Week Ending August 25, 2024

Developer News

KubeCon + CloudNativeCon + Open Source Summit China 2024 happened last week in Hong Kong. The event had various talks on AI, running AI workloads on Kubernetes and the CNCF ecosystem, and updates from various maintainers of different CNCF projects. There was also a keynote by Linus Torvalds. Videos will be posted in the CNCF YouTube channel soon.

Release Schedule

Next Deadline: 1.32 cycle begins, September 9

We’re in the period between releases. Shadow applications for the v1.32 release team are open until September 6. The tentative dates for the v1.32 cycle are from September 9th to December 11th, 2024.

Featured PRs

#126745: Improve PVC protection controller’s scalability by batch-processing PVCs by namespace & caching live pod list results [fixed dead loop issue with idle work queue]

This PR significantly enhances the scalability of the PVC Protection Controller by implementing batch processing of PVCs by namespace and caching live pod list results. It resolves a critical dead loop issue in the idle work queue and addresses performance bottlenecks in large clusters by reducing the number of API calls required for PVC deletion. As a result, the kube-controller-manager’s CPU usage is optimized, ensuring more efficient and reliable operation, especially in environments with high pod and PVC churn.

KEP of the Week

KEP 3998: Job success/completion policy

This (KEP) aims to enhance Indexed Jobs by allowing custom success criteria, so a job can be marked as succeeded based on specific pod indexes, such as leader pods, rather than requiring all pods to succeed. It supports distributed computing frameworks like MPI and PyTorch, where only certain pods determine job success. The proposal does not alter the default behavior for jobs without a SuccessPolicy or extend this feature to NonIndexed Jobs in its first iteration.

This KEP is tracked for beta release in v1.31.

Other Merges

kubeadm now sorts the result of MergeKubeadmEnvVars, and allows mixing of flags ``–print-manifest and –config`
Printer unit tests added for DRA resources
transformation_operations_total metric gets additional resource label
pkg/kubelet/cm/dra migrated to contextual logging
Fix for estimated cost for Kubernetes defined CEL types for equals
Common apiserver for all testcases in CEL tests
kube-scheduler removes non-csi volumelimit plugins
Scheduling throughput thresholds set in scheduler_perf tests
Fix to DRA with structured params to make unschedulable pods schedulable again after ResourceSlice cluster events
kube-proxy now uses field-selector clusterIP!=None on Services to avoid watching for Headless Services
NominatedPodsForNode moved to scheduling queue to make the invocations more direct
Events cached in the scheduling queue are cleared as soon as possible when SchedulerQueueingHints is enabled so that scheduler consumes less memory.
New e2e tests for Node endpoints

Deprecated

Graduated feature gates being removed: ValiatingAdmissionPolicy, StableLoadBalancerNodeSet, CloudDualStackNodeIPs, LegacyServiceAccountTokenCleanUp
kubeadm removes the deprecated flag ‘–experimental-output’
kubeadm removes the deprecated sub-phase of ‘init kubelet-finilize’ called experimental-cert-rotation

Version Updates

corefile-migration to v1.0.24

Subprojects and Dependency Updates

prometheus v2.54.1 allow multiple samples on same series, with explicit timestamps
containerd v1.7.21 regenerate introspection UUID if state is empty
grpc v1.66.1 enable EDS dualstack support by default; also v1.66.0

Last Week In Kubernetes Development (LWKD) is a product of multiple contributors participating in Kubernetes SIG Contributor Experience. All original content is licensed Creative Commons Share-Alike, although linked content and images may be differently licensed. LWKD does collect some information on readers, see our privacy notice for details.

You may contribute to LWKD by submitting pull requests or issues on the LWKD github repo.