Centralizes images into agnhost (part 4)

[claudiubelu]

What type of PR is this?

/kind feature

/sig testing
/sig windows
/area conformance

What this PR does / why we need it:

Centralizes the following images into agnhost:

• dnsutils
• mounttest
• resource-consumer-controller
• test-webserver

Adds Windows support for mounttest.

Adds CoreDNS to agnhost image, which can be used in some DNS related tests (dnsmasq is Linux-only).

Bumps agnhost version to 2.9.
Bumps kitten version to 1.1.
Bumps nautilus version to 1.1.

Which issue(s) this PR fixes:

Related: #76342
Fixes #80194

Special notes for your reviewer:

Does this PR introduce a user-facing change?:

NONE

Additional documentation e.g., KEPs (Kubernetes Enhancement Proposals), usage docs, etc.:

[claudiubelu]

/test pull-kubernetes-integration

[spiffxp]

/uncc ixdy
/cc @johnSchnake

[spiffxp]

/cc

[spiffxp]

/priority important-soon
/milestone v1.16

[claudiubelu]

I've raised these questions at the K8S infra WG meeting, you can see the minutes here: https://docs.google.com/document/d/16VBfsFMynA7tObzuZGPpw-sKDKfFc_T5W_E4IeEIaOQ/edit#

But in summary:

So, from what I understand, apparently, there are a few bits of work that will have to be done beforehand in order for the automatic image promotion to be put in place for the testing images.

First of all, we would probably need a staging repo like gcr.io/kubernetes-staging-e2e-test-images. Do we have something like that first? And if not, can anyone create it?

I've sent a PR for this here: kubernetes/k8s.io#400 . The PR will have to merge, then the google group will have to be created, and then the registry.

Secondly, it seems that we'll have to to do a few new PRs in order to get that ready: on k/k, we'll have to add cloudbuild.yaml which will contain the steps necessary to build the images( Ihave some questions / concerns regarding this, though***), and one k/test-infra which will define image-pushing job.

I've added PRs for this: #84058 and kubernetes/test-infra#14833

*** There are a few things I am wondering about:

* when it comes to test images, we still have ~31 images in `kubernetes/test/images`, even after all the centralization effort. Will this mean that the job will have to rebuild / publish / push / promote all the images, even if we're modifying a single image?

It seems like test-infra has a lazy-build based on directory modifications, which would solve this.

* there is still no validation / verification that the changes made to the images do not break anything in testgrid. Unlike most other images, any breaks here will then cause the whole merging process to stop. Ofc, this can still be mitigated by having the image updates and the image versions used by E2E tests in separate PRs, like this is. Either way, making sure there's 0% chances to break something is a bit unlikely (as we're aproaching 5K E2E tests), so there could be some follow-up PRs with some fixes. But a job using the staging repo and runnining only the `[Conformance]` tests would be helpful, IMO.

Seems that this will remain as is. So, the PRs would look something like this:

• PR1: make the necessary changes to an image.
• Merge the PR.
• The Image Promoter will build and publish the image to the staging registry, from which it will be promoted to the regular registry.
• PR2: bump the image version used im kubernetes/test/utils/image/manifest.go and merge. If something breaks, revert.

* hm, I think we might need a script, or some updates in `kubernetes/test/images` which will build / push **all** the images. Right now, you can `make` only one image at a time: `make WHAT=agnhost`. We'll have to take into account that some images depend on other images: (`mounttest-user` depends on `mounttest` (`mounttest-user` is planned to be removed), `nautilus` and `kitten` depends on `test-webserver`).

Added here: #84058

@BCLAU Bug triage for 1.17 here with a gentle reminder that code freeze for this release is on November 18. Taking into account the extra work mentioned in your comment, is this PR still intended for 1.17?

I've completed the items listed above. They have to be reviewed and merged. Then, we should probably be done with that and we can then merge this.

[smourapina]

Thank you for the comprehensive update @BCLAU. Would you then say that it is feasible to have everything merged until the code freeze date? And would it make sense to add the open PRs mentioned in your comment to the 1.17 milestone?

[claudiubelu]

Thank you for the comprehensive update @BCLAU. Would you then say that it is feasible to have everything merged until the code freeze date? And would it make sense to add the open PRs mentioned in your comment to the 1.17 milestone?

I would say yes. I've got a few reviews on them as well, so they're getting closer to getting merged, hopefully.

And I would say yes, it'll definitely be helpful to have them added to the milestone. It'll help the test image changes get in smoother and without having to rely on a human to build and publish the affected images.

[claudiubelu]

/test pull-kubernetes-e2e-gce-100-performance

[smourapina]

@BCLAU Friendly reminder: code freeze is rapidly approaching and will take place in November 14th. Is this PR still intended for 1.17? Otherwise we will need to move it to 1.18 milestone.

[claudiubelu]

/hold cancel

Iti is the next PR in the queue for agnhost, it is approved, and it had the lgtm label at one point.

[claudiubelu]

Hm. I'd rather let the reviewers focus on the oher, more important PRs, like the Image Promoter for v1.17. We can punt this to the next release.

[josiahbjorgaard]

/milestone v1.18

[claudiubelu]

The Image Promoter bits are in place, so, merging this PR should cause the next agnhost image to be automatically built and published. We should be able to see the job here: https://testgrid.k8s.io/sig-testing-images#post-kubernetes-push-images

So, I think we can proceed with this PR.

[dims]

/lgtm

let's give this a shot!