Kubernetes engineering blog

Loading
Scheduling the removal of the kube-proxy IPVS backend
Dan Winship of Red Hat explains why the Kubernetes kube-proxy ipvs backend is being deprecated and scheduled for removal in Kubernetes 1.40, covering its original data-plane and control-plane performance motivations and its later compatibility, batch-update, and scheduler limitations. The post recommends migrating to the nftables mode, or to the substantially improved iptables mode on older clusters, and gives concrete configuration guidance for kubeadm clusters.
From Working Group to SIG Architecture: spotlight on AI Conformance
This SIG Architecture spotlight interviews Janet Kuo and Yuan Tang about how the Kubernetes AI Conformance effort moved from a temporary Working Group to a permanent SIG Architecture subproject. It explains the tiered graduation model that introduces AI infrastructure requirements as SHOULD and later promotes them to MUST, and the decision to test infrastructure capabilities such as Dynamic Resource Allocation rather than specific frameworks.
Announcing the Workload Conformance Working Group
The Kubernetes community has formed a Workload Conformance Working Group to write a shared, objectively testable specification for how applications behave at runtime, covering security, disruption resilience, resource use, networking, storage, API usage, and observability. Sponsored by SIG Architecture and SIG Apps, the group plans to deliver a per-minor-version specification, a testing method, and upstream tooling, and is seeking input from cluster operators and application maintainers.
Why Is My Kubernetes Job Still Pending? Tracing Kueue Admission with Headlamp
A Kubernetes blog walkthrough of debugging why a Kueue-managed Job stays suspended, separating Kueue admission from Pod scheduling. It traces the Workload, LocalQueue, ClusterQueue, ResourceFlavor, and Cohort path, using a quota mismatch as the worked example. The post also explains which problems are admission-capacity issues and which belong to the kube-scheduler.
The Shift to cgroup v2 in Kubernetes: What You Need to Know
This guide explains Kubernetes’ migration from cgroup v1 to v2, including compatibility requirements, kubelet and runtime configuration, memory QoS, OOM behavior, PSI metrics, and CPU weight changes. It also provides troubleshooting commands and guidance for validating resource enforcement during migration.
Scaling Kubernetes Workloads with Node Swap
The authors benchmark Kubernetes node swap backed by NVMe Local SSDs across kernel builds, headless browser sandboxes, and isolated Python runtimes. They explain cgroup v2 accounting and LimitedSwap configuration, showing up to 3× higher pod density while identifying the latency trade-offs of overcommitting active memory.
Spotlight on SIG Apps
Kubernetes SIG Apps chairs explain how workload controllers such as Deployments, StatefulSets, Jobs, and DaemonSets manage application lifecycles and resilience. The discussion covers node failures, coordinated restarts for distributed AI workloads, backward-compatible API evolution, and the proposed KEP-4443 enhancement for distinguishing Job failure causes.
Kubernetes v1.37: Pod-Level Resource Managers graduated to Beta
Kubernetes v1.37 graduates Pod-Level Resource Managers to beta, enabling the Kubelet’s CPU, memory, and topology managers to use pod-level resource declarations for NUMA-aware placement. The update supports hybrid allocation for primary containers and sidecars and adds pod-level exclusive assignment reporting to the PodResources API.
Kubernetes v1.37: Memory QoS Graduates to Beta
Kubernetes v1.37 promotes Memory QoS to Beta and enables its feature gate by default while preserving existing workload behavior through a null memory throttling default. The post explains cgroup v2 configuration for throttling and tiered reservation, upgrade considerations, and the node-wide limitations of memory protection.
Kubernetes Changed Block Tracking API - Beta Differences
Kubernetes’ Changed Block Tracking API has moved from Alpha to Beta with the v1.0.0 release of external-snapshot-metadata. The post details the v1beta1 CRD migration, compatibility requirements, required upgrade steps, and how CSI drivers and backup applications can try the API.
Kubernetes v1.37: Native Histograms Graduates to Beta
Kubernetes v1.37 enables Prometheus native histograms by default, providing higher-resolution quantiles with fewer time series and lower storage overhead than classic bucketed histograms. The post explains dual exposition, exponential bucket configuration, PromQL queries, Prometheus scraping options, dashboard migration, and rollback strategies.
Kubernetes v1.37: Scheduler Preemption for In-Place Pod Resize (Alpha)
The post explains how Kubernetes v1.37 enables scheduler preemption for deferred in-place Pod resizes, allowing higher-priority workloads to reclaim capacity from lower-priority Pods without restarting. It covers the scheduling architecture, node-level controls, race handling, feature-gate requirements, and a kind-based demonstration.
Kubernetes v1.37: Introducing Node Lifecycle Conditions
Kubernetes v1.37 introduces five well-known Node lifecycle conditions for reporting drains, maintenance, and graceful shutdowns through a shared status channel. The post explains their semantics, current alpha limitations, operational usage, and how they could enable more lifecycle-aware controller behavior.
Kubernetes v1.37: Advancing Workload-Aware Scheduling
Kubernetes v1.37 advances Workload-Aware Scheduling with Beta Workload and PodGroup APIs, gang scheduling, workload-aware preemption, and shared DRA ResourceClaims. It introduces hierarchical CompositePodGroups, multi-level topology constraints, controller integration APIs, and explicit Job scheduling configuration for complex distributed workloads.
Kubernetes v1.37: DRA Updates
Kashish Verma details the Kubernetes 1.37 Dynamic Resource Allocation updates, including GA support for extended resources, device taints, standardized attributes, and workload ResourceClaims. The post also covers new alpha features such as derived attributes, compatibility groups, fractional capacity, and scheduler performance improvements.
Kubernetes v1.37: etcd RangeStream Cuts Memory Use on Large List Reads
The post explains how Kubernetes v1.37 uses etcd v3.7’s RangeStream RPC to stream large list reads in adaptive, byte-bounded chunks, reducing peak memory use in etcd and the API server. It covers feature-gate requirements, fallback behavior, and the metrics query for confirming that streaming is active.