Mateusz Gozdek

Currently, we do not test availability impact of our upgrade process on the cluster. This means that **ideally**, when user is doing the upgrade process, all production traffic should be...

As a follow up #704, we should enable --wait flag in bootkube Helm and switch the policy to fail, so the check cannot be bypassed.

This way, downloading the binary using `go get` gives more representative version. See https://github.com/sigstore/rekor/issues/212#issuecomment-894805808 for details.

Enable pod checkpointing for kube-scheduler and kube-controller-manager

1

Right now, if all scheduler or controller manager pods gets killed, there is no way to recover the cluster. We should address that. As a temporary workaround, we could enable...

Would be nice if documentation on website could say at least what version of Kubernetes we ship. We could also consider mentioning version of components.

It would be nice to have a changelog (maybe per minor version?) on the website, instead of only on GitHub.

While master is updated to 1.21 already.

Investigate "error building openapi models for" logs in kube-apiserver

3

Following logs can be found in a newly deployed cluster. ``` E1118 18:50:07.321385 1 customresource_handler.go:668] error building openapi models for alertmanagers.monitoring.coreos.com: ERROR $root.definitions.com.coreos.monitoring.v1.Alertmanager.properties.spec.properties.containers.items..properties.lifecycle.properties.postStart.properties.httpGet.properties.port has invalid property: anyOf ERROR $root.definitions.com.coreos.monitoring.v1.Alertmanager.properties.spec.properties.containers.items..properties.lifecycle.properties.postStart.properties.tcpSocket.properties.port has...

https://github.com/projectcalico/calico/issues/4227 has been fixed and we use right now custom-uploaded manifests.