Distributed reindexing for any storage backend

Nobody has claimed this yet.

Maybe the title is not the best but I have noticed one disturbing thing: MapReduceIndexManagement class has direct dependencies on the concrete classes for HBase AND Cassandra. I see two issues with this: first, the dependencies themselves, the second - the lack of support for any other storage backends. For example, AWS DynamoDB.

It seems that, at least, janusgraph-hadoop should be backend-independent and something like janusgraph-hadoop-cassandra and janusgraph-hadoop-hbase should be created to deal with backend-specific implementations. I suspect this may require some additional layer to implement the backend-independent partititione, scanner and input formats.

Start by reading the MapReduceIndexManagement class and the janusgraph-hadoop module to map its direct HBase and Cassandra dependencies. Determine the boundaries needed for backend-independent partitioning, scanning, and input formats, then define backend-specific implementations such as janusgraph-hadoop-cassandra and janusgraph-hadoop-hbase; done means another storage backend can be supported without coupling the core module to concrete backends.

Written by the indexing model from the issue text.