513 posts tagged with "Kubernetes"
Container orchestration, Cluster API, and self-hosted control planes
Daytona Went Closed-Source: What 'Self-Hostable' Means Now That the Control Plane Is Private
Daytona moved its core to a private codebase in June 2026, freezing the public repo at v0.190.0. A before/after accounting of what self-hosting still covers, a five-question checklist for any vendor's claim, and how E2B, Modal, Beam, Microsandbox, and the Nightona fork score.
Dragonfly Without the Database: Stop Shipping the Same Gigabytes to Every Node
A 10-node fleet pulling a 1.2 GB image downloads 12 GB per deploy; P2P distribution serves it once. We price the duplicate-byte tax across fleet sizes, unpack Dragonfly's no-database lightweight mode, and compare it with Spegel, Kraken, and a plain registry mirror.
GPU Autoscaling on Kubernetes With KEDA: Scaling on the One Metric HPA Can't See
HPA can't see GPU utilization — nvidia.com/gpu is an integer device count, so a vLLM pod at 8% CPU and a saturated GPU never scales. A worked KEDA pattern: dcgm-exporter to Prometheus to ScaledObject YAML, a minimal external scaler serving utilization plus queue depth, the scale-to-zero cost math, and where KEDA stops and DRA begins.
Headlamp's Cluster API Plugin: The Free Ops Console Your PaaS Team Doesn't Have to Build
Kubernetes retired its Dashboard in January 2026 and Headlamp's June 2026 Cluster API plugin gives small platform teams a free, upstream-maintained fleet-ops console — Machine health, scaling, and ownership maps included. What to adopt, what remains alpha, and why tenants still need their own dashboard.
Two Dashboard Plugins Exposed the Scheduler Gap Every Git-Push PaaS Ignores Until the First GPU Job
Headlamp shipped dashboard plugins for Volcano and Kubeflow eighteen days apart — Job/Queue/PodGroup map views and API-server-native ML panels. A concrete scheduling-gap table, the three primitives a git-push PaaS lacks, and the exact tripwire for adopting Volcano queues.
IPVS Is Dying in Kubernetes: Your Migration Timeline From the v1.37 Feature Gate to v1.43 Removal
Kubernetes 1.37 introduces the KubeProxyIPVS feature gate, starting a staged removal that disables IPVS by default in v1.40 and deletes it in v1.43. The full timeline, what breaks at each gate, and a detect-switch-validate recipe for moving self-hosted clusters to nftables mode.
Stop Buying a Control-Plane Machine Per Tenant: What Kamaji's Pod-Based Control Planes Actually Save
Running each tenant's Kubernetes control plane as pods instead of dedicated machines cuts the per-tenant control-plane bill from €7.50 a month to under €2 — starting at about six tenants. A worked Hetzner-priced model of Kamaji's consolidation, its limits, and the isolation tradeoff that decides whether the saving is safe to take.
kro Hits CNCF's 'Adopt' Ring: What Declarative CRD Composition Buys a PaaS Control Plane That Wants to Stop Hand-Writing Operators
CNCF's Q1 2026 Technology Radar put kro in the Adopt ring alongside Helm and Backstage. What declarative CRD composition actually buys a Cluster API-based PaaS — the 40-line RGD that replaces a hand-written expansion controller — and the four layer-on-layer costs to price in first.
Your Scheduler Config Has No Staging Environment: Rehearsing Bin-Packing Changes With kube-scheduler-simulator
Scheduler config changes ship straight to production because a dev cluster cannot reproduce real contention. kube-scheduler-simulator mirrors your fleet's actual nodes and pods so you can replay bin-packing changes and read per-plugin placement verdicts before anything moves.