On Tue, Jan 27, 2026 at 6:06 AM Jakub Wartak <[email protected]>
wrote:

> On Mon, Jan 26, 2026 at 4:08 PM Andres Freund <[email protected]> wrote:
>
...

> > For measuring particularly stuck things, I've been wondering about
> having a
> > regular timer that starts to collect more information if stuck in a
> place for
> > a while. That would probably end up being lower overhead than constantly
> > measuring... But it would also be a lot more work.
>
> Well if something is really stuck, I think the wait events are covering us
> on that,
> aren't they? One can argue if they carry enough information (for me they
> mostly
> do, but I'm trying to squeeze some more stuff into them in a nearby thread
> [1],
> BTW: it's kind of "blocked" due to that 56-bit relfilenode idea/question,
> any thoughts on that?)
>

One scenario where wait events won't help at all is if you have a backend
stuck somewhere that's not calling CHECK_FOR_INTERRUPTS(). Or at least that
was the case as of a few years ago; it wasn't an uncommon thing to see in a
very large fleet. My guess is that such a backend also wouldn't be
responding to internal timers though...

Reply via email to