maxdemarzi / boolean_filtering Goto Github PK

Hello Max,

I'm struggling to implement a dynamic filtering use case, a bit similar to the example you use in your blog post https://maxdemarzi.com/2020/08/26/faux-bitmap-indexes-in-neo4j-part-two/

I have a database with millions of publication nodes. Now I want to be able to dynamically filter (AND, OR, NOT) the publications. Some of the filterable fields are missing (empty) for some publications.

I initially tried to implement this with composite indexes on all filterable fields but found out that it's too restrictive: the fields should exist for all nodes, and there is no support for OR's. As a result the index is ignored leading to slow searches (between 2 and 10 seconds).

The solution you propose here looks quite powerful but I still have a couple of questions:

• you mention that, except for range and contains fields, no indexes must be created. So since there is no "upfront optimization" why is your solution so much faster than a regular cypher query? Or in other words why doesn't the cypher implementation use the bitmaps internally?
• can ordering of the results be supported? Since in my use case the filters can be quite loose and the results therefore quite large. So I guess ordering should already be done by an index upfront.

Many thanks