Kubernetes 1.12 has been released! More on this below.
Community Meeting Summary
Yinan Li shared the activities of SIG-Big-Data, mostly Spark. Apache Spark 2.4 will include the following Kubernetes features: python & R support, client mode for spark-shell, mounting PVs, and scheduler improvements. The Spark Operator now has Prometheus support, and the Airflow operator is 1.0. They’re still working on pod templates, Kerberos authentication, and other Spark integration features.
SIG Scalability’s Shyam Jeedigunta pitched in with an update, incuding the SIG’s new charter. He went over performance test issues in 1.12 (see below), mentioned that the Cluster Loader is now ready for users, and explained the effort to refactor scale-testing infrastructure. They also want to make presubmits faster.
Next Deadline: begin work on 1.13, October 1st
Version 1.12 is released, and Pengfei Ni has now taken over as patch manager. Any known issues should be targeted at 1.12.1. The 1.12 retro will be discussed for the Community Meeting on Oct. 4th.
1.9.11 was released by patch manager Mehdy Bohlool on Sept. 27th, as the last update to 1.9. If you are still using 1.9, you need to either upgrade soon or get legacy commercial support.
The 1.13 release cycle is now open and the Release Team meets Oct. 1st. Please tag any feature patches with Milestone v.1.13. As 1.13 is a short cycle (10 weeks), SIGs should focus on stability instead of new features.
A quick feature merge now that freeze is behind us, this adds a
MayRunAs group strategy for volumes and containers in the PodSecurityPolicy system.
A bit of forward motion away from old Docker-y defaults, Kubernetes will no longer populate the
docker --link-style environment variables by default, in favor of using DNS-based service lookup through kube-dns or CoreDNS. It can be turned back on if needed using the
A small PR which should be nice for people bouncing kubelets a lot during tests, this should prevent pods from coming up before the network is ready.
Another (very) small PR, but should be a speedup for running gofmt.
And finally, another bit of momentum from the Kubeadm team, removing the alpha2 API. Existing kubeadm config objects can be upgraded using
kubeadm config migrate.
Switching the default DNS plugin to CoreDNS was booted from 1.12 at the last minute. This was due to performance regressions found by the SIG-scalability team during Code Freeze. Hopefully they can be fixed for 1.13.
- Reduce privileges required for kube-proxy containers
only_cpu_and_memoryparameter for retrieving metrics.
- Container-runtime e2e tests are now conformance tests
- All control plane images now have an nsswitch.conf
Fix panic triggered by
- Stop checking the disruption budget for pods we’re terminating
- Add a dynamic lister to client-go
- kubectl creates go profiles now
- Failing priority map functions no longer crash
- Stop bogging down distributed storage with chown commands
- Dropped IPVS endpoints now gracefully terminate
- All in-tree cloud providers now carry a deprecation warning as part of moving all cloud providers to the API.
- We’d deprecated & removed node.Spec.Unschedulable, but had to put it back for compatibility
Graph of the Week
Aaron Crickenberger shared the Triage Dashboard, a place where contributors can go to track down why tests are failing. You can use this interface to see gubernator logs, history of test failures, and other useful troubleshooting data.