[jira] [Commented] (JENA-89) Avoid a total sort for ORDER BY + LIMIT queries

["Paolo Castagna (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/JENA-89?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13083012#comment-13083012 ] 

Paolo Castagna commented on JENA-89:
------------------------------------

The optimizer applies the transformation if and only if LIMIT is less than the threshold and no OFFSET is used.
Are we happy with a fixed TOPN_LIMIT_THRESHOLD = 100? 

The optimization is active by default, but it can be turned off via --set http://jena.hpl.hp.com/ARQ#optTopNSorting=false

The changes in BuilderOp are the ones I am less sure about:

-            BuilderLib.checkLength(4, list,  Tags.tagTopN) ;

-            int N = BuilderNode.buildInt(list, 1, -1) ;

-            ItemList conditions = list.get(2).getList() ;

+            BuilderLib.checkLength(3, list,  Tags.tagTopN) ;

+            long N = BuilderNode.buildInt(list.get(1).getList(), 0, -1) ;

+            ItemList conditions = list.get(1).getList().cdr() ;

An example of the optimization in action follows:

SPARQL query:

SELECT  *
WHERE
  { ?s ?p ?o }
ORDER BY ?p ?s
LIMIT   10

SPARQL algebra expression unchanged:

(slice _ 10
  (order (?p ?s)
    (bgp (triple ?s ?p ?o))))

SPARQL algebra expression with the optimization applied:

(top (10 ?p ?s)
  (bgp (triple ?s ?p ?o)))


> Avoid a total sort for ORDER BY + LIMIT queries
> -----------------------------------------------
>
>                 Key: JENA-89
>                 URL: https://issues.apache.org/jira/browse/JENA-89
>             Project: Jena
>          Issue Type: Improvement
>          Components: ARQ
>            Reporter: Paolo Castagna
>            Assignee: Paolo Castagna
>            Priority: Minor
>              Labels: arq, optimization, scalability, sparql
>         Attachments: ARQ_JENA-89_r1156212.patch
>
>
> In case of SPARQL queries with ORDER BY + LIMIT, ARQ sorts the entire result set and then produces the first N, according to the specified LIMIT.
> As an alternative, discussed on jena-dev [1], we can use a PriorityQueue [2] (which is based on a priority heap) to avoid a sort operation.
> ARQ's algebra package contains already a OpTopN [3] operator. The OpExecutor [4] will need to use a new QueryIterTopN instead of QueryIterSort + QueryIterSlice. A new TransformOrderByLimit to be used by Optimize is also necessary.
> ORDER BY + LIMIT queries are typically used to construct the first page when results are paginated. Then the following query is ORDER BY + OFFSET + LIMIT. (Often users stop at the first page). Ideally, we could cache the ORDER BY and implement the OFFSET|LIMIT using results from the cache. However, the improvement described by this issue is limited to the ORDER BY + LIMIT case for which a priority heap is a good enough solution.
> Hopefully, this would improve the scalability of ORDER BY + LIMIT queries in case of small values specified on the LIMIT. 
>   [1] http://markmail.org/thread/5d2gtazkoxsa2ayv
>   [2] http://download.oracle.com/javase/6/docs/api/java/util/PriorityQueue.html
>   [3] https://svn.apache.org/repos/asf/incubator/jena/Jena2/ARQ/trunk/src/com/hp/hpl/jena/sparql/algebra/op/OpTopN.java
>   [4] https://svn.apache.org/repos/asf/incubator/jena/Jena2/ARQ/trunk/src/com/hp/hpl/jena/sparql/engine/main/OpExecutor.java

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira