It's difficult to say without knowing your situation. Personally I would wait a few months at least to be sure that the biggest bugs are killed. Then if it's a new project or something not critical, I may upgrade without any precaution and hope for the best. For something more critical I would rely on the integration tests and manual testing in a pro-production environment for a few weeks. Before the final upgrade, I would finally do a few practice upgrade runs and be ready to rollback if needed.
> Personally I would wait a few months at least to be sure that the biggest bugs are killed.
This is my default behavior with any database product. Solid advice.
This is just a data point but in my experience over the past few years Postgres updates are usually nothing to worry about. There's other systems where updates come with a lot more baggage attached (Like ElasticSearch) where you have to refactor schemas and stay up to date with the documentation way more than with PG.
Homebrew on mac has a long-standing upgrade bug with Postgres extensions like postgis. It uninstalls the old postgis before copying the data over and then gets stuck in a hard to recover state. I always assume I’m going to lose all my data whenever I update major versions.