On 4/5/06, Chris Hostetter <[EMAIL PROTECTED]> wrote: > : IMO, it's probably a bad idea unless your index is simply too large to > : service queries in a reasonable amount of time. Things become much > : harder when distributed (for example, how would you take the > : intersection of two sets of documents in a distributed manner)? Any > : type of distributed approach would be limited to a subset of > : functionality. > > I think there would definitely be benefit in supporting the same level of > functionality as the existing MultiSearcher class does across multiple > instances of Solr -- complex DocSet operations may not be feasible > for plugins across multiple Solr instances, but I would imagine that > having a "SolrSubSearcherRequestHandler" plugin that is capable of > returning all of the data eccessary for a client to merge the responses > from multiple servers would be possible. > > (i'm notcertian though ... i've never really used MultiSearcher so i'm not > sure what all is involved in the merging of results)
Lucene's MultiSearcher can already get you the top "n" documents, filtered, and sorted by score or other criteria. We could create a SolrMultiSearcher that builds on that. One gets into issues of how to do high availability and redundancy though (which MultiSearcher doesn't currently handle AFAIK) -Yonik
