On Sun, May 8, 2022 at 11:36 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Michael Paquier <michael@paquier.xyz> writes:
> > How long can the backend remain unresponsive? I don't think that
> > anybody would object to the addition of some CHECK_FOR_INTERRUPTS() in
> > areas where it would be efficient to make the shutdown quicker, but
> > we need to think carefully about the places where we'd want to add
> > these.
>
> CHECK_FOR_INTERRUPTS is really quite cheap, just a test-and-branch.
> I wouldn't put it in a *very* tight loop, but one test per row
> processed while gathering stats is unlikely to be a problem.
+1. If we're finding things stalling that would be fixed by adding
CHECK_FOR_INTERRUPTS(), we should generally just add it. In the
unlikely event that this causes a performance problem, we can try to
figure out some other solution, but not responding to interrupts isn't
the right way to economize.
--
Robert Haas
EDB: http://www.enterprisedb.com