Last Week in Kubernetes Development
Stay up-to-date on Kubernetes development in 15 minutes a week.
Week Ending June 11, 2023
Developer News
Please migrate your test jobs off the old Prow Default Cluster as soon as you can, please.
The Kustomize Project is looking to train new maintainers; apply if you love Kustomize and you can spend some serious time on it.
The Kubernetes Contributor Summit Chicago is looking for event volunteers to help run the Summit. Note that you must be a Kubernetes contributor to be eligible.
Release Schedule
Next Deadline: Enhancements Due, June 16th (01:00 UTC)
Hopefully you opted in your Enhancements for Production Readiness Review, so by this Thursday you’ll have a completed, or at least in-progress, PRR. If not, you’re going to need to seek an Exception.
Patch releases, including a go version bump are expected out this Wednesday.
Featured PRs
#114394: Document versioning of CRI API
SIG-Node has added some documentation to k/k to clarify how CRI as a library and standard will be versioned over time, and how Kubernetes should interact with it. This solidifies the use of SemVer in a similar way as Kubernetes itself and a similar N-3 version skew compatibility rule. It also includes a changelog back through CRI v1 being introduced in Kubernetes 1.20. For both developers and CRI admins, this document is well worth a read to better understand the goals for how CRI will evolve over time.
#117793: use the cgroup aware OOM killer if available
Support for cgroups v2 went GA back in 1.25 but one new feature we haven’t previously made use of is the cgroup-aware OOM support added in kernel 4.19. In short, this mode means that if any task in a cgroup is selected for OOM termination, the whole group is terminated. This avoids common issues with things like forking web servers where a subprocess inside the container OOMs but Kubernetes is unaware of this because PID 1 keeps running. Some tools handle this internally but most don’t and either blindly retry the failing operation or, in the worst cases, corrupt some internal state.
Other Merges
- The number of concurrent Cronjob workers is admin-configurable
- CRI-streaming is now part of the Kubelet staging repo
- Job status updates are batched every 1s to prevent update storms
- PodReadyToStartContainers is the new PodHasNetwork
- SelectHost highest-scoring node using heap, so that we can report the selection details
- kube-proxy: Don’t drop packets when enabling LocalModeNodeCIDR, and more network detection cleanup
- Overhaul how the EtcdOptions config and kube-apiserver as well
- Set the
CloudDualStackNodeIPsannotations correctly - New metrics: webhook conversion
rest.DefaultServerUrlForis reusable now for easy URL generation- If we’re gonna list a ton of volumes in an error message, sort them
- Contextual logging migration: inter-pod affinity
Testing Cleanup: changed behavior of master taint, cleanup cacher testing, DRA node tests
Promotions
Deprecated
--azure-container-registry-configis deprecated; use Credential Providers instead- In-tree CephFS Driver is deprecated
- Unused function getSeccompProfilePath is gone
- Kubelet’s
--provider-id flaghas been un-deprecated; we need it after all
Version Updates
- New Go versions: 1.19.10 in 1.24, 1.25, and 1.26; verson 1.20.5 in 1.27 and 1.28
- distroless-iptables image to v0.2.5
- Kernel Management Module is v1.1.0
- Python client is 27.2.0
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.