[
https://issues.apache.org/jira/browse/JENA-441?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13636529#comment-13636529
]
Rob Vesse commented on JENA-441:
--------------------------------
Andy - Yes that was a typo, I'm well aware that distinct has to happen after
the project, was trying to write this up hurriedly last thing before leaving
work, sorry
I think for our database REDUCED may not be supported so that doesn't help us
much.
Stephen - Yes I remember that discussion
As I said in the original description I will run some actual tests and report
back with some figures, I'll try the various different suggestions and see what
happens
> In some cases it may be useful to apply DISTINCT before applying ORDER BY
> -------------------------------------------------------------------------
>
> Key: JENA-441
> URL: https://issues.apache.org/jira/browse/JENA-441
> Project: Apache Jena
> Issue Type: Improvement
> Components: ARQ, Optimizer
> Affects Versions: Jena 2.10.0
> Reporter: Rob Vesse
> Assignee: Rob Vesse
> Priority: Minor
> Labels: distinct, optimization, order, project
>
> One of our internal users highlighted an interesting query where changing the
> plan makes a big difference in performance.
> The query is essentially the following:
> SELECT DISTINCT ?p
> WHERE
> {
> ?s ?p ?o
> } ORDER BY ?p
> Leaving the fact that it is a fundamentally dumb query to write the user had
> an interesting suggestion about the query plan, currently this generates the
> following:
> (distinct
> (project (?predicate)
> (order (?predicate)
> (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?s ?predicate ?o)))))
> For cases like this it may actually be much more performant to do the
> distinct first, because of the associated semantics of the various operators
> you can't just simply put distinct before the order but if you rewrite the
> query as follows:
> SELECT ?p
> WHERE
> {
> { SELECT DISTINCT ?p WHERE { ?s ?p ?o } }
> } ORDER BY ?p
> You get the likely much more performant plan:
> (project (?predicate)
> (order (?predicate)
> (distinct
> (project (?predicate)
> (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?/s ?predicate
> ?/o))))))
> Clearly this optimization does not apply in the general case, I think it only
> applies in the case where you have a DISTINCT and all ORDER BY conditions are
> simple variables and only those variables are projected in which case I think
> you could produce a plan like the following:
> (order (?predicate)
> (project (?predicate)
> (distinct
> (quadpattern (quad <urn:x-arq:DefaultGraphNode> ?s ?predicate ?o)))))
> I am pretty certain this applies in only a few cases and I haven't reproduced
> this with TDB to see if the performance difference is noticeable yet.
> Andy - I will try and gather some more information and experiment with this
> so don't feel you have to look at this one for the time being.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
