Postgres as a search engine: how far can you push it?

Before reaching for a dedicated search cluster, I gave Postgres a real workload: full-text ranking, trigram fuzzy matching, and vector similarity, all on the same rows. The goal was to find the wall, not to avoid it.

It went further than I expected. The wall is real, but it is much further out than the reflex to add infrastructure assumes — and everything before it is one fewer system to operate.