Hello,
Many many moons ago, back when Faunus was a thing, Matthias Bröcheler said --
"Can you make it so we can infer which edge labels are needed for the OLAP job
so I can create a Titan-speciifc push down predicate?"
Well, that was Gremlin 2.x and traversal introspection in those days was not as
easy as it is with Gremlin 3.x. In order to satiate Matthias' (dying) wish, I
created GraphFilterStrategy (now in master/ slated for 3.2.1).
https://github.com/apache/incubator-tinkerpop/blob/master/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/process/computer/traversal/strategy/optimization/GraphFilterStrategy.java
In 3.2.0, we introduced the concept of a GraphFilter which allows the user to
specify a push-down predicate for selecting a subgraph needed for an OLAP job.
https://github.com/apache/incubator-tinkerpop/blob/master/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/process/computer/GraphFilter.java
In short, you can do things like:
g = graph.traversal().withComputer(compute.edges(bothE().limit(0)))
g.V().count()
g.V().has("age",gt(30)).count()
…
In the example above, if your traversals won't touch edges, then filter out the
edges. With GraphFilterStrategy, the GraphFilter no longer needs to be
specified by the user and in fact, a GraphFilter is automatically generated
based on traversal introspection.
https://github.com/apache/incubator-tinkerpop/blob/master/gremlin-core/src/main/java/org/apache/tinkerpop/gremlin/process/computer/traversal/strategy/optimization/GraphFilterStrategy.java#L110-L111
Now you can just do:
g = graph.traversal().withComputer()
g.V().count()
g.V().has("age",gt(30)).count()
…
…and each traversal will have a custom created GraphFilter.
For providers that have low-level support for GraphFilter, this can be huge as
only slices of the graph that will actually be touched by a traversal will be
selected from the source graph reducing the amount of data transferred around
the cluster, the amount of data held in memory, the amount of data needed to be
GC inspected, etc.
Enjoy,
Marko.
http://markorodriguez.com
