Community Meeting Summary
Andrew Chen delivered the SIG-Docs update, sharing upcoming work like the new docs contributor guide, an alternative Chinese search engine and trying to straighten out generated docs in time for 1.12. They are also looking for folks to diagram architectural concepts, and covered the search outage postmortem
Sahdev Zala introduced SIG-IBMCloud, who support compatibilty with IKS and IBM Cloud Private. Currently they are working on supporting the Cloud Provider API and their Charter. SIG-Autoscaling’s Solly Ross explained their current work on the Horizontal Pod Autoscaler (HPA), such as making scale limits smarter and more algorithmic, in time for the feature to be beta in 1.12. They are also figuring out how to integrate the Cluster Autoscaler with the Cluster API.
Reminder: The Steering Committee election process will start (delayed) this week.
New contributors will now ask for Org Membership by filing an issue in kubernetes/org. Aaron Crickenberger went over upcoming automation and homogenezation coming to the Kubernetes repos and namespaces soon:
- Be in SIGs.yaml
- All repos should have OWNERs
- Required label set for repos
Use the same merge automation
Some questions remain open, such as whether all repos should have
Next Deadline: Placeholder PRs for Docs, August 21st
Release Lead Tim Pepper warned contributors that Code Freeze is in just two weeks. Version-branch CI should be delivered this week.
KEP of the Week
Chris Hodge went over KEP 0019, which is the docs portion of moving Cloud Providers out of the main repo. To transfer responsibility, SIG-Cloud-Provider needs to set standards for docs, and provide docs on how to activate a new cloud provider.
The new design for kubectl plugins has landed. In addition to fundamentally changing how you extend kubectl, this also adds kubectl plugin and supports users loading plugins just by dropping a
kubectl-* binary in their PATH.
Volume resizing handlers have been moved from the general controller to the specific storage controllers, in preparation for handling volume resizing on a per-CSI-driver basis.
- Azure storage has stopped hard-coding storage devices, and now handles unzoned disks
- Fix buffering performance for audit logs
- Stop mis-assigning protobuf types
- Stop loading the kube-aggregator container image since aggregation is now included in the apiserver
- Don’t fail when filesystems EIO on unmount
- Reduce the work that anti-affinity checks do
- Refactor attach/detach for local storage
- Stop scoring every node on every scheduler cycle to lower overhead
- Print long public cert strings as DATA+OMITTED
- Add a backoff to the Daemonset scheduler
- Extend the ability to set debug logging dynamically to all components
Heapster is deprecated in 1.12, and will be removed entirely in 1.13. Time to migrate your services away from Heapster.
Graph of the Week
Crickenberger went over several boards and graphs that allow contributors to track flaky tests, including the above bigquery chart of PR flakiness. Together with the presubmits test grid, PR authors and code owners should regularly check if their code is making the builds or tests unreliable.