Blog
Notes from the build
Engineering notes from Austin on container orchestration, deploy reliability, edge routing, and running Kubernetes without a dedicated platform team.
14 notes · Newest first
· Infrastructure · 5 min read · By Ethan Cole
Edge routing and TLS automation without the ingress tax
Most teams only want HTTPS in front of their web service and clean connection draining during a deploy. They should not have to maintain an ingress controller to get it.
· Reliability · 5 min read · By Ethan Cole
Why container abstractions fail at 2 a.m.
Developers do not want a black box that hides logs, networking, and memory limits. They want sane defaults when deploys go right, and plain host state when something breaks.
· Architecture · 5 min read · By Ethan Cole
Scale-to-zero without surprising user-facing traffic
Autoscaling is only helpful when it respects clear floors and ceilings. Here is how we separate background workers that can sleep from APIs that cannot.
· Infrastructure · 5 min read · By Ethan Cole
Multi-environment secrets without dotenv drift
Staging and production issues often trace back to the same mistake: a copied .env file that drifted, or a release step that leaked credentials across boundaries.
· Architecture · 6 min read · By Ethan Cole
What runs in our control plane, and what stays in your cloud account
Running containers on AWS, GCP, or Azure works best when the boundary is explicit: Nimbus coordinates rollouts, while your clusters, images, and data stay yours.
· Product · 5 min read · By Ethan Cole
Dependency-ordered deploys: starting Postgres before the API
Container start order is not readiness. A service graph only stops boot-time crash loops when upstream dependencies pass health checks before downstream services start.
· Reliability · 6 min read · By Ethan Cole
Readiness probes, graceful shutdowns, and boring rollouts
Zero-downtime deploys rarely fail because of the scheduler. They fail because an app ignored SIGTERM or received traffic before its dependencies were warm.
· Product · 5 min read · By Ethan Cole
Every deploy is a Git commit: designing one-step rollbacks
Rebuilding a container to roll back production wastes critical minutes when latency spikes. Promoting and reverting immutable commit SHAs makes recovery predictable.
· Architecture · 5 min read · By Ethan Cole
What “Kubernetes-compatible” actually means
Compatibility is not a kubeconfig and a wave goodbye. It is ordinary images, a real cluster in your account, and a way out.
· Reliability · 5 min read · By Ethan Cole
Logs, metrics, and traces in one place before the first incident
Small engineering teams rarely skip observability on purpose. They skip it because stitching three telemetry agents into every service eats a sprint.
· Product · 5 min read · By Ethan Cole
Keeping a visual service graph and a YAML file in sync
Visual infrastructure tools usually rot because they fight Git. A canvas only holds up in production when the diagram and service.yaml compile from the same schema.
· Architecture · 6 min read · By Ethan Cole
Leaving Docker Compose without adopting a cluster
Compose is a great way to run a laptop. It is a thin way to run production. Here is the gap teams actually hit.
· Company · 5 min read · By Ethan Cole
Why a five-person team builds simpler defaults
Staying lean and bootstrapped forces an engineering constraint that matches our users: every config knob we add is one more failure mode both of us have to carry.
· Company · 5 min read · By Ethan Cole
Why we built Nimbus
We started Nimbus in Austin because product engineers were spending quarters learning ingress instead of shipping.