Right now someone could execute test.sh in a production container and (partially) destroy the production database. Clearly this is less than desirable and needs to be addressed ASAP.

We've not settled on a solution, but some candidates are:

• If tables have more than 10k rows (more rows than tests create), abort.
• Check for the existence of tables/schemas when tests start, and abort if present.
• Check for the existence of a hand created table that indicates the DB server is a production server.

We'll post more details here on how we plan to solve this before we start implementation.