[MarkLogic Dev General] query execution plan question

Mon Jul 16 11:15:34 PDT 2007

Hi Mike,

You don't mention it, but <body> is a fragment root, right?

-jh-

Mike Sokolov wrote:
> I stumbled on a surprising behavior and wanted to know if anybody has a 
> good heuristic for identifying this odd little performance bottleneck in 
> MarkLogic so as to avoid it (there is an easy alternative expression 
> that works better).
> 
> If I run the query:
> 
> cts:search(//doc, cts:element-attribute-value-query (xs:QName('doc'),  
> xs:QName('e:id'),  '/opr/t125/e1646'))/body,
> I get results quickly (.002 sec)
> 
> However the very similar query
> 
> //doc[@id='/opr/t125/e1646']/body
> 
> takes 1.37 sec, almost 1000x slower!
> 
> query-meters reveals that it involves 85,000 tree-cache-hits and touches 
> various documents, when in fact there is only a single fragment in the 
> system matching the given attribute value.  Similarly query-trace says 
> "Selected 7212 fragments to filter," although I'm not sure why.
> 
> Somehow adding the /body element seems to be the culprit since:
> 
> //doc[@id='/opr/t125/e1646']
> 
> is quite fast, as is
> 
> let $doc := //doc[@id='/opr/t125/e1646']
>  return $doc/body
> 
> 
> Does this seem surprising to anyone else?  I have been trying to use 
> xpath whenever I can due to its elegant simplicity in comparison with 
> the cts: functional operators, and I thought I had a good internal sense 
> of when an expression could get run quickly using indexes, but now I am 
> beginning to worry about other possible performance bottlenecks I may 
> have overlooked.
> _______________________________________________
> General mailing list
> General at developer.marklogic.com
> http://xqzone.com/mailman/listinfo/general