We’re trying out a slightly altered format for this issue. Feedback? Tell us on Twitter or Discuss.
Gubernator, the UI that allows you to dig into your test failures on the test framework, is being replaced by a new UI. The new project, called Spyglass is a Prow-native UI developed by the Test-Infra team with more test information and a vastly better looking interface. The switchover should happen in the next few days.
Next Deadline: Doc placeholders for all Enhancements, March 1
Code Freeze is March 7th.
The logic that powers
kubectl drain has been broken out into a
pkg/kubectl/drain library so that it can be used in other tools. This process was needed for just about every Kubernetes deployment system, so many had to either re-implement things themselves or shell out to
kubectl. Moving forward, we can standardize on this library to provide a unified experience and fewer bugs!
A revision to #72619, the timestamp on the Pod object will no longer be updated every time the scheduler fails to place the pod. While the feature was solid, the write load for clusters with many unschedulable pods was getting to be a problem. So now the same timestamps will be tracked within the scheduler. All the same resource allocation-y goodness, now with much less Etcd writing.
If you had written code in the past few weeks which depends on that Pod timestamp, you will need to rework it.
A great feature from the Windows team, kube-proxy will now support Overlay mode networks in addition to the existing L2Bridge support. Overlay-mode networks use VxLAN encapsulation to provide a multi-node network fabric. This also allows interoperation with existing VxLAN users such as Flannel and Canal-based Linux clusters.
Check out the MSDN docs on container networking for more info about Windows networking modes.
A simple move but important for anything integrating with Ingresses. Anything importing the types will need to update from
pkg/networking, and manifests will need to be upgraded in the coming months. This is a nice step forward for the Ingress feature though, as it moves closer to GA.
kubectl autoscalehas been refactored to potentially support CRDs
- VolumeSubpathEnvExpansion has been expanded as a breaking change to the alpha feature
- Cinder volume limits can be configured from openstack.conf
- Multiple fixes for issues on Azure, including devicepath, loadbalancer, timeouts, and detaching disks
- Make admission plugins work with CRDs
- CoreDNS has a readiness probe
kubeadm resetnow removes the etcd nodes
- Decrease TTL for local cache to avoid incoherency
- Split up the test tarball into several smaller ones
- Improve performance of the scheduler cache, speeding up scheduling by over 20%
kubeadm initsupports uploading certificates
kubectl --quietnow hides deletion messages
- Shut down gracefully on Windows
- Untag docker images before removal, so that they actually delete
kubectl logsnow allows using both follow and label queries together
- The underlying kubelet changes for Windows group managed service accounts are in
- PriorityClass v1beta1 and v1alpha1 are Deprecated since PriorityClass is now GA
- Versionless client API accessors like
clientset.Apps(), long-deprecated, are gone
kubelet --containerized, which never really worked, is deprecated
- kubeadm will no longer load deprecated configs
- API export queries are deprecated due to issues, and will be removed down the line