TinkerPop API support in DseGraphFrame

DseGraphFrame supports a subset of the Apache TinkerPop traversal API.

DseGraphFrame supports a subset of the Apache TinkerPop traversal API.

DseGraphFrame does not support org.apache.tinkerpop.gremlin.process.traversal.Traverser or org.apache.tinkerpop.gremlin.process.traversal.TraversalSideEffects.

Supported methods

DseGraphFrame mimics the TinkerPop graph traversal source by defining two methods: E() and V(). These methods return a GraphTraversal that has all methods defined below. Only a limit set of TinkerPop's Step classes are supported. Steps other than the ones in the following table will throw an UnsupportedException.

Table 1. TinkerPop read methods
Steps Methods
CountGlobalStep count()
GroupCountStep groupCount()
IdStep id()
PropertyValuesStep values()
PropertyMapStep propertyMap()
HasStep has(), hasLabel()
IsStep is()
VertexStep to(), out(), in(), both(), toE(), outE(), inE(), bothE()
EdgeVertexStep toV(), inV(), outV(), bothV()
NotStep not()
TraversalFilterStep where()
AndStep and(A,B)
PageRankVertexProgramStep pageRank()
DedupGlobalStep dedup()
OrderGlobalStep order()
LimitGlobalStep limit()
SelectStep as() and select()
OrStep or()

This query finds people who know each other and demonstrates the as() and select() methods:

g.V().as("a").out("knows").as("b").out("knows")
  .where(P.eq("a")).select("a", "b").by("name").show
+-----+-----+
|  a  |  b  |
+-----+-----+
|Alice| Bob |
| Bob |Alice|
+-----+-----+
Table 2. TinkerPop update steps and methods
Steps Methods
DropStep V().drop(), E().drop(), properties().drop()
AddPropertyStep property(name, value, ...)

DseGraphFrame can be used to drop millions of vertices or edges at once, and is much faster for bulk property updates than Gremlin OLAP or OLTP.

For example this query drops all person vertices and their associated edges:

g.V().hasLabel("person").drop().iterate()

Using DseGraphFrame in Scala

GraphTraversal is a Java interface, and extends the Java Iterator interface. To iterate through the results of a traversal as a DataFrame use the df() method. DseGraphFrame supports implicit conversion to DataFrame.

The following example will traverse the vertices of a graph using TinkerPop and then show the result as a DataFrame.

g.V().out().show

In some cases you may need to use the TinkerPop Java API to get the correct TinkerPop objects.

For example, to extract the DSE Graph Id object the Traversal Java iterator can be converted to a Scala iterator which allows direct access to the TinkerPop representation of the Id. This method allows you to use the original Id instead of the DataFrame methods which return the DataFrame String representation of the Id, you can also use the toList() and toSet() methods to set the appropriate ID.

import scala.collection.JavaConverters._
for(i <-g.V().id().asScala) println (i)
{~label=vertex, community_id=748226688, member_id=0}
{~label=custom, name=Name, value=1}
g.V.id.toSet
res18: java.util.Set[Object] = [{~label=demigod, community_id=224391936, member_id=0}, ...

The TinkerPop P (predicate) and T (constant) classes are imported by the Spark shell automatically.

g.E().groupCount().by(T.label)
g.V().has("age", P.gt(30)).show

For standalone applications, import theses classes.

import org.apache.tinkerpop.gremlin.structure.T
import org.apache.tinkerpop.gremlin.process.traversal.P
import org.apache.tinkerpop.gremlin.process.traversal.dsl.graph.__

Scala is not always able to infer the return type, especially in the Spark shell. The property values of the type should be provided explicitly.

g.V().values[Any]("name").next()

Or similarly:

val n: String = g.V().values("name").next()

Explicitly set the type when dropping properties.

g.V().properties[Any]("age", "name").drop().iterate()

In this case, using the DataFrame API is easier as you do not need to specify the type.

g.V().properties("age", "name").drop().show()
++
||
++
++
g.V().values("age").show()
+-----+
|  age|
+-----+
|10000|
Table 3. Using Java methods in DseGraphFrame Scala applications
Method Use case Example
hasNext() You want to know if there's a result, but you don't care about the value. Did Alice create any other vertices
g.V().has("name", "Alice").outE("created").hasNext()
next() You know that there is at least 1 result and you want to get the first one (or the second if you call it twice, and so on). Get the vertex label distribution. Group steps will always return exactly 1 result.
g.V().groupCount().by(label).next()
iterate() You just want to execute the traversal, but don't care about the result and whether it did anything at all. Set all person's ages to 10.
g.V().property("age", 10).iterate()
toList(), toSet() You expect the result to contain an arbitrary number of items and you want to get all of them. Get all the people Alice knows.
g.V().has("name", "Alice").out("knows").toList()