Create test jobs to validate that debs/rpms can be installed

[justaugustus]

Per @spiffxp's review comment:

Is there an open issue for this somewhere? Would rather we use an issue to track this rather than commented out code.

creating an issue instead of having this as a code TODO.

# TODO(justaugustus): Add tests to verify installation of debs and rpms
# We comment out the current job, as the logic needs to be modified for it to be useful.
# The job has failed consistently since the 1.14.0 release:
# - https://testgrid.k8s.io/sig-release-misc#periodic-packages-install-deb
#
# Example:
#- name: periodic-kubernetes-e2e-packages-install-deb
#  interval: 24h
#  decorate: true
#  extra_refs:
#  - org: kubernetes
#    repo: kubeadm
#    base_ref: master
#    path_alias: k8s.io/kubeadm
#  spec:
#    containers:
#    - image: gcr.io/k8s-testimages/kubekins-e2e:latest-master
#      imagePullPolicy: Always
#      command:
#      - ./tests/e2e/packages/verify_packages_install_deb.sh

/assign
/area release-eng
/milestone v1.16
/sig release
/priority important-soon

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[fejta-bot]

Stale issues rot after 30d of inactivity.
Mark the issue as fresh with /remove-lifecycle rotten.
Rotten issues close after an additional 30d of inactivity.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle rotten

[justaugustus]

/remove-lifecycle rotten
/unassign

@kubernetes/release-engineering -- If someone is interested in taking this, go for it!
/milestone v1.18

[xmudrii]

@justaugustus This might be an interesting one. Is it okay if I gave it a try and ask for some initial guidance?

[justaugustus]

Sounds good, @xmudrii!
We can discuss next week.

/assign @xmudrii

[xmudrii]

I want to start working on this, but I need a few pointers to get started.

• What is expected for this job to do?
• In the kubeadm repo, there is still a script that has been used by the old job. Should we re-use the same script or build a new one?
  • The old script seems to be only for deb packages. We should eventually extend it (or add a new one) for rpm packages.
• The testgrid link in the comment doesn't work anymore since it's old. Does anyone remember why the old job was failing?

[justaugustus]

@xmudrii -- Sorry I didn't get to this this week. Still working through the post-holiday queue. Will write something up for you next week.

[xmudrii]

@justaugustus Reminder to take a look at this issue if you have some time. 🙂

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[xmudrii]

/remove-lifecycle stale

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[lasomethingsomething]

@xmudrii Hey, checking in on this -- do your questions from December 2019 still stand/do you still need the help you requested back then? @sladyn98 is interested in this item as well.

/assign @sladyn98

[xmudrii]

I've discussed this issue with @justaugustus a few months ago and I'll try to recap our conversation.

There are two ways to fix this issue: a simple and a complete way.

A simple way means that we move the old script to this repo and make it work. The script can be found here: https://github.com/kubernetes/kubeadm/blob/master/tests/e2e/packages/verify_packages_install_deb.sh

A complete way means that we do something along:

• generate the packages
• verify the packages
• install the packages on a clean system
• run e2e tests against the new cluster

The complete way is much more complicated and it's unclear how it should look like. We should start simple, port the old script, and create a job that would use it.

[sladyn98]

So if I understand correctly the steps we are likely to take are:
a) Move the sh file to this repo: I see a verify-published.sh inside the hacks/packages folder so would we be moving this script there as well.
b) Create a job that runs this sh file: Would we be adding a setp to the cloudbuild.yaml file which runs this after a certain step

[xmudrii]

a) Move the sh file to this repo: I see a verify-published.sh inside the hacks/packages folder so would we be moving this script there as well.

I would only add that we might want to slightly revisit the script. I see that it has a part for verifying stable packages. I think that this might be covered by verify-published.sh, and in that case, we should drop that part from that script.

b) Create a job that runs this sh file: Would we be adding a setp to the cloudbuild.yaml file which runs this after a certain step

I'm not sure that we need to update cloudbuild.yaml. I think that, at least at the beginning, a periodic job should be enough. Here's how the periodic for verifying published debs (verify-published.sh) looks like: https://github.com/kubernetes/test-infra/blob/675a42cb78d8eb6239cf4117e39a78f21c4e6891/config/jobs/kubernetes/release/release-config.yaml#L442-L463

[sladyn98]

So looking at current script, it does not check the versions of the installed packages and does not check kubeadm, kubelet, kubectl etc

TODO: Add logic to support checking all the existence of all packages, not just kubeadm
#       The full list is: kubeadm, kubelet, kubectl, cri-tools, kubernetes-cni

So would we be better off keeping the check or maybe adding it to the verify-published.sh and then dropping it from the new script.

[xmudrii]

We've discussed this issue in the #release-management channel on Slack. I'll try to recap the most important points.

The original script from the kubeadm repo doesn't work anymore because we are no longer building debs and rpms for each CI build. Historically, we would build debs/rpms for each CI build and put them in a bucket, from where anyone could grab them. However, we are not doing this since a while ago, so it's not possible to use the original script at all.

Instead, we want to use kubepkg to build debs and rpms, and then test those packages. However:

1. kubepkg is still not completed (some functionalities are missing), so it might not be possible to use it for building and testing packages
2. kubepkg is currently not used anywhere, so we would end up testing packages that are not used at all. That doesn't make much sense.

Because of the reasons stated above, we've concluded that it makes sense to put this issue on hold until we don't start using packages built with kubepkg.

Also, @puerco proposed that instead of running this job as a periodic, we run it as part of the release process (e.g. when staging the release).

/lifecycle frozen

[afbjorklund]

This job should probably also verify that the installed software (from the packages) has the expected version ?

Currently the kubernetes-cni 0.8.7 deb contains a repackaging of the 0.8.6 cni-plugins tgz, i.e. the wrong one

https://apt.kubernetes.io/pool/kubernetes-cni_0.8.7-00_amd64_ca2303ea0eecadf379c65bad855f9ad7c95c16502c0e7b3d50edcb53403c500f.deb

https://github.com/containernetworking/plugins/releases/download/v0.8.6/cni-plugins-linux-amd64-v0.8.6.tgz

[afbjorklund]

Failed again, for the 1.1.1 deb. It also contains 0.8.6

https://apt.kubernetes.io/pool/kubernetes-cni_1.1.1-00_amd64_910d5920eab8d7cf9a66cc4b0faf998fb3ee909bd6ee0bd7d1c6605b6a82f98f.deb

[saschagrunert]

Failed again, for the 1.1.1 deb. It also contains 0.8.6

https://apt.kubernetes.io/pool/kubernetes-cni_1.1.1-00_amd64_910d5920eab8d7cf9a66cc4b0faf998fb3ee909bd6ee0bd7d1c6605b6a82f98f.deb

Ugh, this may fix it: #2673

[afbjorklund]

@saschagrunert : would it be possible to release new packages to apt, with the correct contents ?

kubernetes-cni_0.8.6-00 (this one seems OK, no new release needed)
kubernetes-cni_0.8.7-01
kubernetes-cni_1.1.1-01

EDIT: These would be imaginary new packages, to replace the old ones with the wrong content:

$ apt list kubernetes-cni -a
Listing... Done
kubernetes-cni/kubernetes-xenial 1.1.1-00 amd64
kubernetes-cni/kubernetes-xenial 0.8.7-00 amd64
kubernetes-cni/kubernetes-xenial 0.8.6-00 amd64
kubernetes-cni/kubernetes-xenial 0.7.5-00 amd64
kubernetes-cni/kubernetes-xenial 0.6.0-00 amd64
kubernetes-cni/kubernetes-xenial 0.5.1-00 amd64
kubernetes-cni/kubernetes-xenial 0.3.0.1-07a8a2-00 amd64

[saschagrunert]

@saschagrunert : would it be possible to release new packages to apt, with the correct contents ?

kubernetes-cni_0.8.6-00 (this one seems OK, no new release needed) kubernetes-cni_0.8.7-01 kubernetes-cni_1.1.1-01

They should be automatically generated with the October patch releases. Not sure if we have to bump the package rev, though.

[afbjorklund]

Not sure if we have to bump the package rev, though.

If you do a "stealth" update, the old ones might be used from the cache (depending on how people set up their mirroring, when filenames are same)

EDIT: My bad, the filename would change:
pool/kubernetes-cni_1.1.1-00_amd64_910d5920eab8d7cf9a66cc4b0faf998fb3ee909bd6ee0bd7d1c6605b6a82f98f.deb

Also I don't think apt and yum will see it as an update, if it has the same EVR ?

But we can try it, something simple like /opt/cni/bin/bridge --version should do

[saschagrunert]

@afbjorklund I'm wondering why we are at revision 01 for those packages, it should be 00 per:

release/packages/deb/build.go

Line 95 in dbc13bd

revision = "00"

[afbjorklund]

We are at 00, unfortunately the contents are 0.8.6. So I was thinking that 01 was the next revision after ?

Normally the default debian revision is "0", so I'm not sure where the extra zero came from to start with...

[afbjorklund]

Still happening in Kubernetes 1.26.0:

anders@lima-k8s:~$ apt list | grep kubernetes-xenial

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

cri-tools/kubernetes-xenial,now 1.25.0-00 amd64 [installed]
docker-engine/kubernetes-xenial 1.11.2-0~xenial amd64
kubeadm/kubernetes-xenial,now 1.26.0-00 amd64 [installed]
kubectl/kubernetes-xenial,now 1.26.0-00 amd64 [installed]
kubelet/kubernetes-xenial,now 1.26.0-00 amd64 [installed]
kubernetes-cni/kubernetes-xenial,now 1.1.1-00 amd64 [installed]
rkt/kubernetes-xenial 1.29.0-1 amd64
anders@lima-k8s:~$ /usr/bin/crictl --version
crictl version v1.25.0
anders@lima-k8s:~$ /opt/cni/bin/portmap --version
CNI portmap plugin v0.8.6

Due to the packages not being bumped.

[xmudrii]

I'm not working on this at the moment.
/unassign