Using JanusGraph 0.2.0 with Java, I have a simple traversal like this to get the neighborhood of a vertex:

Iterator<Vertex> neighbors=graph.traversal().V(vertexId).bothE(edgeLabels).otherV().dedup();

vertexId is the ID of the vertex of interest and edgeLabels are edge labels of interest (not all edge labels contribute to this view of the vertex's neighborhood).

When run as-is above, this works perfectly. However, in the interest of performance, I wanted to try using a GraphComputer, because the real graph has some vertices with on the order of one million neighbors.

If I simply add withComputer() after traversal(), like this, the query throws a series of exceptions.

Iterator<Vertex> neighbors=graph.traversal().withComputer().V(vertexId).bothE(edgeLabels).otherV().dedup();

The exceptions are similar to this:

2018-02-09 12:46:22.541 ERROR [o.j.d.k.s.StandardScannerExecutor       ] - Exception processing row [ 24-  0-  0-  0-  0-  0-  0-128]:  java.lang.IllegalArgumentException: Undefined type used in query: gremlin.traversalVertexProgram.haltedTraversers
    at org.janusgraph.graphdb.query.QueryUtil.getType(QueryUtil.java:63)

 
I am still somewhat new to Gremlin, but have extensive experience with Blueprints. Is my assumption that I can "automagically" benefit from Gremlin OLAP simply by adding withComputer() to an existing traversal (subject to the caveats in the documentation) incorrect? Am I doing something else?

Thanks in advance!

-- John J. Szucs

Hi John,

In short,

• use withComputer(SparkGraphComputer.class)
• take conf/hadoop-graph/read-hbase.properties or conf/hadoop-graph/read-cassandra3.properties as an example

The withComputer() without SparkGraphComputer is merely to force a breadth-first search, but does not introduce any parallellism (in this syntax). Therefore, I guess it is neither tested for JanusGraph nor needed.

If you do not use HBase or Cassandra as a storage backend you will have to export your data and use the ScriptInputFormat of TinkerPop's HadoopGraph.

Also OLAP might have problems with your supernodes if it would cause an uneven distribution of work over the Spark executors.

HTH,     Marc

Op vrijdag 9 februari 2018 18:52:04 UTC+1 schreef John J. Szucs:

Using JanusGraph 0.2.0 with Java, I have a simple traversal like this to get the neighborhood of a vertex:

Iterator<Vertex> neighbors=graph.traversal().V(vertexId).bothE(edgeLabels).otherV().dedup();

vertexId is the ID of the vertex of interest and edgeLabels are edge labels of interest (not all edge labels contribute to this view of the vertex's neighborhood).

When run as-is above, this works perfectly. However, in the interest of performance, I wanted to try using a GraphComputer, because the real graph has some vertices with on the order of one million neighbors.

If I simply add withComputer() after traversal(), like this, the query throws a series of exceptions.

Iterator<Vertex> neighbors=graph.traversal().withComputer().V(vertexId).bothE(edgeLabels).otherV().dedup();

The exceptions are similar to this:

2018-02-09 12:46:22.541 ERROR [o.j.d.k.s.StandardScannerExecutor       ] - Exception processing row [ 24-  0-  0-  0-  0-  0-  0-128]:  java.lang.IllegalArgumentException: Undefined type used in query: gremlin.traversalVertexProgram.haltedTraversers
    at org.janusgraph.graphdb.query.QueryUtil.getType(QueryUtil.java:63)

 
I am still somewhat new to Gremlin, but have extensive experience with Blueprints. Is my assumption that I can "automagically" benefit from Gremlin OLAP simply by adding withComputer() to an existing traversal (subject to the caveats in the documentation) incorrect? Am I doing something else?

Thanks in advance!

-- John J. Szucs

Hi John,

I have encountered the same trouble. If there any solution to solve the ERROR [o.j.d.k.s.StandardScannerExecutor]? And why it happened? 

Thanks for any suggestions

On Saturday, February 10, 2018 at 1:52:04 AM UTC+8 jo...@... wrote:

Using JanusGraph 0.2.0 with Java, I have a simple traversal like this to get the neighborhood of a vertex:

Iterator<Vertex> neighbors=graph.traversal().V(vertexId).bothE(edgeLabels).otherV().dedup();

vertexId is the ID of the vertex of interest and edgeLabels are edge labels of interest (not all edge labels contribute to this view of the vertex's neighborhood).

When run as-is above, this works perfectly. However, in the interest of performance, I wanted to try using a GraphComputer, because the real graph has some vertices with on the order of one million neighbors.

If I simply add withComputer() after traversal(), like this, the query throws a series of exceptions.

Iterator<Vertex> neighbors=graph.traversal().withComputer().V(vertexId).bothE(edgeLabels).otherV().dedup();

The exceptions are similar to this:

2018-02-09 12:46:22.541 ERROR [o.j.d.k.s.StandardScannerExecutor       ] - Exception processing row [ 24-  0-  0-  0-  0-  0-  0-128]:  java.lang.IllegalArgumentException: Undefined type used in query: gremlin.traversalVertexProgram.haltedTraversers
    at org.janusgraph.graphdb.query.QueryUtil.getType(QueryUtil.java:63)

 
I am still somewhat new to Gremlin, but have extensive experience with Blueprints. Is my assumption that I can "automagically" benefit from Gremlin OLAP simply by adding withComputer() to an existing traversal (subject to the caveats in the documentation) incorrect? Am I doing something else?

Thanks in advance!

-- John J. Szucs