Heikki Linnakangas <hlinnaka@iki.fi> writes:
> On Wed, 29 Mar 2006, Tom Lane wrote:
>> That loses the ability to reflect tuple deadness back into LP_DELETE
>> flags, no?
> At first glance, it doesn't look so hard. index_getmulti could mark
> those tids that are dead, and btgetmulti would rescan the index page and
> set LP_DELETE on all tuples that are still there.
> We don't have to care about splits; if the index tuple is no longer where
> it used to be, just ignore it. Right, no?
True --- as long as there's even a reasonable probability of the tuple
getting marked, we'll get the performance benefit. I don't see a way to
make it work for bitmap indexscans though --- by the time we visit the
heap, the index has long since forgotten where those index entries were.
I think this may be worth doing even disregarding any possible vacuum
speedup, simply because it'll reduce the number of index page lock/unlock
cycles needed during a regular indexscan.
regards, tom lane