Happy New Year! Due to the holidays, there wasn’t a lot of work on Kubernetes so this edition rolls up two weeks of changes.
Community Meeting Summary
Matt Farina continued with SIG-Apps update. For the last release, they completed their charter, set GA criteria for Cronjobs, started work on Portable Service Definitions and an Application Controller. This will continue into 1.14, but they need a lead dev for Cronjobs, which needs substantial refactoring.
SIG UI’s Jeffrey Sica said they also merged their charter, and they did two releases of the dashboard including some CVE fixes. They plan to add metrics support and better Oath integration.
Steve Wong briefed on SIG VMware, especially the alpha of the new external cloud provider for vSphere, including a stable CSI provider. They’re working on stabilizing the cloud provider, and want to license Fusion for Minikube.
Next Deadline: 1.14 Dev Cycle Begins, January 7th
This week should be Week 1 for 1.14, which also had its first release meeting on January 7th. A full release schedule will be published soon. Important: SIG Release is proposing that all enhancements must be linked to a KEP in order to merge in 1.14.
The Patch Release Team is working on making a patch schedule.
A quiet week over the New Year, but we did have a few cool PRs to check out!
A new UI to visualize what actions Tide is taking. This can help with debugging odd behaviors or getting a better idea of how Tide is operating internally.
A newly accepted Kubernetes Enhancement Proposal, this proposal outlines a new
PodGroup object. This will be used as a way to pre-reserve resources for multiple pods at once for situations where you want a heterogeneous collection of containers to launch as an all-or-nothing package deal. Some examples from the KEP include machine learning batch jobs and Spark pods.
Top committer Jordan Liggitt seems to have used his holidays to clean up many minor bugs and bits of obsolete code. Aside from that, there were some minor changes:
- Workqueue metrics now use Prometheus naming conventions
- Stop running an etcd compactor for every service in Kubernetes
- RuntimeClass display has more informative columns
- Azure MaximumLoadBalancerRuleCount is now 250
- Prevent improper subpath deletion when cleaning up pods, then clean them up properly
- The APIserver now checks the informer for service accounts first to reduce load
- Let Kubernetes work even when docker is set to private IPC mode