Hi,
I don't have the answer about why cache still shows "true", but as far as 
memory usage goes, based on your description I'd guess the memory is allocated 
and used by the JVM which typically  tries not to run GC unless it needs to.  
So if you want to get rid of that used memory, you need to talk to the JVM and 
persuade it to run GC.  I don't think there is a way to manage memory usage 
directly.  There is System.gc() that you can call, but that's only a 
"suggestion" for the JVM to run GC.


Otis --
Sematext -- http://sematext.com/ -- Lucene - Solr - Nutch


----- Original Message ----
> From: Roberto Nieto <[EMAIL PROTECTED]>
> To: solr-user <solr-user@lucene.apache.org>
> Sent: Wednesday, June 18, 2008 7:43:12 AM
> Subject: never desallocate RAM...during search
> 
> Hi users,
> 
> Somedays ago I made a question about RAM use during searchs but I didn't
> solve my problem with the ideas that some expert users told me. After making
> somes test I can make a more specific question hoping someone can help me.
> 
> My problem is that i need highlighting and i have quite big docs (txt of
> 40MB). The conclusion of my tests is that if I set "rows" to 10, the content
> of the first 10 results are cached. This if something normal because its
> probable needed for the highlighting, but this memory is never desallocate
> although I set solr's caches to 0. With this, the memory grows up until is
> close to the heap, then the gc start to desallocate memory..but at that
> point the searches are quite slow. Is this a normal behavior? Can I
> configure some solr parameter to force the desallocation of results after
> each search? [I´m using solr 1.2]
> 
> Another thing that I found is that although I comment (in solrconfig) all
> this options:
> ----> filterCache, queryResultCache, documentCache, enableLazyFieldLoading,
> useFilterForSortedQuery, boolTofilterOptimizer
> In the stats always appear "caching:true".
> 
> I'm probably leaving some stupid thing but I can't find it.
> 
> If anyone can help me..i'm quite desperate.
> 
> 
> Rober.

Reply via email to