Dora Noda
Software Engineer
2304 posts · View all authors
Headlamp's Knative Plugin Surfaces Traffic Splitting and Scale-to-Zero in One View: What a Serverless Runtime Actually Needs From a Cluster API Fleet's Ingress Layer
The June 2026 Headlamp plugin makes Knative traffic splits and scale-to-zero state visible in one view. Here is what those primitives buy a git-push PaaS, the Gateway API stack a Cluster API fleet must run underneath, the cold-start bill tenants pay, and why scale-to-zero should be opt-in for production routes.
Healthchecks.io Walks Off Managed S3: The Egress-and-Request-Fee Math of Self-Hosting Object Storage
Healthchecks.io moved 14M objects off managed object storage onto one dedicated server — and paid more. The real math: ~$390/mo in S3 PUT fees vs $2.74 storage, the egress trap priced out, what self-hosting adds to the pager, and when running state on your own metal wins.
The 4GB RAM Price Gap: What Hetzner at $4 vs Heroku at $260 Actually Buys You
A 4GB-RAM machine costs $4 on Hetzner and $250+ on Heroku. Re-verified 2026 prices, an itemized look at what the 14x–125x premium buys, and a break-even table that prices in your own ops hours.
Hetzner Raised Prices Three Times in Six Months: The "Just Use a Cheap Hetzner Box" Math, Recomputed
Hetzner repriced three times in six months, pushing dedicated-vCPU and bare-metal lines up 2.3-3.7x on June 15, 2026. A line-by-line recompute of the self-hosting-vs-managed-PaaS math on the new price list — including the tier where an 11x advantage compressed to 4x, and why self-healing fleets forfeit grandfathered pricing.
Metering Your Tenants: Build a Kubernetes Metrics Exporter for Railway-Style Usage Showback
Build a small Go exporter that turns cAdvisor's CPU, working-set memory, and network counters into per-app usage statements priced at Railway's public meters — plus the five cAdvisor pitfalls that silently corrupt naive per-container accounting.
Kubernetes v1.37's Deprecation Sweep: The Migration Inventory a Cluster API Fleet Needs Before August 26
Kubernetes v1.37 lands August 26, 2026, and the July 31 sneak peek already names every deprecation and breaking change. Here is the full migration inventory for a Cluster API fleet — ipvs kube-proxy, static-pod Secret refs, cgroup v1, SELinuxMount GA — with the audit command and deadline for each.
45-Day Let's Encrypt Certificates Are Here: The Renewal-Automation Bar Every Self-Hosted PaaS Now Has to Clear
Let's Encrypt's tlsserver profile dropped to 45-day certificates on May 13, 2026, and the default follows by 2028. A five-point audit of what a halved renewal window demands from a self-hosted PaaS's ACME pipeline: percentage-based margins, retry isolation, external expiry alerting, ARI, and 1-hour authorizations.
OpenAI's Presence Ships With Engineers, Not a Deploy Button: What the Gap Between Enterprise Agents and Git-Push Reveals
OpenAI launched Presence with engineer-led deployments instead of a self-service console. A decomposition of its six-stage process shows which half genuinely needs consultants — and which half is deploy-and-operate mechanics a git-push platform already made a one-command path.
OpenCost 1.121 Brings Inference Cost Tracking to Kubernetes: Metering GPU Token Workloads Like Any Other Tenant
OpenCost 1.121 adds per-inference cost metering to Kubernetes, so self-hosters can finally measure dollars per million tokens per tenant. We decompose owned vs leased vs rented GPU vs API costs, find the utilization crossover, and wire vLLM token metrics into a multi-tenant showback report.