Hi again, I'm working on pg_duckdb these days. In that project I'm actually calling into Postgres FFI from different threads. This is safe because I manage a global lock there. I'm running into the problem though, that when calling SPI, rust extensions on the same system will abort the whole Postgres process tree due to this check.

So I'm wondering if there's a way in which pg_duckdb could either tell pgrx extensions what the "current" thread id is. Or a way to temporarily disable that check. Because in this environment the FFI calling behavior is actually safe, but pgrx extensions are a so conservative that it still doesn't work.