Multitenancy trade-offs, plus posts on DDL vs DML and Docker Sandboxes

A day split between a hard architecture question on a client system and turning recent lessons into public posts.

Client project (NDA)

The big question was multitenancy. With one client today and a plan to grow toward a few hundred, I worked through the trade-off between one isolated infrastructure per company and a single shared database that separates tenants at the query layer. The honest starting point is that the data we have today already fits a single tenant, so the decision is about where the product is going, not where it is. I surveyed open-source references (Twenty, NocoBase, ERPNext, Odoo) not to adopt any of them, but to study how mature systems model dynamic entities, roles, and permissions when one codebase serves many tenants. Some of the day was also plumbing: folding meeting notes into tracked issues so the decisions do not live only in my head.

tgmarinho-ai-website

I shipped two bilingual posts: DDL vs DML, and running builds inside Docker sandboxes. The DDL vs DML one started as a small confusion I cleared up while writing, which is usually a sign the post is worth publishing.

I also spent time researching UUID versus sequential IDs across many sources: database engineers, Postgres, Snowflake, Cassandra, MongoDB, and a few ERP and cloud writeups. The question I wanted the post to answer is not "which is better" but when to prefer one, and who has migrated between them and why.

personal-os

A small side thread: converting a legacy Microsoft Access database into a clean Postgres DDL. It is the kind of one-off data plumbing that is worth turning into a repeatable recipe instead of a manual chore.

The through-line all day was the same: a good default is still a decision, so it should be able to explain itself.