MySQL-Solr for geospatial search

Question

The site currently does mainly range searches (latitude & longitude) with some filtering like WHERE color = "red" type of clauses. However using MySQL with geospatial index is still quite slow and I need to speed it up.

Problem: Will using Solr to do the search be a good idea?

If so, should I only duplicate the range columns from MySQL into Solr, and do the WHERE clauses in MySQL, or do both type of queries in Solr?

I've read that Solr is not for storing data like a database (ie. MySQL). Does this mean that if my search can take place over 10 different columns (or field in Solr terms), and the MySQL table that I replicated Solr's from only has 11 tables, I would still keep the MySQL table even though that will use up almost twice as much storage space half of which is redundant?

It appears that I'm using structured data (because each row has many columns defined?) and storing the entire table in Solr instead having redundant data on MySQL and Solr will save storage space and number of database access operations when writing. Is Solr a good choice here?

In terms of speed, would it be better to use PostGIS or Solr?

Answer 1

Solr has very fast numerical/date range queries. Solr 3 geospatial takes advantage of that, and I wrote a plugin that does even better. I doubt MySQL is faster.

That said, if the sole problem you are trying to solve is slow geospatial queries then bringing in Solr may solve it but will add a lot of overall complexity to your system since it isn't designed to replace relational databases--it works alongside them. Don't get me wrong; Solr is awesome, particularly for faceted navigation and text search. But you didn't state you wanted to take advantage of Solr's primary features.

PostGIS is by far the most mature open-source GIS storage system. I suggest you try it as an experiment to see if it's better. I would try a lat + lon pair of columns approach like what you are doing now with MySQL, and I would also try using the PostGIS native geospatial way to do it, whatever that is exactly.

One thing you could try in either MySQL or PostGIS is to round your latitude and longitude value to the number of decimals to get an appropriate level of precision you need, which is surely far less than the full precision of a double. And if you store them in floats rather than doubles, right there the precision is capped to 2.37 meters. The system you use will probably have a much easier time doing range queries if there are fewer distinct values to scan over.