Labels look very cool, but at first glance they seem too “linear” for typing.

If a `movie -IS_A-> Movie` its an excellent boost for performance and constraining.

But how can we capture the relations between fine-grained `type` hierarchies ?

For example `movie -IS_A->ActionMovie’ and `ActionMovie-KIND_OF->Movie` ?

1 -> 5
3 -> 2
6 -> 1

the rel-file should be:

1 -> 5
6 -> 1
3 -> 2

It also helps for the normal inserter (speeds it up b/c adding rels doesn’t have to walk across the whole relationship-store-file.

Could you elaborate on how exactly do they need to be sorted (and what happens if they are not sorted as required)?

I am asking because rels.csv example in https://github.com/jexp/batch-import#readme is not sorted by min(start, end) — the last line is out of place.

if you have streaming in the client you can easily consume a result from a non-streaming server so that the client has less memory consumption (at least if you don’t keep the results around but just use them to calculate something).

In Neo4j Server 1.8 we implemented streaming for the format that was there before (aka colums:[], data: [[row1],[row2]] ). So you should use that format.

Our problem now is that we sometimes get a result from the Neo4j server that is very large and having in saved in one Java object is a problem.

The system is in production so the hope is to implement a quick client fix using the non-streamed Neo4j 1.7.1 server and later upgrade to 1.8 and make use of server-side streaming.

I see that the sample output for the streamed server has “rows” whereas our non-streamed server has “data”. I also see that the sample streamed server output has type:value pairs whereas our non-streamed server results has only values.

My question is:

Can a Jackson Cypher streaming client work correctly with a non-streamed Neo4j server?

Thank you for your time,
Karen