Are there any easily foreseeable problems with implementing an r-
tree box
indexing/searching extension to Solr, in the spirit of localsolr? If
anyone
has any pointers I'm all ears.
I have implemented an R-Tree based integration for solr. It is pretty
ugly and memory intensive, but works for now. I plan to release it
(or something like it) in the lucene spatial contrib sometime. I am
waiting to see what the "Flexible indexing" thread leads since that
may be a good way to just build the R-Tree at index time rather then
every time you open a searcher.
In my current approach, I store a string representing a bounding box.
When you open a searcher, it walks through every document and builds
an in memory R-Tree. Then I have a solr query component that applies
a filter based on what matches.
- - - - -
Patrick (of local lucene fame) thinks it is possible to do extent
queries with the cartesian grid method -- essentially you select the
"best fit" level and cell, and that should be set for anything within
the extent. The advantage of this approach is that it is super-fast
and scaleable. The disadvantage is that it is only as accurate as the
grid.
ryan
