Re: [geomesa-dev] stats based query optimization on attribute indexes at

[Emilio Lahr-Vivaz <elahrvivaz@xxxxxxxx>]

Jim has it right - it estimates that there would be fewer rows scanned 
in the z2 index than hitting the attribute index. The high-cardinality 
hint causes the cost of an attribute query to be the estimated count 
divided by 10:

https://github.com/locationtech/geomesa/blob/master/geomesa-accumulo/geomesa-accumulo-datastore/src/main/scala/org/locationtech/geomesa/accumulo/index/attribute/AttributeQueryableIndex.scala#L264

If you know that you want to hit the attribute index, you can use the 
geoserver 'view params' to to skip the cost-based query planning. In 
1.3.0 this would be "QUERY_INDEX" (to correspond to the query hint 
name), in previous versions it would be "STRATEGY":

https://github.com/locationtech/geomesa/blob/master/geomesa-index-api/src/main/scala/org/locationtech/geomesa/index/geoserver/ViewParams.scala#L54

http://docs.geoserver.org/stable/en/user/data/database/sqlview.html#using-a-parametric-sql-view

so something like: &viewparams=QUERY_INDEX:attr

We could definitely add something like 'highest' if people want that.


On 12/27/2016 11:23 AM, Jim Hughes wrote:
> Hi Anthony,
>
> I'd say that the CBO should be relatively 'free' to pick between the 
> Z2 index and the attribute index.
>
> If one zoomed into a few blocks of NYC and was searching for 'pizza', 
> I'd imagine that pulling back POI from a geo-index to be a faster than 
> checking to see which pizza shops were in that part of Queens.  On the 
> other hand, if you were using Twitter ID, maybe it makes sense to use 
> the attribute index.
>
> Perhaps, we need a 'cardinality=highest' hint which indicates to 
> always prefer the attribute index and never choose...
>
> Cheers,
>
> Jim
>
> On 12/26/2016 3:28 PM, Anthony Fox wrote:
> > Hi all,
> >
> > I've got a quick question about the query planner.  Say I have an
> > attribute index on an attribute with a high cardinality. Something like
> > the following sft:
> >
> > group:String:index=full:cardinality=high,dtg:Date,*geom:Point:srid=4326
> >
> > When running queries with stats based optimization turned on, I've
> > noticed that if I am at a high zoom level, the query planner tries to
> > hit the z2 table.  In my case, I'd expect it to only ever hit the
> > attribute index.
> >
> > I think I can use the override to not use stats when query planning but
> > I wonder if there's a better way.
> >
> > Thoughts?
> >
> > Thanks,
> > Anthony
> > _______________________________________________
> > geomesa-dev mailing list
> > geomesa-dev@xxxxxxxxxxxxxxxx
> > To change your delivery options, retrieve your password, or 
> > unsubscribe from this list, visit
> > http://locationtech.org/mailman/listinfo/geomesa-dev
>
>
> _______________________________________________
> geomesa-dev mailing list
> geomesa-dev@xxxxxxxxxxxxxxxx
> To change your delivery options, retrieve your password, or 
> unsubscribe from this list, visit
> http://locationtech.org/mailman/listinfo/geomesa-dev