Ummm, 400k documents is _tiny_ by Solr/Lucene standards. I've seen 150M docs fit in 16G on Solr. I put 11M docs on my laptop....
So I would _strongly_ advise that you don't worry about space at all as a first approach and freely copy as many fields as you need to support your use-case. Only after you've proved that this is untenable would I recommend you develop custom code. You'll be in production much faster that way ;) Of course this is irrelevant if each doc is "War and Peace", but.... Best, Erick On Thu, Jul 31, 2014 at 3:29 PM, Juan Pablo Albuja <jpalb...@dustland.com> wrote: > Good afternoon guys, I really appreciate if someone on the community can > help me with the following issue: > > I need to implement a Solr autosuggest that supports: > > 1. Get autosuggestion over multivalued fields > > 2. Case - Insensitiveness > > 3. Look for content in the middle for example I have the value > "Hello World" indexed, and I need to get that value when the user types > "wor" > > 4. Filter by an additional field. > > I was using the terms component because with it I can satisfy 1 to 3, but > for point 4 is not possible. I also was looking at faceting searches and > Ngram.Edge-Ngrams, but the problem with those approaches is that I need to > copy fields over to make them tokenized or apply grams to those, and I > don't want to do that because I have more than 6 fields that needs > autosuggest, my index is big I have more than 400k documents and I don't > want to increase the size. > I was trying to Extend the terms component in order to add an additional > filter but it uses TermsEnum that is a vector over an specific field and I > couldn't figure out how to filter it in a really efficient way. > Do you guys have an idea on how can I satisfy my requirements in an > efficient way? If there is another way without using the terms component > for me is also awesome. > > Thanks > > > > > Juan Pablo Albuja > Senior Developer > > >
