On Thursday the 11th, Jérôme Petazzoni is hosting a special session with the CNCF Maintainers Circle to share his experiences as a container evangelist. Reach out to
#maintainers-circle on CNCF Slack for more information.
Next Deadline: Exceptions Deadline, March 1
The 1.21 Enhancements Freeze is in effect by EOD Feb 9th. Please check your feature in the tracking sheet; if it’s not up to date there you’re gonna need an Exception.
The deadline for cherrypicks for the next batch of update releases is this Friday.
If you run large-scale NUMA servers, the Kubelet will now be able to manage NUMA memory affinity. This mirrors the existing CPU manager subsystem, with the only policy defined so far being
--memory-manager-policy=static. An example of a memory policy:
--feature-gates=MemoryManager=true --kube-reserved=cpu=4,memory=4Gi --system-reserved=cpu=1,memory=1Gi --memory-manager-policy=static --reserved-memory 0:memory=1Gi,hugepages-1M=2Gi --reserved-memory 1:memory=2Gi
If you are using Kubernetes in a NUMA environment, give it a shot and provide feedback to SIG-Node!
If you run the end-to-end tests locally and are hitting Docker Hub pull limits, this change should resolve them. This may be useful to locally cherry-pick to older versions if you are still maintaining things.
And to reiterate from the PR, these images are part of the e2e testing system and not for general use.
After many years as an alpha feature, the TTLAfterFinished system is finally moving up to beta. This feature is enabled through a
ttlSecondsAfterFinished in both
PodSpec, and if set the controller will automatically delete jobs or pods after they reach a finished state. This functionality has long been relegated to out-of-tree tools but as a common stumbling block, it’s great that it will be available to everyone in the next release.
- PodDisruptionBudgets should theoretically be going GA in 1.21.
- Stop auto-deleting data in Volumes if creation errors.
- Allow e2e tests to check things during disruption phases, and use explicit zone choices on GCE.
- Make dockershim pay attention to HostIP. (hey, remember dockershim?)
- Message overhaul for
kubeadm upgrade plan.
- Get scheduler to truncate messages instead of refusing to send ones that are too large.
- Open up healthcheck nodeport in IPTABLES.
kubectl waitutilizes observedGeneration to avoid stale data.
- Structured Logging Migration: taint manager this week.
A whole bunch of v1beta1 API versions are going away when 1.22 gets released, including CRDs and webhooks. Time to refactor your YAML files again!
- golang to 1.15.8 in 1.21
Original Source: Week Ending February 7, 2021 | Last Week in Kubernetes Development