Dora Noda
Software Engineer
2304 posts · View all authors
AWS DevOps Agent Goes GA But Still Can't Deploy the Fix: The Diagnose-Not-Act Line
AWS DevOps Agent reached GA finding root causes in minutes — but it still refuses to execute the fix. Why every enterprise agentic-ops vendor stops at diagnosis, and the five pieces of platform machinery needed before an AI agent can safely deploy the rollback itself.
Bitwarden's July 2026 Update Broke Every Vaultwarden Overnight: The Real Cost of Reimplementing a Protocol You Don't Own
Bitwarden's 2026.7.0 client silently repurposed a wire-format field Vaultwarden had emitted for eight years, breaking every self-hosted deployment overnight. A concrete cost accounting of running a compatible reimplementation of a protocol you don't control — and why bex's Render-compatible API carries the same risk.
Cloudflare Containers Bill 5-10x More Than an Always-On VPS: What the 10ms Meter Actually Costs a Long-Running Service
A line-by-line recompute of what an always-on service costs on Cloudflare Containers' 10ms meter versus a flat-rate VPS: $35-58 vs $7-15 for a small web+worker, a ~9-hour-a-day duty-cycle crossover, and why memory billed on provisioned size makes the gap structural.
Cloudflare OS Is Apache-2.0 — But the Only Supported Substrate Is Cloudflare's Network
Cloudflare OS launched Apache-2.0 with zero-access agents and Gatekeeper Workers — but the only supported production path deploys into a Cloudflare account. A component-by-component portability ledger, the Hacker News stress test, and a four-question audit separating open code from an open substrate.
Coolify v4 Shipped an MCP Server. Dokploy Didn't. The Self-Hosted PaaS Race Just Split in Two
Coolify's v4.x releases made a native MCP server its headline differentiator while Dokploy stayed deliberately minimal. A side-by-side comparison of both platforms — idle RAM, install, MCP surface — and why agent-operability's real cost is API discipline, not footprint.
Crossplane vs Cluster API: Two CNCF Answers to "Should Infrastructure Be an API?" — and Where a Self-Hosted PaaS Draws the Line
Crossplane models everything from databases to GPU inference fleets as one declarative API; Cluster API keeps a written list of what it refuses to manage beyond cluster lifecycle. A side-by-side of what each CNCF project actually optimizes for, why they compose as layers rather than compete, and why a self-hosted PaaS builds on Cluster API while skipping multi-cloud abstraction.
Daytona's 90ms vs E2B's 150ms: What a 40% Cold-Start Gap at Price Parity Actually Means at 10,000 Sandboxes a Day
2026 benchmarks show Daytona starting sandboxes in ~90ms vs E2B's ~150ms at identical ~$0.0504/vCPU-hour pricing. The worked math at 1K–100K sandboxes a day shows the gap costs pennies and only matters for serial agent loops — concurrency caps, isolation, and lifecycle hygiene decide the rest.
Heroku Is in Sustaining-Engineering Mode. Dokku Is the Obvious Exit — Here's What It Actually Gets You (and What It Doesn't)
Heroku's February 2026 shift to sustaining engineering has teams planning exits. A grounded look at migrating to Dokku: the Heroku-to-Dokku concept map, the honest limits of a single-server PaaS, the ten-command migration, and where the one-box ceiling ends.
Gitea's Docker Image Trusted Every Proxy on Earth: A Five-Question Audit for the Defaults Baked Into Your Images
Gitea's container image shipped REVERSE_PROXY_TRUSTED_PROXIES = * while its own docs said loopback-only — CVSS 9.8, armed by the very SSO flag a platform is supposed to set. Five runnable questions, plus a CI check, for auditing the defaults hiding in every image and chart your platform ships.