Re: Page at a time index scan

Heikki Linnakangas <hlinnaka@iki.fi> writes:
> The first solution that occurs to me is to force page splits to choose the
> target page so that it's blkno > the original page's blkno during vacuum.

I thought about that too, but don't like it for three reasons:

* it encourages index bloat, the more the longer the vacuum runs.  Not
good, especially if you've got aggressive vacuum cost delay settings.

* there's a locking problem with respect to how you turn that behavior on.

* there's a failure mode where the behavior doesn't get turned off if
vacuum fails partway through.

            regards, tom lane