Insights

Notes from the infrastructure floor.

Field notes on cloud platforms, GPU fabric, and networking — written by the engineers who build and run it.

Kubernetes

Designing cluster networking that survives a bad Tuesday

What actually breaks in production Kubernetes networking, and how to design around it before it does.

6 min read
 AI Infrastructure

RoCE vs InfiniBand for GPU clusters, in practice

A field comparison of fabric choices for training workloads — beyond the spec-sheet bandwidth numbers.

8 min read
 Platform Engineering

The Terraform module structure we default to

A pragmatic module layout that scales from one environment to a dozen without a rewrite.

5 min read
 OpenStack

Dual-domain LDAP in Keystone, without the footguns

A walkthrough of integrating two LDAP domains into a single Keystone identity backend cleanly.

7 min read
 Reliability

What we check before calling a migration "done"

The unglamorous checklist that catches the failures a clean cutover run doesn't show you.

4 min read
 Automation

Ansible at scale: keeping playbooks boring on purpose

Why the least exciting playbook is usually the safest one to run at 2am.

6 min read