Community Meeting Summary
To start the community meeting, Chuck Ha showed off the ClusterAPI provider for AWS so far. Starting with an existing Kubernetes cluster (possibly Minikube), you can use it to provision one or multiple AWS clusters.
clusterawsadm sets up your IAM roles and resources, you deploy CRDs with
makemanifest, and then
clusterctl deploys the cluster.
The SIG-Docs update, delivered by Jennifer Rondeau and Zach Corleissen, covered their big changes to localization. All languages are now in the master website repo, with new guidelines. They have new automation, Rondeau is now the Docs Lead, and Doc Sprints are happening in Shanghai and Seattle. Doc tooling has been split off into its own WG. Reminder: docs always needs technical reviewers.
Saad Ali’s status of SIG-Storage, covered their work for last quarter, including Topology Aware Volume Scheduling and Snapshot & Restore. CSI will be GA this quarter, so now they need to begin migrating in-tree storage providers to the API. They’re also extracting the mount library to a separate repo. Other features maturing this quarter include Ephemeral Volumes, CSI Topology, CRD automated installation, and Block Volume support (in-tree). They are looking into replacing CRDs for storage support, longer-term.
There will be a Meet Our Contributors Nov. 7.
Next Deadline: Beta Release, November 6.
The 1.12.2 update release came out on October 25th.
A big generated code purge, all the internal informers and listers are gone to force using the standard handler chain APIs. This was also joined by #70028 moving more internal API types out of public packages and #70210 unforking our version of typecheck now that upstream has merged our patches. Overall a solid batch of cleanup PRs!
This is a partial merge of #62692, which is aiming to improve the authentication token system to support audience-limited tokens. The future goal will allow issuing JWTs which are only valid against specific services, this this specific PR updating some of the API plumbing to make room for the additional data.
Kubeadm can now automatically set up HA etcd, a great improvement in making Kubernetes deployments both easier and more standardized. Now every deployment tool doesn’t need to maintain its own tools and systems for setting up etcd.
And finally for the week, new Github Issues templates. I would encourage everyone working in or around Kubernetes to check out the new templates, and please talk to the ContribX SIG if you have any suggestions on improving the issue reporting flow.
- Now using debian-base instead of busybox for standard server images
- Azure metadata is now cached
kubectl getnow prints all results even if erroring
kubectl describe nodedisplays EphemeralStorage usage
- kubeadm now checks for enough CPUs on master, and sorts Volumes for updating
- Windows runtime endpoints have been switched to npipe from TCP
- Internal versions are now handled differently for Custom Resources, and
- ipsets for IPv6 now work correctly
- getPids stops reading all directories
- Volume scheduling operations have metrics now
- Critical pod annotation is gone; use Pod Priority instead
- e2e tests are allowing unknown providers again, after several breakages; this will be deprecated again in the future, though
--service-account-api-audiencesin kube-apiserver has been replaced with
system:csi-external-attachercluster roles are deprecated
say goodbye to
Developer Tip of the Week
Aaron Crickenberger explained the hold command, used to prevent merging. It’s good to explain why you are holding, and bad to use hold (or cancel it) if you’re not a PR author or reviewer. Anyone can add or remove a hold.