[
https://issues.apache.org/jira/browse/JENA-383?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13633966#comment-13633966
]
Andy Seaborne commented on JENA-383:
------------------------------------
Jena ARQ 2.9.4 could mis-optimize. The fix is safe but can miss a case when
the inner optional (both sides of the leftjoin) don't involve the assigned
variable.
> The query optimizer generates a suboptimal query plan in case of nested
> optionals followed by a filter
> ------------------------------------------------------------------------------------------------------
>
> Key: JENA-383
> URL: https://issues.apache.org/jira/browse/JENA-383
> Project: Apache Jena
> Issue Type: Improvement
> Components: ARQ, Optimizer
> Affects Versions: ARQ 2.9.4
> Reporter: Giovanni Mels
> Labels: filter, nested, optimization, optional
> Attachments: simple-query.rq
>
>
> The query optimizer generates a suboptimal query plan in case of nested
> optionals followed by a filter (?var = <uri>).
> The problem is similar to the problem discussed here
> http://markmail.org/message/zy5qx47hxfo2vdlg, except we have nested optionals
> in our query (There are no scope problems, all optionals are translated to
> conditionals). It is only the filter that is translated differently:
> arq-2.8.8
> (extend ((?var <uri>))
> (conditional
> (conditional
> (conditional
> (conditional
> (conditional
> (conditional
> (bgp
> ...
>
> jena-arq-2.9.4
> (filter (= ?var <uri>)
> (conditional
> (conditional
> (conditional
> (conditional
> (conditional
> (conditional
> (bgp
> ...
> Query times went from milliseconds to over 7 minutes because of this.
> I reduced the query to this test case:
> PREFIX ex: <http://example.org/test#>
> SELECT * WHERE {
> ?var ex:p1 ?x.
>
> OPTIONAL {
> ?x ex:p2 ?y.
>
> OPTIONAL {
> ?y ex:p3 ?z
> }
> }
>
> FILTER (?var = ex:i)
> }
> (filter (= ?var <http://example.org/test#i>)
> (conditional
> (bgp (triple ?var <http://example.org/test#p1> ?x))
> (conditional
> (bgp (triple ?x <http://example.org/test#p2> ?y))
> (bgp (triple ?y <http://example.org/test#p3> ?z)))))
>
>
> The nested OPTIONAL seems to be the problem here, removing it gives
> (assign ((?var <http://example.org/test#i>))
> (conditional
> (bgp (triple <http://example.org/test#i> <http://example.org/test#p1>
> ?x))
> (bgp
> (triple ?x <http://example.org/test#p2> ?y)
> (triple ?y <http://example.org/test#p3> ?z)
> )))
>
> This might be related to JENA-294.
--
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
