Re: Performance improvement for joins where outer side is unique

On 13 March 2015 at 20:34, Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp> wrote:

Unfortunately I can't decide this to be 'ready for commiter' for

now. I think we should have this on smaller footprint, in a
method without separate exhauxtive searching. I also had very
similar problem in the past but I haven't find such a way for my
problem..

I don't think it's ready yet either. I've just been going over a few things and looking at Tom's recent commit b557226 in pathnode.c I've got a feeling that this patch would need to re-factor some code that's been modified around the usage of relation_has_unique_index_for() as when this code is called, the semi joins have already been analysed to see if they're unique, so it could be just a case of ripping all of that out of create_unique_path() and just putting a check to say rel->is_unique_join in there. But if I do that then I'm not quite sure if SpecialJoinInfo->semi_rhs_exprs and SpecialJoinInfo->semi_operators would still be needed at all. These were only added in b557226. Changing this would help reduce the extra planning time when the query contains semi-joins. To be quite honest, this type of analysis belongs in analyzejoin.c anyway. I'm tempted to hack at this part some more, but I'd rather Tom had a quick glance at what I'm trying to do here first.