Gremlin Snippets are typically short and fun dissections of some aspect of the Gremlin language. For a full list of all steps in the Gremlin language see the Reference Documentation of Apache TinkerPop™. This snippet is based on Gremlin 3.4.6.Please consider bringing any discussion or questions about this snippet to the Gremlin Users Mailing List.



The following dataset represents text messages being sent among a group of “person” vertices:

g.addV('person').property('name', 'marko').as('a').
  addV('person').property('name', 'vadas').as('b').
  addV('person').property('name', 'josh').as('c').
  addV('person').property('name', 'peter').as('d').
  addV('person').property('name', 'daniel').as('e').
  addE('text').property('count', 10).from('a').to('b').
  addE('text').property('count', 12).from('a').to('b').
  addE('text').property('count', 20).from('a').to('e').
  addE('text').property('count', 25).from('b').to('a').
  addE('text').property('count', 30).from('d').to('a').
  addE('text').property('count', 25).from('d').to('a').
  addE('text').property('count', 10).from('e').to('c')

We might become interested in doing some sort of analysis that answered the question of “Who does ‘marko’ send messages to and receive messages from and how many of those messages are there per person?” The start to such a traversal surely involves finding “marko”, traversing incoming and outgoing edges, and then grouping on the “person” vertices to which “marko” is connected:

gremlin> g.V().has('person','name','marko').
......1>   bothE().
......2>   group().
......3>     by(otherV().values('name')).next()
==>daniel=[e[12][0-text->8]]
==>peter=[e[14][6-text->0], e[15][6-text->0]]
==>vadas=[e[10][0-text->2], e[11][0-text->2], e[13][2-text->0]]

The structure of the result is beginning to form itself as we can see the names of the people who send and receive messages to and from “marko”, but an important step remains. The values in that Map are lists of Edge objects and need to be converted to a form that explains the number of messages sent and received for each person. That value would be best represented as a Map within “sent” and “received” keys where their values represented the total number of messages.

While the first by() modulator to group()-step defines the value on which the grouping should occur (i.e. the key to the Map), the second by() modulator describes how to process the resultant values in the Map. As a quick demonstration, the following code demonstrates how the second by() will extract the “count” for each edge:

gremlin> g.V().has('person','name','marko').
......1>   bothE().
......2>   group().
......3>     by(otherV().values('name')).
......4>     by('count')
==>[daniel:[20],peter:[25,30],vadas:[10,12,25]]

Going back to the original purpose of our analysis, we can see that while these numbers could provide us with the “total number of messages sent and received per person” we can’t use them as-is to determine how many were sent and how many were received as we need to utlize the direction of the edge to figure that out. Since we want to reduce this Edge list into a Map with “sent” and “received” keys, it would be good to use project(). This step is helpful when we know the names of the keys ahead of time and want to push values into them.

gremlin> g.V().has('person','name','marko').
......1>   bothE().
......2>   group().
......3>     by(otherV().values('name')).
......4>     by(project('sent','received').fold()).next()
==>daniel=[{sent=e[29][17-text->25], received=e[29][17-text->25]}]
==>peter=[{sent=e[32][23-text->17], received=e[32][23-text->17]}, {sent=e[31][23-text->17], received=e[31][23-text->17]}]
==>vadas=[{sent=e[27][17-text->19], received=e[27][17-text->19]}, {sent=e[28][17-text->19], received=e[28][17-text->19]}, {sent=e[30][19-text->17], received=e[30][19-text->17]}]

The above demonstrates the basic Map structure we want to have, but it doesn’t sort out which edge represents “sent” and which are “received”. Those edges need to be filtered accordingly for each Map pair. One way to do this is with coalesce():

gremlin> g.V().has('person','name','marko').
......1>   bothE().
......2>   group().
......3>     by(otherV().values('name')).
......4>     by(project('sent','received').
......5>          by(coalesce(filter(outV().has('name','marko')).values('count'), constant(0))).
......6>          by(coalesce(filter(inV().has('name','marko')).values('count'), constant(0))).
......7>        fold()).next()
==>daniel=[{sent=20, received=0}]
==>peter=[{sent=0, received=25}, {sent=0, received=30}]
==>vadas=[{sent=10, received=0}, {sent=12, received=0}, {sent=0, received=25}]

Note in the above bit of Gremlin that we use the inV() and outV() of each Edge instance to determine directionality. If the outV() (i.e. the outgoing Vertex of the Edge) is “marko” then we know he was the sender of the message and the reverse is true for inV() (i.e. the incoming Vertex of the Edge). Since any given Edge can only be one of “sent” or “received” but not “both” then one of those by() modulators would have to return no elements. By using coalesce() and a constant(0) to represent that condition we first prevent a traversal error and second provide a numeric value of zero as a default which makes the next step of summation of all these Map instances quite straightforward:

gremlin> g.V().has('person','name','marko').
......1>   bothE().
......2>   group().
......3>     by(otherV().values('name')).
......4>     by(project('sent','received').
......5>          by(coalesce(filter(outV().has('name','marko')).values('count'), constant(0))).
......6>          by(coalesce(filter(inV().has('name','marko')).values('count'), constant(0))).
......7>        unfold().
......8>        group().
......9>          by(keys).
.....10>          by(select(values).sum())).next()
==>daniel={received=0, sent=20}
==>peter={received=55, sent=0}
==>vadas={received=25, sent=22}

Each Map is deconstructed to key/value pairs by unfold() and is then reconstructed with group() and the values reduced by way of sum(). This approach is a common pattern in collection manipulation and should be recgonizable to more advanced Gremlin users.