Improve release process

kubernetes / node-problem-detector

This is a place for various problem detectors running on the Kubernetes nodes.

Apache License 2.0

2.94k stars 627 forks source link

Improve release process #874

Closed wangzhen127 closed 6 months ago

wangzhen127 commented 7 months ago

Currently the release process is very manual. And only @Random-Liu can run make push. We should simplify the release process.

CC @Random-Liu @vteratipally @hakman

wangzhen127 commented 7 months ago

Discussed with @BenTheElder offline. The main pain point of pushing and pulling from gs://kubernetes-release can be removed by just using the github.com file hosting in release notes. Since we already upload tar files in release notes, this would be a simple change.

See https://github.com/kubernetes/kubernetes/pull/123741

hakman commented 7 months ago

@wangzhen127 We can also promote binaries, like we do for kOps to the bucket (via k8s.io repo) and just link them in release notes. Ref: https://github.com/kubernetes/k8s.io/pull/6368

wangzhen127 commented 7 months ago

That's also fine. But using the github hosting file is the simplest way for now.

Does your organization set up NPD as system daemon of daemonset? So far, cluster/gce/gci/configure.sh is the only place setting it up as system daemon as far as I know. Given kube-up.sh will switch to kops in the future, and kops decides to treat NPD as addons, it may not worth the effort to set up pipelines to promote binaries.

hakman commented 7 months ago

I am fine with images at the moment. kOps already has NPD as an addon 😄 . See https://github.com/kubernetes/kops/pull/11381. We should update the NPD tests to run via kOps in the near future.

wangzhen127 commented 6 months ago

Just realized a few things

NPD releases tar files to gs://kubernetes-release/node-problem-detector/ historically. No one has permission to push files to it now. We only have tar files up to v0.8.10 in the GCS bucket.
Currently k/k's default NPD version is:
- k/k 1.28 uses NPD v0.8.9
- k/k 1.29 uses NPD v0.8.13
- k/k 1.30 uses NPD v0.8.13
NPD's standalone mode configuration in k/k depends on the tar files in the above GCS bucket. Given none of k/k's release blocking tests fail, we already lost the test coverage for about half a year (for both k/k 1.29 and 1.30). I filed https://github.com/kubernetes/node-problem-detector/issues/878 for tracking separately.

For the release process, we should stop pushing tar files to gs://kubernetes-release/node-problem-detector/. We lost push permission for a long time already. Relying on github file hosting should be fine for k/k.

wangzhen127 commented 6 months ago

I have removed the need of pushing tar files to gs://kubernetes-release during release. So any approvers should be able to make the release now. It is still a manual process. But given the low frequency of NPD releases. I can live with it for now. We can revisit if situations change.