If you never tracked what happened in production, it may have worked most of the time well enough that you never saw how bad it was.

But read https://aphyr.com/posts/284-call-me-maybe-mongodb and https://aphyr.com/posts/284-call-me-maybe-mongodb for an idea of how the promises in MongoDB documentation compared to the reality of the software under stress. And it wasn't just hypothetical either - there are plenty of horror stories floating around from people who ran into those problems in production for uses cases that were supposed to be a fit for MongoDB.

And the performance argument didn't hold water either. As benchmarks like https://www.enterprisedb.com/node/3441 showed, decent relational databases consistently beat MongoDB on the same hardware. Yes, lots of people rewrote bad relational models and saw performance improve. But apples to apples, writing an application against a relational databases in the same way you would against MongoDB resulted in a win for the relational database.

So yes, there were lots of people saying exactly what you are saying now. But the ones who actually tested their systems and ran performance tests came to a very, very different conclusion.