Re: Have vacuum emit a warning when it runs out of maintenance_work_mem

Heikki Linnakangas <heikki@enterprisedb.com> writes:
> Jim C. Nasby wrote:
>> If nothing else, it would likely be
>> worth special-casing an entire page being dead, which is a common case
>> for queue tables. That could be done by making an entry in the page
>> number array with a special offset value.

> That won't work, because someone might add live tuples to the page after
> the 1st vacuum pass. You could only invoke that special case when
> there's no room on the page for new tuples, but that's a hack and not as
> common.

The bitmap case seems to me to be plenty efficient already for an
all-dead page.  The regime where my proposal seems to leave something
to be desired is just a few dead tuples per page --- it's not very much
better than the existing code in that case.

            regards, tom lane