Dora Noda
Software Engineer
2304 posts · View all authors
The 5x Spread on an Identical Workload: What Railway, Render, Fly.io, and Vercel Actually Bill
A June 2026 audit priced the same API, worker, Postgres, and 200 GB egress workload across Railway, Render, Fly.io, and Vercel and found a 5x bill spread. The invoice, why it hides, and a model to price your own migration.
Fly.io After the Free Tier Died: What Per-Second Usage Billing With New 2026 Line Items Costs Against a Fixed Hetzner Invoice
Fly.io killed Hobby/Launch/Scale plans in October 2024 for pure per-second billing, then added new metered line items through 2026 — what that does to bill predictability for a small team versus a flat Hetzner invoice, priced workload by workload.
Fly.io at $11M ARR With Zero Open Roles: How to Read a PaaS Vendor's Vital Signs Before You Bet Your Stack
Mid-2026 snapshots peg Fly.io around $11M ARR with no open roles while Render touts a $1.5B valuation on 100%+ YoY growth — and Fly's own plan deprecations plus new 2026 billing lines look like optimizing the base, not sprinting. A five-signal checklist for auditing any PaaS before you migrate, a scored walkthrough on Fly.io, and why the hedge isn't multi-cloud but a real exit path you can test.
Mooring Skipped the Orchestrator: Why a Typed YAML PaaS Still Hits a One-Machine Ceiling
Mooring ships a security-first, typed mooring.yaml control plane that owns your Compose file and handles HTTPS, monitoring, and self-healing — without Swarm or Kubernetes. What that model gets right on one machine, and the five things that concretely break the moment you need a second.
Vercel Repriced Four Times Since 2024: Why a Moving Formula Hurts More Than a High Price
Four Vercel repricings since 2024 reshuffled which metered line item dominates a bill, so the same workload costs different amounts quarter to quarter. A vintage-by-vintage ledger, a worked variance model, and what a fixed-price Hetzner contract does to a 12-month hosting forecast.
Railway Finally Shipped DNS Logs: What You Can Debug From a Dashboard vs What You Get From CoreDNS on Day One
Railway's July 24, 2026 changelog added DNS logs after years of opaque name-resolution failures. A side-by-side tour of debugging ndots, search-domain expansion, and stale service records on rented DNS visibility versus CoreDNS logs, metrics, and tcpdump on a cluster you own.
Analytics for the Price of Disk: Self-Hosting Umami as a Tenant Workload Instead of Paying Per Event
Product analytics is just a web service plus a Postgres table. A dollar-for-dollar comparison of Plausible, PostHog, and GA4 bills at 100K, 1M, and 10M events against running Umami or Plausible CE as a tenant workload where 10 million pageviews cost cents in disk.
Railway's Access Groups Are Group RBAC Behind a SaaS UI: Here's the Same Boundary as Kubernetes YAML
Railway shipped Access Groups on August 14, 2026 — group-based RBAC for its hosted PaaS. A concrete mapping of the same boundary as Kubernetes RBAC, RoleBindings, and Gateway API policy on a control plane you own.
Healthchecks.io Left the Cloud Bucket: What a Real SaaS Paid to Bring Object Storage Home
Healthchecks.io moved production object storage off managed cloud in April 2026. A case-study read of the egress and replication trade and a decision framework for when owning the disks beats renting the bucket.