7 hours ago
- The author used a SaaS/PaaS stack (Supabase, Vercel, Render) for over three years without traditional cloud providers like AWS or GCP.
- Changes in Render's pricing (reduced bandwidth from 100GB to 5GB) and higher costs for managed compute prompted a migration to AWS.
- Terraform was adopted from the start to manage multi-project, multi-environment infrastructure repeatably.
- Backend services moved from Render to ECS on Fargate, with GitHub Actions for CI/CD and RDS for the database.
- A key issue was RDS automatic password rotation causing production outages because the application used a static DATABASE_URL instead of the managed secret.
- The database experience on RDS was inferior to Supabase's table viewer and SQL editor, making casual data inspection cumbersome.
- The migration saved about $15-20 per month but introduced significant time spent on learning, debugging, and maintaining infrastructure.
- The author concludes that moving to a cloud provider is only worthwhile when the team has enough bandwidth to own the infrastructure long-term.