Azure v17.0.0-alpha1 release

[alex-dabija]

ℹ️ Team Phoenix is drafting Azure v17.0.0 release to update to Kubernetes v1.22 .
This will be an alpha release with no possibility to upgrade from or to.
Target release date: 2022-02-24 .
Release PR: giantswarm/releases#937 .

Scope

• Kubernetes v1.22;
• Latest Flatcar with cgroups v2;
• Automatically roll nodes after Control Groups v1 are enabled;
• Disable rpcbind if not needed;

Pre-Release

• Create release ticket on Releases board
• Make an announcement on #sig-product that team is doing a release. Include a link to the ticket from the release board.
• Make sure release was tested by creating new cluster, upgrading from older version and with HelloWorld app.
• Aggregate release notes from all changed components and add to README.md in the release directory.
• Send draft release announcement to #sig-product for review ASAP.
• Verify documentation is up to date.

Releasing

• Create release PR
  Instructions
  • Define new release in release.yaml file.
  • Activate new release by setting state to active.
  • Deactivate old release if it's within same major version (state: deprecated). Verify with PO if you have doubts.
  • Check versions of all components included. They shouldn't be lower than those in the previous release unless there was a rollback.

Example PR

• Release relevant operators.
  • Tag release branch.
  • Bump operator version in draft release.
• Make sure that WIP releases in releases repo branch are removed and not merged to master since the master branch is deployed automatically after merging PR and that would make WIP releases visible in customers’ installations which should generally not happen.

Post-Release

• Update front-end cluster with the new release.
• Verify release is available on all installations. You can use this script.
• Send out release announcement to all customers according to these instructions.
• PO subscribes to release posts on Slack to get notifications if there are follow-up messages.
• Close release task on Releases board.

[whites11]

There is a bug in cluster autoscaler that prevents it from running in our 1.17+ clusters:

• bug was introduced in 1.22.0, that's why we didn't witness it so far.
• behaviour was changed completely in 1.23.0 and bug disappeared because of that.

Outcome is that with 1.22+ we need to provide subscription ID explicitly to cluster autoscaler as it won't get it automatically any more.

Options:

• change cluster-operator and cluster-apps-operator to provide subscriptionID to helm charts.
• add back the removed feature in cluster operator and backport to 1.22.
• add an initContainer to get the subscriptionID from the metadata endpoint and provide it to the cluster autoscaler container through an emptyDir

neither solution shines, not sure how to move on.

[whites11]

This could be the PR for upstream. kubernetes/autoscaler@master...whites11:azure-subscription-id-from-metadata

[whites11]

PR to fix 1.22 is out kubernetes/autoscaler#4600

[whites11]

Blocked until upstream deals with our PR

[alex-dabija]

Blocked until cluster-autoscaler v1.22.3 is out.

[alex-dabija]

The release is still not out.

[whites11]

The release is still not out.

Should I release a patched version of CA 1.22?

[alex-dabija]

The release is out.