i totally get that position. I used to be in that camp, but in my last 3 jobs at companies of >10mil users, we had to get rid of FKs for other reasons (perf, dist system stuff). I kind of expected bad data corruption issues.... but they just never happened ¯\_(ツ)_/¯. Not sure if that was strong cultures of automated testing, or what. So now i no longer think they're worth the tradeoffs.
YMMV with this one, i could see how it might pan out differently in other environments
Did these involve financial transactions? Other situations I will have dangling rows without issues, but whenever money comes into the picture I like to have as much consistency enforced as I can. But I'm aware that doesn't play nicely with distributed systems.
Yes actually. financial transactions are often used as an example that demands strong consistency. But it’s not true in the real world for the most part. for any given money movement (a debit and a credit) there’s a good chance one of those legs is over the network - you’re either interacting with a third party api, or a proprietary internal product, or similar. So you can’t have transactional consistency, or fks. Further, for every movement there is layers of reconciliation that happens hours or days later, that fixes any inconsistency. Finance / fintech is actually among the most eventually consistent systems out there, so it’s sort of ironic its use in a lot of transactional examples.
Exception is true banking ledgers maintained at banks, but even then you’d be surprised. Entirely different systems handle loans and checking accounts
YMMV with this one, i could see how it might pan out differently in other environments