LWKD: Week Ending August 11, 2024
Week: 2024-08-11
Developer News
It's Release Week! Kubernetes 1.31 "Elli" is released, with many new features. In addition to the list of features in the main blog post, note that cgroups v1 is going into maintenance, several things have been removed (most notably Ceph in-tree driver), and the addition of lastTransitionTime for PVs. More 1.31 features below.
Steering Committee nominations are open.
The Kubernetes Contributor Summit is looking for artists to create designs. Registration and CfP is still open.
Release Schedule
Next Deadline: v1.31.0 release day, August 13th
Kubernetes 1.31 was released on August 13.
Patch releases are expected later this week.
Lesser-known 1.31 Features
These features didn't make the 1.31 release blog, but are interesting to contributors:
4355: Coordinated Leader Elections
This Enhancement makes control plane leader elections function in a way that is compatible with upgrading one control plane component at a time, by keeping everyone on the old APIserver until everything else is upgraded. This should make for a smoother, and more reliable, upgrade experience. Alpha and opt-in only for 1.31.
4368: Job API managed-by mechanism
A small part of the MultiKueue initiative of the Kueue job manager, this enhancement adds tracking for which controller "owns" a job. While potentially useful for any multi-controller environment, the change is intended to make multi-cluster job scheduling possible. Alpha in 1.31.
4176 and 4622: HPC Features
Two features make Kubernetes more useful on bigger, beefier machines. We can spread hyperthreads across physical CPUs, making better use of high-core-count machines. And we can configure topology rules for more than eight NUMA nodes, supporting very high memory systems. 4176 is Alpha and 4622 is Beta in 1.31.
KEP of the Week
KEP 4420: Retry Generate Name
This KEP implements automated retry of generateName create requests when a name conflict occurs. Despite generating over 14 million possible names per prefix with a 5-character random suffix, conflicts are frequent, with a 50% chance after 5,000 names. Currently, a conflict triggers an HTTP 409 response, leaving it to clients to retry, which many fail to do, causing production issues.
This feature became Beta in 1.31.
Subprojects and Dependency Updates
- containerd v1.6.35 regenerate UUID if state is empty in introspection service
- prometheus v2.54.0 remote-Write: Version 2.0 experimental, plus metadata in WAL via feature flag metadata-wal-records; also v2.53.2