The goal here is to support something similar to SQL's HAVING clause where one can filter documents based on data that is not available in the index. For example, filter the output of a reduce(....) based on the calculated metrics.
This example would return all where the total spent by each distinct customer is >= 500. The total spent is calculated via the sum(cost) metric in the reduce stream.
The intent is to support as the filters in the having(...) clause the full query syntax of a search(...) clause. I see this being possible in one of two ways.
1. Use Lucene's MemoryIndex and as each tuple is read out of the underlying stream creating an instance of MemoryIndex and apply the query to it. If the result of that is >0 then the tuple should be returned from the HavingStream.
2. Create an in-memory solr index via something like RamDirectory, read all tuples into that in-memory index using the UpdateStream, and then stream out of that all the matching tuples from the query.
There are benefits to each approach but I think the easiest and most direct one is the MemoryIndex approach. With MemoryIndex it isn't necessary to read all incoming tuples before returning a single tuple. With a MemoryIndex there is a need to parse the solr query parameters and create a valid Lucene query but I suspect that can be done using existing QParser implementations.