etcd 3.2 works with Kubernetes 1.18 but not 1.19

[James Troup]

We followed the CDK 1.18 -> 1.19 upgrade documentation available here:

  https://ubuntu.com/kubernetes/docs/1.19/upgrading

At the end of the process, we were still on etcd 3.2, but it was not working reliably with Kubernetes 1.19 - etcd was crashing a lot with backtraces. Either this is a bug in 3.2 which should be fixed, and/or our documentation should explicitly tell you to upgrade to 3.4.

[James Troup (elmo)]

Subscribing field-high as (I think) this will bite anyone who follows our upgrade docs; if etcd 3.2 is meant to work with Kubernetes 1.19, let me know and I can provide logs.

[George Kraft (cynerva)]

I can reproduce this easily enough. Deployed with k8s 1.18 and etcd 3.2, then upgraded to k8s 1.19, and etcd started crashing.

There are known issues with etcd 3.2 in Kubernetes[1] so I think the correct action here is to add an upgrade note instructing users to upgrade to etcd 3.4 prior to upgrading to k8s 1.19.

[1]: https://kubernetes.io/docs/tasks/administer-cluster/configure-upgrade-etcd/#known-issue-etcd-client-balancer-with-secure-endpoints

[George Kraft (cynerva)]

Docs PR: https://github.com/charmed-kubernetes/kubernetes-docs/pull/530

[George Kraft (cynerva)]

Currently awaiting review from the web and design team: https://github.com/canonical-web-and-design/ubuntu.com/pull/9121

[Nick Veitch (evilnick)]

These changes are now live on ubuntu.com