ICYMI, Kubernetes 1.20 has been released. This particular release has 42 major enhancements, making it the most featureful release in a while, and probably the most disruptive as well. Maybe use the holidays for testing instead of deploying it on Christmas Eve. Just a tip.
The chairs of the Component Standard WG are retiring and are looking for new leaders for that WG.
Next Deadline: 1.21 cycle begins, January 5th
We are currently in the limbo period between the release of 1.20 and the start of official 1.21 development, otherwise known as “packaging up 1.20.1”. Expect an announcement of the start of 1.21 after New Year’s, and even consider being a release team shadow.
A good reminder and example of updating code using the old
failure-domain.beta.kubernetes.io labels, this PR switches a lot of systems over to the newer
topology.kubernetes.io equivalents. If you’ve been putting off a similar update in your in-tree code or external automation, now is a great time to start!
If you use the AlwaysPullImages admission controller, this should ease switching pains when enabling it. Previously all pods would have to be recreated due to imagePullPolicy being an immutable field, now updates to existing pods created before enabling the admission plugin will still be updatable except to change their image.
To improve usability by humans,
kubectl logs -l will now display an error banner if no pods match the requested label selector. If you use commands like that in automation, notably in CI jobs where the pods don’t always exist yet, be sure to check for compatibility.
- Carlos Panato gives kubectl a Brazillian translation
- Auto-detect global ip addresses attached to the loopback interface on nodes, and support BGP Unnumbered
- APIserver’s locking lease in etcd becomes configurable
- Results of
kubectl logs --tail #are now repeatable
- No more panic in the kubelet if some status fields are missing
- Add AcceleratorStats to cri_stats_provider (backported fix)
- Don’t corrupt Fiberchannel volumes on disconnect
- Cloud provider routes stop depending on
--allocate-node-cidrsby side effect, are set only by
- Several deprecated scheduler metrics are being removed in 1.21.
Service.topologyKeysis deprecated in 1.21, to be replaced by Topology Aware Subsetting
- cadvisor to v0.38.6 in 1.21 and 1.20, to v0.37.3 in 1.19
- Cluster Autoscaler to v1.20.0
- etcd to v3.4.13 in kubeadm deployments of 1.19 and 1.20
Original Source: http://lwkd.info/2020/20201215