Had a similar experience with Hasura. They have done some amazing things leveraging Postgres and GraphQL. But there were just too many things that got really questionable. Things like migrations becoming inconsistent with metadata, schema lock in, poor ability to do rate limiting, having to use stored procedures for everything, weird SQL that had performance issues, unexplained row level locking, and so on. Local development was a total mess.

Ultimately we were making architectural decisions to please Hasura, not because it was in the best interests of what or how we were building.