jackrabbit-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ard Schrijvers <a.schrijv...@onehippo.com>
Subject Re: Query that sorts a large result set.
Date Thu, 18 Jun 2009 15:30:52 GMT
> I have been having these thoughts as well, did not implement anything,
> but already expected the 50% drop: I namely think, that you dropped
> the SharedFieldCache memory usage with for example 90%, where the
> SharedFieldCache is a little more then the actual used memory: namely,
> the other 50% is kept by the opened indexReader internal in lucene.
> *If* we are able to actively flush this (TermEnum cache) in our opened
> lucene indexes, I think we are there: namely the SharedFieldCache
> contains the sorting data already, which is kept. Then, we might drop
> memory to way beyond 50%. I looked at this quite some time ago, but
> did not have time to see how we could flush the lucene termenums...

Small addon:

>From lucene api:

/lucene/docs/api/org/apache/lucene/index/TermEnum.html

close()
         Closes the enumeration to further activity, freeing resources.

As the SharedFieldCache does end in the finally with

termEnum.close();

one would expect Lucene to free the in memory terms....I did not test
it, but from the top of my head, memory snapshots I saw in the past do
indicate memory is not freed, and the terms are kept in memory

Regards Ard

>
>

Mime
View raw message