Background workers don't react to postmaster death

[hlinnaka]

This seems to be a pg_rag bug: neondatabase/neon#10726

If postmaster dies unexpectedly, the background worker processes should exit quickly. Normally that's handled by using the WL_EXIT_ON_PM_DEATH flag in WaitLatch() calls, and / or checking PostmasterIsAlive() periodically.

[jawj]

Hmm. This is meant to be handled here:

pgrag/exts/rag_bge_small_en_v15/src/lib.rs

Line 169 in 66186ce

while !BackgroundWorker::sigterm_received() {

It's not the first problem we've had with this (and pgrx), since in other situations we've also had the extension appearing to crash on ordinary Postgres shutdowns.

[jawj]

I note that the standard approach to this seems to be BackgroundWorker::wait_latch (such as in this example), but I'm not clear if/how that can be combined with out use of tonic::transport::Server.

@conradludgate As a more committed Rustacean and pgrag contributor: any thoughts here?

[conradludgate]

We could spawn a thread whose only job is wait_latch, which sends a message on an in-memory async channel on completion