Community Meeting Summary
This week’s community meeting started with a demo of multicluster-scheduler, which is part of a suite of tools for managing multiple clusters. Adrien Trouillaud showed off running an Argo workflow on two clusters.
SIG-Azure has been working on support for the various kinds of Azure disks and availability zone management. They will soon be moving the cloud provider out of tree (alpha in 1.14), supporting the Cluster API, and adding CSI drivers.
SIG-Release has proposed the creation of an LTS Working Group, and plans to create several other subprojects. It’s also continuing to tinker with the release process (require KEPs, no more code slush) and is pushing to “deflake” CI. Tim Pepper also explained the Release Engineering subproject, and the problems with current build tools and automation.
Next Deadline: KEPs must be Implementable Feb. 4th
We completed the 4th week of 1.14 development, and released v1.14-alpha.2. Enhancement Freeze was Tuesday, with 8 alpha, 8 beta, and 9 stable features, plus 11 KEPs that still need work. This has meant a lot of exception requests.
For CI Signal, tests have now been divided into Blocking and Informing, and Blocking jobs all need SIG owners (including an alerts email). Also, the KOPS presubmit test has been removed from automatic pull tests because of AWS account issues.
Update v1.13.3 was released last week, including update to Go 1.11.5. Since this contains a security update, please update your 1.13 clusters ASAP.
- Support Windows nodes on GCE
- The DiscoveryAPI now offers a StorageVersionHash for appropriate objects, per the KEP
- Missing plugin directories won’t crash kubectl anymore, and kubectl also supports
[#:#]syntax for custom columns as well as
waiton a label selector
- Don’t drop all watchers in mixed-version cluster
- Make Daemonsets delete pods assigned to non-existent nodes
- Make scheduler move unschedulable pods to active queue as part of a MoveRequest
- Forward plugin is now default for CoreDNS
rulesfield is now optional for RBAC Roles
- Pod eviction now uses graceful deletion
- Fix time units in admission metrics
- Support AWS Custom Endpoints and SSL health checks
- kube-proxy restarts when its config changes
--repair-malformed-updatesis gone from the APIserver
--experimental-fail-swap-on, deprecated in 1.8, removed