Feature Request: Allow for Indexing about batch-import HOT 4 CLOSED

Ok, so how. Secondary files for indexing which only contain the data to index or what I'd like more a marker in the column header (property) that points out that this property shall be indexed.

from batch-import.

nodes_index.csv
index_name type node_id key value

rels_index.csv
index_name type rel_id key value

node_id and rel_id gathered by line number (the same way relationships does it).

An alternative is we could have an index.csv with index_name and type, and just reference the index ids.

What do you think?

from batch-import.

So I started working on this today, and pretty much have the POC figured out. Now I just need to expand this into user passable args.

Pass in the index type, index name, full text or exact and finally the name of the index file in command line quads:

node_index users fulltext users_names_index.csv
node_index user_keys exact user_keys_index.csv
rel_index worked fulltext worked_index.csv

I'll try to finish it tomorrow and send in a pull request after I test it.

from batch-import.

Thanks Max, what about my initial suggestion of using the table header row for that?

You could use indexname:key as title for those columns that are to be indexed ?

After all the direction of Neo4j indexing is moving towards only indexing values that are actually derivable from the graph data (to allow index build/rebuild).

What would be an interesting (though separate) feature is to use a index-lookup for the start and end-node, e.g. by putting the same indexname:key in the first two column headers and using the acutal value in that row for lookup (this should then employ the faster lookup mechanisms, that e.g. Tero implemented (or lucene-kv store / babudb).

from batch-import.