On Thu, Dec 29, 2022 at 12:00 PM Andres Freund <andres@anarazel.de> wrote:
> > I could just move the same tests from heap_prepare_freeze_tuple() to
> > heap_freeze_execute_prepared(), without changing any of the details.
>
> That might work, yes.
Attached patch shows how that could work.
> It seems somewhat wrong that we discard all the work that
> heap_prepare_freeze_tuple() did. Yes, we force freezing to actually happen in
> a bunch of important cases (e.g. creating a new multixact), but even so,
> e.g. GetMultiXactIdMembers() is awfully expensive to do for nought. Nor is
> just creating the freeze plan free.
I'm not sure what you mean by that. I believe that the page-level
freezing changes do not allow FreezeMultiXactId() to call
GetMultiXactIdMembers() any more often than before. Are you concerned
about a regression, or something more general than that?
The only case that we *don't* force xmax freezing in
FreezeMultiXactId() is the FRM_NOOP case. Note in particular that we
will reliably force freezing for any Multi < OldestMxact (not <
MultiXactCutoff).
> I wonder how often it'd be worthwhile to also do opportunistic freezing during
> lazy_vacuum_heap_page(), given that we already will WAL log (and often issue
> an FPI).
Yeah, we don't actually need a cleanup lock for that. It might also
make sense to teach lazy_scan_prune() to anticipate what will happen
later on, in lazy_vacuum_heap_page(), so that it can freeze based on
the same observation about the cost. (It already does a certain amount
of this kind of thing, in fact.)
--
Peter Geoghegan