Gremlin in Gaffer
Warning
GafferPop is still under development and has some limitations.
The implementation may not allow some advanced features of Gremlin and it's
performance is unknown in comparison to standard Gaffer OperationChains.
Gremlin is a query language for traversing graphs. It is a core component of the Apache Tinkerpop library and allows users to easily express more complex graph queries.
GafferPop is a lightweight Gaffer implementation of the TinkerPop framework, where TinkerPop methods are delegated to Gaffer graph operations.
The addition of Gremlin as query language in Gaffer allows users to represent
complex graph queries in a simpler language akin to other querying languages
used in traditional and NoSQL databases. It also has wide support for various
languages so for example, you can write queries in Python via the gremlinpython library
Tip
In-depth tutorials on Gremlin as a query language and its associated libraries can be found in the Apache Tinkerpop Gremlin docs.
Using Gremlin Queries in Gaffer
Gremlin was added to Gaffer in version 2.1 as a new graph query language and since version 2.2 a container image is provided allowing a Gremlin layer to be added to existing 2.x graphs. A full tutorial on setting up this image is provided in the administration guide.
This guide will use the Python API for Gremlin to demonstrate some basic capabilities and how they compare to standard Gaffer syntax.
To start querying in Gremlin we first need a reference to what is known as the
Graph Traversal. To obtain this we need to connect to a running Gremlin server,
similar to how a connection to the Gaffer REST API is needed if using
gafferpy. We can do this by first importing the required
libraries like so (many of these will be needed later for queries):
from gremlin_python.process.anonymous_traversal import traversal
from gremlin_python.driver.driver_remote_connection import DriverRemoteConnection
from gremlin_python.process.graph_traversal import __
We can then establish a connection to the Gremlin server and save a reference to
this (typically called g):
# Setup a connection with the gremlin server running on localhost
g = traversal().with_remote(DriverRemoteConnection('ws://localhost:8182/gremlin', 'g'))
Now that we have the traversal reference this can be used to spawn graph traversals and get results back.
Basic Gremlin Queries
Gremlin queries (similar to Gaffer queries) usually require a starting set of
entities to query from. Commonly Gremlin queries will be left without any IDs in
the starting seed which would be analogous to asking for all vertexes with
g.V() or, all edges with g.E(). However, if this type of query is used with
the Gafferpop library this will in effect call a GetAllElements operation which
is less than ideal. Its therefore, highly recommended to always seed the query with
IDs.
We will use the following graph to demonstrate the basic usage of gremlin compared to Gaffer.
graph LR
A(["Person
ID: John"])
--
"Created
weight: 0.2"
-->
B(["Software
ID: 1"])
A
--
"Created
weight: 0.6"
-->
C(["Software
ID: 2"])Now say we wanted to get all the vertexes connected from the Person
vertex "John" via a Created edge (essentially all the things "John"
has created):
Gremlin is 'lazy' so will only execute your query if you request it
to by using a to_list() or calling next() on the iterator.
# We seed with "John" and traverse over any "Created" out edges
g.V("John").out("Created").element_map().to_list()
Result:
Note standard Gaffer you must know the group of the target vertexes you want returned otherwise edges will be also present in the result.
{
"class": "OperationChain",
"operations": [
{
"class": "GetAdjacentIds",
"input": [
{
"class": "EntitySeed",
"vertex": "John"
}
],
"view": {
"edges": {
"Created": {}
}
}
},
{
"class": "GetElements",
"view": {
"entities": {
"Software": {}
}
}
}
]
}
Result:
As you can see the Gremlin query is quite a bit easier to write and provides the results in a handy output to be reused. Now say if you wanted to apply some filtering on the same graph, the following is an example of how Gremlin handles this:
Get all the Created edges from vertex "John" that have a "weight"
greater than 0.4:
# If needed we run this through an 'element_map()' call to get more info on the edge
g.V("John").outE("Created").has("weight", P.gt(0.4)).to_list()
Result:
There are more example queries to be found in the Gremlin Getting Started docs.
Mapping Gaffer to TinkerPop
Some of the terminology is slightly different between TinkerPop and Gaffer, a table of how different parts are mapped is as follows:
| Gaffer | TinkerPop |
|---|---|
| Group | Label |
| Vertex | Vertex with default label of id |
| Entity | Vertex |
| Edge | Edge |
| Edge ID | A list with the source and destination of the Edge e.g. [dest, source] |
Created: September 13, 2023