[Mondrian] Thread hangs with RolapStar.clearAggregationRequests

Julian Hyde julianhyde at speakeasy.net
Mon Feb 26 06:33:26 EST 2007


OK, I'll put my changes in. The overhead of creating the HashSet should
be negligible if things are as you describe, and well worth if it avoids
a quadratic worst-case scenario.

Julian

> -----Original Message-----
> From: mondrian-bounces at pentaho.org 
> [mailto:mondrian-bounces at pentaho.org] On Behalf Of Pappyn Bart
> Sent: Monday, February 26, 2007 1:57 AM
> To: Mondrian developer mailing list
> Subject: RE: [Mondrian] Thread hangs with 
> RolapStar.clearAggregationRequests
> 
> Julian,
> 
> Most queries do not generate a lot of aggregation requests.  But the
> query I wrote today, using
> LinRegPoint and Rank, resulted in a lot of aggregation requests (>
> 65000) - before my change.
> 
> After my change, only 3 requests where left in the list.  
> There will be
> no more aggregation requests
> per thread than there will be aggregations, since only the unique
> requests are kept now.
> 
> So the question to be asked is how much aggregations will 
> there be used
> by one query.
> 
> Your proposal will be faster if there are a lot of 
> aggregations used by
> one query and a lot
> of concurrent threads.  Of course, there is some overhead in 
> allocating
> and creating the new HashSet.
> 
> It's up to you if you wish to put these changes in, no problem by me.
> 
> Bart




More information about the Mondrian mailing list