Attached are two patches. The first patch adds assertions that verify
that calls to visibilitymap_set() don't set VISIBILITYMAP_ALL_FROZEN
for a page without also setting VISIBILITYMAP_ALL_VISIBLE (plus a
similar assertion for visibilitymap_clear()). When I run "make
check-world" with just the first patch, several tests fail because one
of the new assertions fails.
The second patch tentatively addresses the issue by making two of the
calls to visibilitymap_set() made by vacuumlazy.c set
VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN -- as opposed to
just setting VISIBILITYMAP_ALL_FROZEN, which is what happens on HEAD.
It's not clear whether or not it's strictly correct to only set
VISIBILITYMAP_ALL_FROZEN, but it seems questionable. It's also likely
to have negative performance implications. Non-aggressive VACUUMs
won't actually use VISIBILITYMAP_ALL_FROZEN to skip -- they could in
principle, but they don't.
VACUUM's skipping logic doesn't just refuse to skip
VISIBILITYMAP_ALL_VISIBLE pages during an aggressive VACUUM (which is
necessary and makes sense); it also refuses to skip
VISIBILITYMAP_ALL_FROZEN pages during non-aggressive VACUUMs. And so
any page that just has its VISIBILITYMAP_ALL_FROZEN bit set is not
skippable by non-aggressive VACUUMs. I'm referring to this code:
while (next_unskippable_block < rel_pages)
{
uint8 vmskipflags;
vmskipflags = visibilitymap_get_status(vacrel->rel,
next_unskippable_block,
&vmbuffer);
if (vacrel->aggressive)
{
if ((vmskipflags & VISIBILITYMAP_ALL_FROZEN) == 0)
break;
}
else
{
if ((vmskipflags & VISIBILITYMAP_ALL_VISIBLE) == 0)
break;
}
vacuum_delay_point();
next_unskippable_block++;
}
--
Peter Geoghegan