Btw, I always wanted to ask why it's always done in two stages. It seems to me that it's purposed for the specific usecase. But shouldn't we have an option to do it in the single stage? eg. if I have a few fields to retrieve and not really deep paging, I'd like to retrieve whole fl-s and merge them by PKs.
Benson, Thanks for contributing the doc for community. I spend a lot of time to get into that machinery on my own. Regards On Thu, Apr 12, 2012 at 4:17 AM, Yonik Seeley <[email protected]>wrote: > On Wed, Apr 11, 2012 at 8:15 PM, Benson Margulies <[email protected]> > wrote: > > I'd like to add some more information to this; to that end, I'd > > appreciate a correctness review of the following: > > > > * QueryComponent distribution algorithm > > > > The query component distributes queries in two passes. First, it sends > > the query out to the shards with the field list trimmed to be just the > > score and the doc ID. Upon receipt of these results, it merges the > > lists, and then issues new queries out to the shards with > > ids=id1,id2,... to retrieve the remaining fields. The results are > > packaged up and flow out through the result writer. > > Yep. > > -Yonik > lucenerevolution.com - Lucene/Solr Open Source Search Conference. > Boston May 7-10 > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > > -- Sincerely yours Mikhail Khludnev [email protected] <http://www.griddynamics.com> <[email protected]>
