Week Ending Feburary 18, 2018

Community Meeting Summary

This week’s Community Meeting demo was AppZ by Cloudbourne, a proprietary online source-to-production development and deployment service based on Kubernetes. Jaice DuMars gave an update on the 1.10 timeline, and 19.3 and 1.8.8 have been released. 1.9.4 is expected about two weeks from the meeting.

Aaron Crickenberger offered an update from SIG-Testing, announcing that Jenkins is no longer used in the project’s Github development, having been entirely replaced by the Kubernetes-native Prow. He thanked Amazon for hosting a Testing face-to-face in Seattle January 26th. He reminded people about Testing Commons, went over some Prow UI updates, and discussed implementation of Tide, a tool designed to replace the submit queue. Testing would like to set firm criteria for blocking test failures, and they’re looking for help implementing Docker-on-Docker in the test infrastructure so that they can simulate a cloud without needing to run one for every E2E test.

Paris Pittman of SIG-Contribex also had many updates, including a work on a new charter and the Contributor Guide, which will be merged into kubernetes.io. The mentoring subproject needs contributors, particularly more mentors for Outreachy, GSOC, Meet Our Contributors and SIGs that can devote time to group mentoring. They’re also working on a User Guide for DevStats graphs, Issue Triage guidelines, and Slack usage guidelines, and would like feedback from contributors on all of these. Contribex is having a face-to-face this week, will be doing a “SIGs Roadshow”, and wants to remind everyone about the Copenhagen Contributor Summit.

SIG-API-Machinery, whose Daniel Smith reminds everyone that they don’t spec the API, they just plumb it, would like feedback on their Advanced Configuration Management design proposal.

Finally, the meeting began an important discussing into the functioning of LGTM and Approval labels. Currently automation settings trigger “auto-approval” of merges when the submitter is an owner. This has caused some SIGs and repos to accidentally merge things that weren’t ready. What should the correct behavior be, and should it be configurable by repository? Please add your thoughts on the issue.

KubeVirt Meeting Summary

Network Device Plugin work in order to get network resources into a pod using a device plugin. Still early research but it’s progressing.

Release Schedule

Last week was Week 7 of 12 for 1.10 development, and included a beta release and setting up the release branch. Tuesday the 20th will begin Code Slush, where issues and features need to get ready for final merge on Code Freeze. All issues in the milestone are considered release-blocking until removed. Currently there are 48 issues open in the milestone.

Feature Work

• kubectl port-forward now supports specifying a service, as well as using Resource Name.
• Added disk support for VMSS nodes on Azure, which also gets a cache.
• Allow specifying an IAM role for the AWS cloud-controller-manager instead of using a default role.
• fluentd-gcp resources can now be modified as part of a ScalingPolicy without cluster restart, and fluentd-gcp is changing its logging config.
• We now monitor the kubepods cgroup directly to determine pod memory usage more accurately, especially for evictions.
• Auditing now supports matching resources by wildcard.
• The KubeletConfig API is now beta for 1.10.

Deprecated

• CSI is breaking more things in the API to rev version 0.2; if you touch CSI, you need to check the changes in 0.2/Kube 1.10.
• Eviction thresholds of 0% or 100% will now be explicitly ignored, as this setting is a bad way to disable eviction.
• Controller-managers (kube- and cloud-) will now use option+config struct for bootstrapping, which also means closing the old insecure HTTP port.
• ConfigOK has been renamed to KubeletConfigOK.
• The APIserver has removed its hard-coded /ui/ redirect.

Version Updates

• Failed upgrades can now be rescued with etcd minor version downgrades.

Other Merges

• Kube-proxy has been refactored to provide more shared code for user-defined proxies, cutting 500 LOC per proxy.
• A fatal race condition in the SharedInformer is fixed.
• Kubernetes-OpenStack environments can now automatically pick IPv6 vs. IPv4 correctly in mixed environments.
• Create Azure blob disks without error.
• GCE has changed the RBAC roles used for its Cloud Provider addon.
• Kubelets will stop caching stale PVC metrics.
• Nodes no longer stall when they run out of processes.
• The Scheduling folks are changing how Volume equivalence is handled.

Other Repositories

• KubeVirt: Native PV based storage will allow KubeVirt to use any regular Kubernetes VP to store VMs. For now the support is limited to storing the VM image in the PV file-system mode (raw block mode is planned).
• KubeVirt: Native Kubernetes networking integration will allow KubeVirt to integrate VMs into the stock Kubernetes network without dedicated support from the underlying network plugin. A user will be able to access and expose VM based services in the same manner as regular pods.

Last Week In Kubernetes Development (LWKD) is a product of multiple contributors participating in Kubernetes SIG Contributor Experience. All original content is licensed Creative Commons Share-Alike, although linked content and images may be differently licensed. LWKD does collect some information on readers, see our privacy notice for details.

You may contribute to LWKD by submitting pull requests or issues on the LWKD github repo.