Kubernetes

Our nightly Velero backups reported 'Completed' every single night. Then we actually looked at what was inside them: nothing. Here's how a silent default in local-path-provisioner backed up 550 KB of metadata and zero files for 136 days — and how we found it, fixed it, and proved the fix with a real restore.

When your hosting provider goes down, you have two options: scramble or switch. We built an automated failover system using Cloudflare Workers, KV storage, and an on-site server that flips DNS within two minutes — at zero ongoing cost.

A production migration from bare-metal LAMP to a single-node K8s cluster — covering namespace isolation, ExternalSecrets, cert-manager TLS, and the database name swap that nearly took down two sites simultaneously.