The field cache is completely reloaded. LUCENE-831 solves this by merging the field caches of the segments. For realtime search systems, merging the field caches is not desirable though.
On Thu, Dec 4, 2008 at 6:45 PM, John Wang <[EMAIL PROTECTED]> wrote: > Glad to be of help. > Understand that FieldCache lives in a map in the static memory and is keyed > by an IndexReader. So if your reader updates often there might be an issue > of cleaning the map. > > This is a question for the Luceners, when you call IndexReader.reopen, how > is FieldCache updated? > > -John > > On Thu, Dec 4, 2008 at 5:46 PM, Muralidharan V <[EMAIL PROTECTED] > >wrote: > > > John, > > > > Using the FieldCache worked well. Thanks! > > > > -Murali > > > > On Thu, Dec 4, 2008 at 3:10 PM, John Wang <[EMAIL PROTECTED]> wrote: > > > > > Easiest way to do this is using the FieldCache. It constructs a > > StringIndex > > > object which gives you very fast lookup to the field value (index) > given > > a > > > docid. Create a parallel count array to the lookup array for the > > > StringIndex. Run your HitCollector thru should be fast. > > > Loading FieldCache maybe expensive, but you do it only once when > opening > > > the > > > reader. > > > > > > This is one way of implementing Faceted Search with Lucene but there > are > > > lotsa things you can do to make this faster. > > > > > > This is of course assuming your Brand field is indexed and not > tokenized. > > > > > > One side-benefit is that if you do this, sorting on Brand would be fast > > > because FieldCache would be already loaded. > > > > > > Hope this helps > > > > > > -John > > > > > > > > > On Thu, Dec 4, 2008 at 3:02 PM, Muralidharan V <[EMAIL PROTECTED] > > > >wrote: > > > > > > > We are evaluating lucene for a product search engine. One requirement > > is > > > > that we be able to suggest the top n brands(the ones with most > products > > > in > > > > the result set) for a given search term to further refine the search > > > query. > > > > The brand is stored in a separate field and searches are performed > > > against > > > > product description and brand. > > > > > > > > One option is to use a custom HitCollector to keep track of the > brands > > in > > > > the result set but that would require reading the brand field for > each > > > doc > > > > that matches the search term. We think this will be an order of > > magnitude > > > > slower. Is there anything else that we can do? > > > > > > > > Thanks, > > > > Murali > > > > > > > > > >
