Thoughts on DevOps, cloud architecture, and software engineering.
If you're migrating a PostgreSQL database hosted on Supabase using AWS DMS, pglogical will fail immediately — Supabase doesn't grant superuser. Here's why pgoutput is the right protocol for this setup, and exactly how to configure it.
If deploying the same Docker image to dev and production requires code changes between environments, your configuration is in the wrong place. Here's the pattern I use to build once and promote the exact artifact — same SHA, same bytes — across isolated AWS accounts with no edits.
Replication slots silently accumulate WAL data until something consumes them. On managed PostgreSQL services like Supabase, an abandoned slot will eventually get dropped — taking your DMS task with it. Here's what's actually happening, and how to keep it alive.
API Gateway 403s are frustrating because they look identical regardless of cause — and there are at least four distinct places they can originate. Here's the diagnostic framework I use, and the two specific issues that caught me during a production setup.
When every application repo owns its own CI/CD workflow, you end up with N copies that can never all be updated at once. Putting the deployment logic in the infrastructure repo gives you a single source of truth — and means fixing a bug in the pipeline fixes it everywhere simultaneously.
Most Terraform modules don't get reused — they get copied, modified, and eventually forked into something unrecognisable. Here's the interface design principles that made a module actually stick across 12 production pipelines.