Re: Pluggable Indexes (was Re: rmgr hooks (v2))

On Thu, 2009-01-22 at 10:09 +0200, Heikki Linnakangas wrote:
> Simon Riggs wrote:
> > Preventing work on new indexes by non-committers has meant that Bitmap
> > indexes, which first came out in 2005 have not been usable with
> > Postgres. That forced people *away* from Postgres towards Bizgres. Lack
> > of Bitmap indexes is a huge issue for many people. It's 2009 now and it
> > seems probable that without this patch it will be 2010 at least before
> > they see BMIs, and later still before they see other index types.
>
> No-one is preventing anyone from working on bitmap indexes.

> Bitmap indexes required other backend changes, in addition to the rmgr
> changes. This rmgr plugin patch is *not* sufficient to enable bitmap
> indexes to live as a plugin.
>
> This patch does *not* bring us any closer to having bitmap indexes.
> Don't raise false hopes.

I agree those changes would be better but those changes are *not*
essential (as has been agreed onlist). They are just a possible tuning
feature, amongst many that must prove themselves before they happen.

Manipulating multiple large bitmaps on a 1 TB table will still be much
more efficient than reading multiple btrees and manipulating those. BMIs
are typically much smaller than btrees, so even if they use some memory
we will avoid significant amounts of real I/O. BMIs also have a
significantly lower time to build, making them much more practical.

It is not a false hope since the case is not black/white, just a matter
of opinion.

--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support