How to avoid cloud vendor lock-in?

06 Tem 21:44

I ran into this exact problem with a client who built their MVP on AWS. They started with EC2 and DynamoDB, but after a year they wanted to move to GCP for better ML tooling. The nightmare began with hardcoded AWS SDK calls everywhere and S3-specific bucket policies. We had to do a full rewrite of their CI/CD just to swap out the deployment files. It took us two months to untangle everything and ended up using Terraform for IaC so we could spin up the same stack anywhere. Now when I build something new, I treat cloud services like Lego bricks: avoid proprietary stuff like AWS Lambda layers or GCP-specific Pub/Sub extensions. Instead, I stick to open standards (K8s, Postgres) and use Terraform modules we can tweak per provider. For storage, we always use S3-compatible APIs so the data layer never cares which vendor hosts it. Last year, we even migrated a client from Azure to DigitalOcean in a weekend—just had to change a few Terraform variables and redeploy.