Backtrack Pattern

Many times its desirable to traverse a particular path and if some criteria is met along that path, then go back to the element from n-steps ago. Examples of such uses cases include:

• “What is the age of my friends who have friends who are older than 30 years old?”
• “What other products have my friends purchased who have also purchased a product of type X?”

g = TinkerGraphFactory.createTinkerGraph()

The query below says, in plain English: “What are the ages of the people that know people that are 30+ years old?” The call to back(2) refers to the elements 2 steps ago that have paths up to the back(2) step (i.e. back to the V step). In the example below, back(2) “wraps” out('knows').filter{it.age > 30}.

gremlin> g.V.out('knows').filter{it.age > 30}.back(2).age
==>29
gremlin> g.V.as('x').outE('knows').inV.filter{it.age > 30}.back('x').age
==>29

A more complicated example is provided over the Grateful Dead graph diagrammed in Defining a More Complex Property Graph.

g = new TinkerGraph()
g.loadGraphML('data/graph-example-2.xml')

The example query below states the following:

• get the song with id 89 (Dark Star).
• get all the songs that follow Dark Star in concert (A)
• get the singers of those songs (B)
• filter to only those songs that are sung by Jerry Garcia (C)
• go back 2 steps to yield those songs that follow Dark Star and are sung by Jerry Garcia.
• get the names of those songs that follow Dark Star and are sung by Jerry Garcia.

gremlin> g.v(89).out('followed_by').out('sung_by').has('name','Garcia').back(2).name
==>EYES OF THE WORLD
==>SING ME BACK HOME
==>MORNING DEW
==>HES GONE
==>CHINA DOLL
==>WHARF RAT
==>BROKEDOWN PALACE
==>TERRAPIN STATION
==>DEAL
==>ATTICS OF MY LIFE
==>COMES A TIME
==>STELLA BLUE
==>BERTHA

In order to determine how many steps to go back, the GremlinPipeline.toString() can be handy for displaying all the steps in an expression.

gremlin> println g.v(89).out('followed_by').out('sung_by').has('name','Garcia')             
[StartPipe, OutPipe(followed_by), OutPipe(sung_by), PropertyFilterPipe(name,EQUAL,Garcia)]
==>null

Now, using the back step, notice how back(3) wraps 3 pipes prior to it. The name of the pipe in Pipes is BackFilterPipe.

gremlin> println g.v(89).out('followed_by').out('sung_by').has('name','Garcia').back(2).name
[StartPipe, OutPipe(followed_by), BackFilterPipe([OutPipe(sung_by), PropertyFilterPipe(name,EQUAL,Garcia)]), PropertyPipe(name)]
==>null

Moreover, you can use as to label a back point.

gremlin> println g.v(89).out('followed_by').as('x').out('sung_by').has('name','Garcia').back('x').name
[StartPipe, AsPipe(x,OutPipe(followed_by)), BackFilterPipe([OutPipe(sung_by), PropertyFilterPipe(name,EQUAL,Garcia)]), PropertyPipe(name)]
==>null