Q on facets performance with CachingWrapperFilter in Lucene 4.4

[Rose, Stuart J]

I've noticed some interesting and unexpected behavior regarding performance of 
the Facets aggregation in Lucene 4.4 and am wondering if anyone has come across 
this before and can offer insight to potential factors.

In a nutshell, the CachingWrapperFilter results in significant performance 
gains when using it to wrap a base query while iterating over about one 
thousand 'sub-queries' . However it also results in lower performance during 
facets aggregation following each sub-query. In contrast, if a 
QueryWrapperFilter is used instead of a CachingWrapperFilter, then the 
sub-queries take 4X as long yet facets aggregation takes much less time.

Below is a summary table containing average times to perform a search 
(sub-query):

                searcher.search(subQuery, baseFilter, facetsCollector);

and to collect the facet results for that search:

                List<FacetResult> results = facetsCollector.getFacetResults();


... average times are in milliseconds...

                                                                CWF             
          QWF
                avg. Search                         39                          
 166
                avg. Accumulate               103                         29


I'm curious to know if the CachingWrapperFilter can be used in a way that does 
not negatively impact the facets collector, if there are known solutions, or if 
this has been resolved in a subsequent Lucene release.


Thanks!
Stuart