Clean up ECR credential patch

[jonjohnsonjr]

In #4084, we added a patch in order to backport some changes in kubernetes to fix #1996. Those changes landed in 1.15.

Once we are able to update our dependencies to 1.15, we can just drop that patch.

X-ref: google/go-containerregistry#355

/area API
/area test-and-release
/kind process

[jonjohnsonjr]

This got merged into 1.14, so we can drop it earlier 🎉

[toVersus]

Hi, @jonjohnsonjr (cc. @mattmoor )

I encountered the following error while trying to deploy ksvcs on HEAD (995194b).

Unable to fetch image "xxxxxxxxxxxx.dkr.ecr.ap-northeast-1.amazonaws.com/rosetta-go/gateway:v0.4.0": unsupported status code 401; body: Not Authorized

I know we dropped this ECR credential patch in #5570 along with bumping up K8s toolchains to v1.15. However, it seems not to get to work on my EKS cluster (v1.14.6). I wonder if we need a little bit more works on solving this problem.

• Update K8s dependency on upstream google/go-containerregistry first as described in k8schain doesn't work with ECR google/go-containerregistry#355 (comment)

• Bump up google/go-containerregistry version
  

  serving/Gopkg.toml

  Lines 45 to 48 in 995194b

  	[[constraint]]
  	name = "github.com/google/go-containerregistry"
  	# HEAD as of 2019-09-10
  	revision = "b02d448a3705facf11018efff34f1d2830be5724"

• Delete dependency no longer needed
  

  serving/Gopkg.toml

  Lines 12 to 13 in 995194b

  	# TODO(#4549): Drop this when we drop our patches.
  	"k8s.io/kubernetes/pkg/version",

If my understanding is correct, do you have any plans for pushing those changes until v0.10 release date?

[jonjohnsonjr]

I'm really confused by the #5570 diff. It looks like it changes the signature of LazyProvide back to a single parameter here, but the ECR credential fix requires this commit which changes LazyProvide to take two parameters, which was included in the patch...

[toVersus]

I reverted back 1996.patch partially and it got to work fine as expected.

https://gist.github.com/toVersus/6a76477e6eb342499ea7242e5ed65a72

[jonjohnsonjr]

That makes sense. I think this is actually impossible to fix until kubernetes/kubernetes#82396 is resolved. I can't update my dependency on kubernetes/kubernetes/pkg/credentialprovider upstream because go-containerregistry is using go modules, and kubernetes has intentionally broken anyone depending on packages under k/k/pkg by setting the go module version to v0.0.0. Trying to update yields:

go: k8s.io/kubernetes@v1.15.3 requires
	k8s.io/api@v0.0.0: reading k8s.io/api/go.mod at revision v0.0.0: unknown revision v0.0.0

We need to update this dependency upstream in go-containerregistry because of the breaking change (as in my comment) or carry a patch that fixes it (which this issue tracks cleaning up).

[knative-housekeeping-robot]

Issues go stale after 90 days of inactivity.
Mark the issue as fresh by adding the comment /remove-lifecycle stale.
Stale issues rot after an additional 30 days of inactivity and eventually close.
If this issue is safe to close now please do so by adding the comment /close.

Send feedback to Knative Productivity Slack channel or file an issue in knative/test-infra.

/lifecycle stale

[toVersus]

/remove-lifecycle stale