You are viewing a plain text version of this content. The canonical link for it is here.

Posted to users@jena.apache.org by Zen 98052 <z9...@outlook.com> on 2016/06/03 17:28:48 UTC

optimize a query

Hi,

Is there a way to rewrite this Sparql query to run more efficiently?


SELECT ?s

WHERE

{

  ?s <http://foo/bar> ?o1, ?o2 .

  ?o2 <http://foo/id> "some id" .

  FILTER (?o1 != ?o2) .
}


It seems to me this can be optimized, I am still trying to figure it out.



Thanks,

Z

Re: optimize a query

Posted by Zen 98052 <z9...@outlook.com>.

Thanks Andy for the pointer!

I turned off that flag, and it got pushed to BGPs as expected.

Before that, I also tried rewrite the query, and I got same behavior like turning off that flag.

The re-written query is:


SELECT ?s
WHERE
{
  ?o2 <http://foo/id> "some id" .
  ?s <http://foo/bar> ?o2 .
  ?s <http://foo/bar> ?o1 .
  FILTER (?o1 != ?o2) .
}


Thanks,

Z


________________________________
From: Andy Seaborne <an...@apache.org>
Sent: Friday, June 3, 2016 6:03:02 PM
To: users@jena.apache.org
Subject: Re: optimize a query

On 03/06/16 18:28, Zen 98052 wrote:
> Hi,
>
> Is there a way to rewrite this Sparql query to run more efficiently?
>
>
> SELECT ?s
>
> WHERE
>
> {
>
>    ?s <http://foo/bar> ?o1, ?o2 .
>
>    ?o2 <http://foo/id> "some id" .
>
>    FILTER (?o1 != ?o2) .
> }
>
>
> It seems to me this can be optimized, I am still trying to figure it out.

Yes - by the filter placement transform TransformFilterPlacement

There is no need to wait for   ?o2 <http://foo/id> "some id" .

The filter can be executed after ?s <http://foo/bar> ?o1, ?o2 . as both
?o1 and ?o2 are bound.

BGPs are left alone if

context.isTrueOrUndef(ARQ.optFilterPlacementBGP) ;

See Optimize.rewrite

TDB turns it off by setting that context values on each dataset:

dsg.getContext().set(ARQ.optFilterPlacementBGP, false) ;

(then does it itself in "NodeId" space, not in "Node")

You can experiment using the command line tools:

arq.qparse --print=opt --query=YourQuery.rq

        Andy

>
>
>
> Thanks,
>
> Z
>

Re: optimize a query

Posted by Andy Seaborne <an...@apache.org>.

On 03/06/16 18:28, Zen 98052 wrote:
> Hi,
>
> Is there a way to rewrite this Sparql query to run more efficiently?
>
>
> SELECT ?s
>
> WHERE
>
> {
>
>    ?s <http://foo/bar> ?o1, ?o2 .
>
>    ?o2 <http://foo/id> "some id" .
>
>    FILTER (?o1 != ?o2) .
> }
>
>
> It seems to me this can be optimized, I am still trying to figure it out.

Yes - by the filter placement transform TransformFilterPlacement

There is no need to wait for   ?o2 <http://foo/id> "some id" .

The filter can be executed after ?s <http://foo/bar> ?o1, ?o2 . as both 
?o1 and ?o2 are bound.

BGPs are left alone if

context.isTrueOrUndef(ARQ.optFilterPlacementBGP) ;

See Optimize.rewrite

TDB turns it off by setting that context values on each dataset:

dsg.getContext().set(ARQ.optFilterPlacementBGP, false) ;

(then does it itself in "NodeId" space, not in "Node")

You can experiment using the command line tools:

arq.qparse --print=opt --query=YourQuery.rq

	Andy

>
>
>
> Thanks,
>
> Z
>

Re: optimize a query

Posted by Zen 98052 <z9...@outlook.com>.

I also notice a weirdness this BGP pattern ?o2 <http://foo/id> "some id" is somehow not passed by Jena, hence it never got optimized [:(]

The prime suspect is the FILTER line, which if I remove it, or change it to FILTER (?o1 = ?o2) (note this is just for experimental purpose since it's not making sense), then I can see pattern ?o2 <http://foo/id> "some id" is being passed now. Weird..

________________________________
From: Zen 98052 <z9...@outlook.com>
Sent: Friday, June 3, 2016 1:28:48 PM
To: users@jena.apache.org
Subject: optimize a query

Hi,

Is there a way to rewrite this Sparql query to run more efficiently?

SELECT ?s

WHERE

{

  ?s <http://foo/bar> ?o1, ?o2 .

  ?o2 <http://foo/id> "some id" .

  FILTER (?o1 != ?o2) .
}

It seems to me this can be optimized, I am still trying to figure it out.

Thanks,

Z