes #49424

Special notes for your reviewer:

Release note:

Custom resources served through CustomResourceDefinition now support field selectors for `metadata.name` and `metadata.namespace`.

tamalsaha · 2017-10-02T20:01:28Z

Thanks @ncdc !

@sttts, can this be picked for client-go 5.0.0 release?

ncdc · 2017-10-02T21:04:39Z

@deads2k where in the apiextensions-apiserver integration tests would you recommend testing this? In testSimpleCRUD?

0xmichalis · 2017-10-02T21:40:19Z

Can you add a test in hack/make-rules/test-cmd-util.sh?

ncdc · 2017-10-02T21:43:49Z

@Kargakis I'd rather put it in the integration tests

ncdc · 2017-10-03T14:23:40Z

I just pushed an integration test in a separate commit. It's largely a copy of testSimpleCRUD but it instead creates 2 resources and tries to ensure that it only gets data for the object requested in the fieldSelector. I also verified that without the first commit in this PR, the new tests fail.

nikhita · 2017-10-03T16:13:15Z

/retest

enisoc · 2017-10-03T16:57:30Z

staging/src/k8s.io/apiextensions-apiserver/test/integration/basic_test.go

+		t.Errorf("expected %v, got %v", e, a)
+	}
+
+	listWithItem, err := noxuResourceClient.List(metav1.ListOptions{FieldSelector: "metadata.name=foo"})


Could you arrange for the objects to differ in a custom spec field, and select for that here? It would be good to test that it works for fields outside metadata.

That doesn't work. I believe it's because of

kubernetes/staging/src/k8s.io/apiextensions-apiserver/pkg/registry/customresource/strategy.go

Lines 98 to 109 in 2bf7a20

// objectMetaFieldsSet returns a fields that represent the ObjectMeta.

func objectMetaFieldsSet(objectMeta metav1.Object, namespaceScoped bool) fields.Set {

if namespaceScoped {

return fields.Set{

"metadata.name": objectMeta.GetName(),

"metadata.namespace": objectMeta.GetNamespace(),

}

}

return fields.Set{

"metadata.name": objectMeta.GetName(),

}

}

I could write some code that creates a fields.Set of every possible field leaf (excluding slices); e.g., if we have this:

a: b c: 1 d: - a - b e: f: g h: i: j

the field set would be

a: b c: 1 e.f: g e.h.i: j

I think you're right. I must have misremembered that it worked for other fields in my earlier experiment. Should we specify in the release note that CRD field selectors only support name and namespace?

Sample code that works: https://gist.github.com/ncdc/edc1ed3abefa36d9650a524ebd5f3f8c

@Kargakis I don't have any concerns with the UX of allowing arbitrary fields; that was my plan when I thought all we needed to do was stop blocking it. However, it turns out we also have to recursively flatten the whole custom object into a map, for every object returned by a list, in order to make filtering work for arbitrary fields.

That seems extreme given how conservative other objects are with their supported field selectors. For example, Pod only picks out 3 fields from spec that you're allowed to use in fieldSelector.

If you have a use case for filtering on arbitrary fields, would you be ok with having to specify a whitelist in the CRD spec, so we don't have to export every field regardless of whether anyone wants it?

We could discuss that on a separate feature request issue. Either way I think we should go ahead and merge this fix for metadata.name, since that's important for watching single objects.

Agreeing to go ahead with this PR before discussing abritrary keys in a follow-up.

@enisoc yeah, a whitelist would be great to have. Please cc me one the new issue when one is made.

@Kargakis I think it's better if you file the issue. I don't actually have any use case in mind for needing arbitrary field selectors, so I can't explain why the feature is being requested.

Opened #53459

enisoc · 2017-10-03T17:52:36Z

staging/src/k8s.io/apiextensions-apiserver/pkg/apiserver/customresource_handler.go

+}
+
+func (crdObjectConverter) ConvertFieldLabel(version, kind, label, value string) (string, string, error) {
+	// Just return the passed-in label and value, as CRDs only support a single version currently.


For other types, we return an error at this point if the field label isn't in a whitelist:

kubernetes/pkg/api/v1/conversion.go

Line 177 in 2bf7a20

return "", "", fmt.Errorf("field label not supported: %s", label)

If we don't error out, it just comes back as an empty list, which is misleading.

Whitelisting metadata.name and metadata.namespace would also mean we can remove the caveat about this only working because CRD currently supports only one version -- at least until ObjectMeta v2.

@enisoc are you saying this should just return the result of k8s.io/apimachinery/pkg/runtime/conversion.go#DefaultMetaV1FieldSelectorConversion()?

Hm the behavior there seems reasonable, although it's slightly different than what other resources do. Other cluster-scoped resources seem to error out on metadata.namespace:

kubernetes/pkg/api/v1/conversion.go

Lines 184 to 195 in 31697a2

err = scheme.AddFieldLabelConversionFunc("v1", "Node",

func(label, value string) (string, string, error) {

switch label {

case "metadata.name":

return label, value, nil

case "spec.unschedulable":

return label, value, nil

default:

return "", "", fmt.Errorf("field label not supported: %s", label)

}

},

)

We know if it's cluster-scoped or not, so I'll just write a custom function

@enisoc just pushed an update - ptal!

enisoc · 2017-10-03T19:19:35Z

/lgtm

BTW I updated the release note to specifically call out the fields that are supported.

ncdc · 2017-10-03T19:21:36Z

Ok thanks. I need to squash, so it'll need 1 more round of lgtm. Will ping in a few.

Signed-off-by: Andy Goldstein <[email protected]>

ncdc · 2017-10-03T19:22:36Z

Squashed

enisoc · 2017-10-03T19:27:16Z

/lgtm

sttts · 2017-10-04T07:17:04Z

staging/src/k8s.io/apiextensions-apiserver/test/integration/basic_test.go

+		}
+
+	case <-time.After(5 * time.Second):
+		t.Errorf("missing watch event")


is 5 seconds enough to avoid flakes?

Perhaps not, but this is not net new (I copied an existing test)

ncdc · 2017-10-04T11:23:24Z

Re-applying lgtm b/c the only new commit was to update bazel.

deads2k · 2017-10-04T13:05:25Z

/approve

ncdc · 2017-10-04T13:31:52Z

Gah I had an extra .go file in the repo root that caused bazel to do stuff it shouldn't have. Fixing.

Signed-off-by: Andy Goldstein <[email protected]>

ncdc · 2017-10-04T13:32:53Z

Ok, that's better. @deads2k mind giving it a quick glance again and tagging?

deads2k · 2017-10-04T13:53:12Z

/lgtm

k8s-github-robot · 2017-10-04T13:54:08Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: deads2k, enisoc, ncdc

Associated issue: 51046

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these OWNERS Files:

~~staging/src/k8s.io/apiextensions-apiserver/OWNERS~~ [deads2k]

You can indicate your approval by writing /approve in a comment
You can cancel your approval by writing /approve cancel in a comment

sttts · 2017-10-04T15:50:41Z

/retest

k8s-github-robot · 2017-10-04T17:33:31Z

/test all [submit-queue is verifying that this PR is safe to merge]

k8s-github-robot · 2017-10-04T18:47:42Z

Automatic merge from submit-queue. If you want to cherry-pick this change to another branch, please follow the instructions here.

tamalsaha · 2017-10-05T11:03:18Z

In case anyone is listening, this used to work with TPRs and was broken when we moved to CRDs.

0xmichalis · 2017-10-05T11:43:03Z

dims · 2017-10-05T11:45:08Z

@Kargakis : it should be RED!

0xmichalis · 2017-10-05T11:54:27Z

In all seriousness, @tamalsaha if there is a backport that needs to be done or what you mention is still an issue, consider opening a separate issue.

tamalsaha · 2017-10-05T14:29:59Z

@Kargakis , as a workaround, we now watch everything in a namespace and skip with an if . That works for now.

resouer · 2018-01-20T23:42:41Z

@ncdc Great work! This is included in 1.9 release right?

nikhita · 2018-01-21T17:42:49Z

This is included in 1.9 release right?

Yes. :)

k8s-ci-robot added release-note Denotes a PR that will be considered when it comes time to generate release notes. size/S Denotes a PR that changes 10-29 lines, ignoring generated files. cncf-cla: yes Indicates the PR's author has signed the CNCF CLA. labels Oct 2, 2017

ncdc mentioned this pull request Oct 2, 2017

Unable to use a fieldSelector with custom resources #51046

Closed

k8s-github-robot assigned deads2k and sttts Oct 2, 2017

ncdc assigned enisoc Oct 2, 2017

k8s-ci-robot added size/L Denotes a PR that changes 100-499 lines, ignoring generated files. and removed size/S Denotes a PR that changes 10-29 lines, ignoring generated files. labels Oct 3, 2017

enisoc reviewed Oct 3, 2017

View reviewed changes

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 3, 2017

Support field selectors for CRDs

2ff8730

Signed-off-by: Andy Goldstein <[email protected]>

ncdc force-pushed the crd-add-fieldSelector-support branch from cd3119c to 2ff8730 Compare October 3, 2017 19:22

k8s-github-robot removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 3, 2017

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 3, 2017

tamalsaha mentioned this pull request Oct 3, 2017

Use field selectors in TLS mounters voyagermesh/voyager#558

Closed

sttts reviewed Oct 4, 2017

View reviewed changes

k8s-github-robot removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 4, 2017

ncdc added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 4, 2017

ncdc removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 4, 2017

Update bazel

74b4db2

Signed-off-by: Andy Goldstein <[email protected]>

ncdc force-pushed the crd-add-fieldSelector-support branch from 283ad90 to 74b4db2 Compare October 4, 2017 13:32

k8s-github-robot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label Oct 4, 2017

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Oct 4, 2017

k8s-github-robot merged commit e9a0b15 into kubernetes:master Oct 4, 2017

dionysius mentioned this pull request Jul 26, 2018

plugin/kubernetes: support crd coredns/coredns#1999

Closed

ncdc deleted the crd-add-fieldSelector-support branch October 22, 2018 15:30

bbeaudreault mentioned this pull request Dec 16, 2020

Fixes around exceptional cases in watchers fabric8io/kubernetes-client#2671

Merged

11 tasks

Aitozi mentioned this pull request May 15, 2022

[FLINK-27163] Add label for the session job to help list the session … apache/flink-kubernetes-operator#215

Closed

ingvagabund mentioned this pull request Mar 6, 2023

WRKLDS-605: e2e: Config v1 client shim for static configuration manifests with read-only operations openshift/origin#27714

Merged

	// objectMetaFieldsSet returns a fields that represent the ObjectMeta.
	func objectMetaFieldsSet(objectMeta metav1.Object, namespaceScoped bool) fields.Set {
	if namespaceScoped {
	return fields.Set{
	"metadata.name": objectMeta.GetName(),
	"metadata.namespace": objectMeta.GetNamespace(),
	}
	}
	return fields.Set{
	"metadata.name": objectMeta.GetName(),
	}
	}

	err = scheme.AddFieldLabelConversionFunc("v1", "Node",
	func(label, value string) (string, string, error) {
	switch label {
	case "metadata.name":
	return label, value, nil
	case "spec.unschedulable":
	return label, value, nil
	default:
	return "", "", fmt.Errorf("field label not supported: %s", label)
	}
	},
	)

Support field selectors for CRDs #53345

Support field selectors for CRDs #53345

Conversation

ncdc commented Oct 2, 2017 • edited by enisoc Loading

tamalsaha commented Oct 2, 2017

ncdc commented Oct 2, 2017

0xmichalis commented Oct 2, 2017

ncdc commented Oct 2, 2017

ncdc commented Oct 3, 2017

nikhita commented Oct 3, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

enisoc commented Oct 3, 2017

ncdc commented Oct 3, 2017

ncdc commented Oct 3, 2017

enisoc commented Oct 3, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ncdc commented Oct 4, 2017

deads2k commented Oct 4, 2017

ncdc commented Oct 4, 2017

ncdc commented Oct 4, 2017

deads2k commented Oct 4, 2017

k8s-github-robot commented Oct 4, 2017

sttts commented Oct 4, 2017

k8s-github-robot commented Oct 4, 2017

k8s-github-robot commented Oct 4, 2017

tamalsaha commented Oct 5, 2017

0xmichalis commented Oct 5, 2017

dims commented Oct 5, 2017

0xmichalis commented Oct 5, 2017

tamalsaha commented Oct 5, 2017 • edited Loading

resouer commented Jan 20, 2018

nikhita commented Jan 21, 2018

ncdc commented Oct 2, 2017 •

edited by enisoc

Loading

tamalsaha commented Oct 5, 2017 •

edited

Loading