LWKD: Week Ending November 14, 2021
Developer News
Dockershim is going away in 1.24. Are you ready? SIG-Node is collecting user feedback on the current state of the removal. It will only take you a few minutes, please fill it out if you can. Thanks!
We are testing moving to EasyCLA for CLA-signing, to replace the older propriety Linux Foundation system. Currently EasyCLA is giving lots of warnings, but it's not blocking your merge. If it doesn't recognize you, comment on the tracking issue.
Release Schedule
Next Deadline: Code Freeze, Nov. 16th 23:59 UTC
It's here! Tomorrow is Code Freeze for 1.23, and it's time to decide if your feature is going to make it or not. 23 of this release's 60 tracked enhancements are "at risk"; if one of those is yours, you need to get in touch with the Enhancements team now. From here until the release, you'll need a milestone maintainer to approve anything that needs to go into 1.23.
The next patch releases are due out on Wednesday.
Featured PRs
#102534: Create HPA v2 Stable API
The long-awaited HPA v2 has finally arrived after many years of v2beta. Mostly this solidifies the changes early in the beta cycle to have much more flexible scaling target descriptors, but if you're already using v2beta2 then look forward to easier kubectl integration. A few internal values have been renamed so if you create an HPA object from operator code in Go, you'll have to update MinPolicySelect to MinChangePolicySelect and MaxPolicySelect to MaxChangePolicySelect.
#106123: package log runner as part of Kubernetes releases
Originally introduced as go-runner, kube-log-runner is a program with exactly one mission in life: redirect process output to a log file. This has been an internal component of our testing systems for a while, letting us test things on the slim distroless base images to save space (and thus speed up testing) while still collecting logs of the test runs. If you have similar "output redirection without a shell" needs in your tooling, check it out!
Other Merges
- SIG-Apps accidentally added time zones to the CronJob API; attend their meeting if you would like to help work out what we do about this
- Pod OS selector is complete now
- Keep kubelet from leaving log file descriptors open
- Kubectl won't panic if you mangle your Secret creation; backported
- kube-proxy uses debian-bullseye-v1.1.0 to get access go gorunner
- kube-scheduler won't break your log reader when logging config
- Prevent managedFields from changing behind webhooks' backs
- RequestedToCapacityRatio strategy requires a ShapePoint
- API Priority & Fairness gets more metrics
- Make sure the volume is actually deleted if that's what should happen
- Windows containerd network stats are a thing
- Kubeadm stops versioning Kubelet ConfigMaps
kubectl cpcan resume after network errors- Don't create a replacement static pod until TerminationGracePeriod is over
- Correctly validate sysctl params even if they have
/
Promotions
- ConfigurableFSGroupPolicy to GA
- IngressClassNamespacedParams to GA
- DownwardAPIHugePages to beta and enabled by default
- WindowsHostProcessContainers to beta
- DelegateFSGroupToCSIDriver to beta, so FSGroups are automatically pushed down to CSI volumes that support them
- A bunch of scheduler metrics are now GA; if you've been using them already, you can feel reassured
Structured Log Migrations: cache, comparer and dumper and note_tree, preemption and stateful and resource_allocation, and finally assume_cache
Version Updates
- go to 1.16.10 in 1.21, 1.22, and 1.23
- etcd to 3.5.1
- kustomize to v4.4.1
- system-validators to 1.6.0