Dora Noda
Software Engineer
2325 posts · View all authors
CVE-2026-50566: How Namespace-Scoped RBAC Became a 9.9 Node Compromise in Fission
A routine namespace-scoped RBAC grant let a Fission tenant escalate to node compromise — why RBAC was never a sandbox boundary, and what it means for platforms scheduling tenant code on shared nodes.
Gateway API on kind: The Local Reproduction Recipe for Ingress Bugs You Can't Debug on Bare Metal
The stock kubernetes.io recipe for running Gateway API on kind gets you one working HTTPRoute. Here's how to extend it to reproduce the three tenant ingress bugs that actually page a self-hosted PaaS's on-call, before they hit bare metal.
GitHub Copilot's Worktree Isolation Is Not a Sandbox — Why That Matters for Deploy-Authority Agents
GitHub's Copilot app isolates parallel agent sessions with git worktrees, not sandboxes — fine for code edits, but not enough for an MCP server that can deploy and rollback real infrastructure. Here's the concrete difference and what it means for agent-operated PaaS platforms.
Do You Still Need kubectl for Cluster API Fleet Management? Headlamp's New Plugin Says Mostly No
Kubernetes retired its Dashboard and pointed operators at Headlamp's new Cluster API plugin — here's exactly what it replaces, what still needs kubectl and GitOps, and why bex should adopt it instead of building a bespoke fleet view.
Hetzner's GEX131 GPU Server: What €889/Month for a Blackwell RTX PRO 6000 Buys Against AWS's g5.xlarge
Hetzner's GEX131 costs more per month than AWS's g5.xlarge — but a third as much per gigabyte of VRAM. Here's the worked cost-per-GB math across on-demand, spot, and reserved AWS pricing, plus what 96GB of VRAM actually unlocks that 24GB can't.
K3s vs K0s Under Cluster API: Which Lightweight Distro Actually Fits a Hetzner Fleet's Machine Lifecycle
Most k3s-vs-k0s comparisons are written for a single Raspberry Pi. Under Cluster API on Hetzner, the real fork is where each distro puts your control plane — and it changes your upgrade and cert-rotation story more than either binary's size does.
Kamaji Turns Kubernetes Control Planes Into Pods — But It Only Cuts the Control-Plane Bill, Not the Worker-Node One
A worked cost model for Kamaji's Cluster API control-plane provider: what running tenant control planes as pods actually removes from a self-hosted fleet's bill, what it leaves untouched, and where its shared-kernel isolation stops being enough.
Kamal Supports Cloud Native Buildpacks, Not Just Dockerfiles: What That Means for bex's Own Build Path
Kamal, 37signals' zero-daemon deploy tool built specifically to avoid PaaS abstraction, now ships a pack builder option for Cloud Native Buildpacks alongside its default Dockerfile path. Here's the actual config, why a minimal-abstraction tool reached for a CNCF spec instead of writing its own builder, and what it means for bex's default.
Kubernetes 1.36 Just Removed gitRepo Volumes and IPVS kube-proxy — Here's the Audit Your Cluster API Fleet Needs Before You Upgrade
Kubernetes 1.36 permanently kills gitRepo volumes (closing a CVE-2024-10220-class RCE) and removes IPVS kube-proxy mode as hard failures, not warnings. Here are the exact audit commands a Cluster API fleet needs before upgrading, including the CAPI config gap that hides IPVS on Hetzner clusters.