gurudev wrote:
One thing that you can go with is using "adjacent" field collapsing rather
than simple collapsing. As internally SOLR would first sort on the collapse
field to use simple collapsing, which is not the case with "adjacent"
collapsing.
This something that I think could be improved in that patch (non
adjacent collapsing). There is a lot of work done on this as its
essentially a groupby in sql and shares a lot of the issues of an sql
join. The current strategy is to use two sorts - thats certainly a valid
approach, but I think there are much better possibilities. Again, I have
like half a dozen PDFs on possible algorithms or whatnot if anyone is
interested - if only i could magic all these damn pdfs I have into some
code :)