On Thu, Aug 26, 2021 at 3:15 PM Denis Laxalde <denis.laxalde@dalibo.com> wrote:
>
> Michael Paquier a écrit :
> >> @@ -5862,6 +5862,9 @@ do_start_bgworker(RegisteredBgWorker *rw)
> >> static bool
> >> bgworker_should_start_now(BgWorkerStartTime start_time)
> >> {
> >> + if (IsBinaryUpgrade)
> >> + return false;
> >> +
> > Using -c max_worker_processes=0 would just have the same effect, no?
> > So we could just patch pg_upgrade's server.c to get the same level of
> > protection?
>
> Yes, same effect indeed. This would log "too many background workers"
> messages in pg_upgrade logs, though.
> See attached patch implementing this suggestion.
I disagree. It can appear to have the same effect but it's not
guaranteed. Any module in shared_preload_libraries could stick a
"max_worker_processes +=X" if it thinks it should account for its own
ressources. That may not be something encouraged, but it's definitely
possible (and I think Andres recently mentioned that some extensions
do things like that, although maybe for other GUCs) and could result
in a corruption of a pg_upgrade'd cluster, so I still think that
changing bgworker_should_start_now() is a better option.
