640 posts tagged with "Kubernetes"
Container orchestration, Cluster API, and self-hosted control planes
OpenCost 1.121 Finally Answers: What Does Each Token Cost on Your Own GPUs?
OpenCost 1.121 adds Kubernetes-native per-token inference metering via llm-d — allocation vs. usage cost, KV-cache-corrected — turning self-hosted LLM spend from a quarterly guess into a Prometheus metric you can alert on.
Proxmox VE Evaluations Up 340%: What the VMware Exodus Means for Your Next Cluster API Provider
Gartner reports Proxmox VE evaluations up 340% as Broadcom's VMware bills rise 800-1500%. What Proxmox 9.1's OCI-to-LXC feature actually does, how CAPMOX compares to CAPH, and when a second Cluster API provider earns its place in a self-hosted fleet.
vLLM vs Ollama in Production: What PagedAttention's 19x Throughput Gap Really Buys on Owned GPUs
Red Hat's 2026 benchmark put vLLM at 793 tokens per second and Ollama at 41 on the same GPU and model — a 19x gap. The same comparison at one concurrent user shows them within 20%. A concrete breakdown of where PagedAttention earns that gap, where it doesn't, and what each engine costs to run on an owned Hetzner GPU fleet.
Who Backs Up the Management Cluster? Rebuilding the CAPI Brain When It Dies
Your workload clusters keep serving traffic when the management cluster dies — but nothing can scale, heal, or upgrade until you rebuild the brain. A line-by-line walkthrough of three recovery paths with concrete commands, the stale-state trap, and a quarterly drill to prove it works.
Waking Idle Apps at the Proxy Layer: Sablier vs KEDA vs Knative for the 30-App Fleet
Sablier wakes idle containers at the reverse-proxy layer — Traefik, Caddy, Nginx — with no operator or sidecar. How its blocking and waiting-page strategies compare to KEDA and Knative for a fleet of dozens of low-traffic apps, what idle-hour math actually saves, and where HTTP-only wake hits its ceiling.
Docker Hub Raised Prices 80 Percent, Then Cancelled Rate Limits — Why Your CI Still Needs a Mirror
Docker Hub raised Pro 80 percent and Team 67 percent, then cancelled planned pull charges — but 100 pulls per 6 hours still blocks burst CI. A line-by-line cost and failure-mode comparison of paying Docker, riding free-tier limits, or running Harbor as a pull-through cache on hardware you already own.
AWS Fixed the Network in 20 Minutes. Your App Stayed Down for 10 Hours.
AWS restored routing in 20 minutes on July 24, 2026, but downstream providers stayed down for 10 more hours. Why a provider's green checkmark understates tenant recovery, the three mechanisms that stretch a blip into a workday, and how a single-tenant fleet on owned Hetzner hardware recovers differently.
Kubero's PaaS on Kubernetes: What a Dashboard Over kubectl Actually Solves — and What You Still Operate Yourself
Kubero gives you a polished PaaS surface on Kubernetes — git-push deploys, pipelines, and add-ons without Helm — but it assumes you already own a healthy cluster. Here is exactly what the dashboard automates, what it leaves you holding, and where a Cluster API fleet closes the gap.
Your Build Queue Is Idle 80% of the Day: Sizing a Git-Push Pool Without Paying for Peak
A git-push PaaS build queue is bursty by nature — peaks 4–6x the average. A fixed Hetzner pool sized for peak sits idle 78% of the day; sized for average, it queues 10+ minutes at morning burst. A worked comparison of three sizing strategies and the hybrid math that beats both.