DEVOURED
Kubernetes v1.37: Garhwal
Kubernetes v1.37 'Garhwal' arrives with 67 enhancements, graduating metrics.k8s.io and KYAML to stable while introducing pod-level checkpointing in alpha.
What: The release team delivered 16 stable features, including metrics.k8s.io, KYAML, and several Dynamic Resource Allocation (DRA) enhancements. Key updates include scale-to-zero for HorizontalPodAutoscaler (beta), pod-level checkpoint/restore (alpha), and support for manifest-based admission control (beta).
Why it matters: This release continues the project's push toward 'stable-by-default' APIs and improved resource management, specifically targeting the needs of AI/ML workloads and large-cluster scalability through better etcd interaction and stale controller mitigation.
Takeaway: If you are running clusters in ipvs mode, begin planning your migration as it has been officially deprecated; also, prepare to transition off of kube-dns, which will stop receiving package updates after v1.40.
Deep dive
- API Machinery: Metrics API v1 is stable, and concurrent watch object decoding is now enabled by default to reduce initialization latency.
- Storage: StorageVersionMigration is stable, and PVC 'last used' tracking is now beta, helping identify idle storage.
- Networking/Proxy: kube-proxy now uses netlink for nftables, improving management performance, and adds localhost access support.
- Scheduling: Gang scheduling is now beta, along with workload-aware preemption to better support AI/ML batch jobs.
- Node: cgroup v1 support is being phased out, with v1.37 failing to initialize by default on nodes using legacy cgroups without an override.
- Security: Pod certificates and ClusterTrustBundles are now stable.
Decoder
- KYAML: A constrained, safer subset of YAML designed for Kubernetes that is fully compatible with standard parsers.
- Gang Scheduling: A strategy where a collection of pods is scheduled only if the entire group can be placed, preventing partial deployment deadlocks.
- DRA (Dynamic Resource Allocation): A framework for managing specialized hardware resources (like GPUs) that moves beyond simple device plugins to allow more complex allocation logic.
- KEP (Kubernetes Enhancement Proposal): The formal design document process used to propose and track new features in the Kubernetes ecosystem.
Original article
Full article content is not available for inline reading.
Explore all 55 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi