maria-developers team mailing list archive

[Sergey Petrunya <psergey@xxxxxxxxxxxx>]

Hello Philip,

One more note about correlated results: I agree that it's true that the
optimizer can't make a good choice if the data is correlated. 

But since this is intersection, we could take this from other end: when 
doing intersection scan, calculate utility of using extra indexes on the fly.

If we scan some amount of rows and see that using extra indexes doesn't 
provide any benefit, we could fall back to "unary intersection" which would 
be very similar to what range+DS-MRR do. 

The question is, is the issue of slowdowns because of correlations is such a
big problem that we need to drop other things and start working on a solution
like the above...

BR
 Sergey
-- 
Sergey Petrunia, Software Developer
Monty Program AB, http://askmonty.org
Blog: http://s.petrunia.net/blog