Value Aggregation along Traversal Gremlin Cosmos DB

Question

I need to perform a query of the following form: I have a tree structure with costs at the leaf node. I need a single query to give me all the aggregated costs under the root node.

For example in the above graph, I would expect an output from my query like

{ 1: 6, 2: 4, 3: 2, 4: 1, 5: 1, 6: 2, 7: 1, 8: 1}

I was looking into using the 'sack' step from the Gremlin API for this, but cosmosDB doesn't seem to support sacks currently. I also tried storing a pseudo-property of "aggregated-cost" and working my way up from the leaf nodes, but I was unable to figure out how to store a dynamic value at each node as a property that is local to only that node. Is this kind of query possible given these constraints?

stephen mallette · Accepted Answer · 2018-07-05T20:58:13.770

When asking questions about Gremlin, it is always best to include a short Gremlin script of sample data:

g.addV().property('id',1).as('1').
  addV().property('id',2).as('2').
  addV().property('id',3).as('3').
  addV().property('id',4).property('cost',1).as('4').
  addV().property('id',5).property('cost',1).as('5').
  addV().property('id',6).property('cost',2).as('6').
  addV().property('id',7).property('cost',1).as('7').
  addV().property('id',8).property('cost',1).as('8').
  addE('link').from('1').to('2').
  addE('link').from('1').to('3').
  addE('link').from('2').to('4').
  addE('link').from('2').to('5').
  addE('link').from('2').to('6').
  addE('link').from('3').to('7').
  addE('link').from('3').to('8').iterate()

With the steps available in CosmosDB I think that the closest you might be able to get is this:

gremlin> g.V().
......1>   group().
......2>     by('id').
......3>     by(emit(has('cost')).
......4>        repeat(out()).
......5>        values('cost').
......6>        fold())
==>[1:[1,1,2,1,1],2:[1,1,2],3:[1,1],4:[1],5:[1],6:[2],7:[1],8:[1]]

The group() helps produce the Map structure you wanted. Then for each vertex you group on you use repeat() to traverse out until you reach the leaf vertices. Note that emit() is ensuring that only those vertices that are leaves with the "cost" property are being returned for purpose of the result.

The reason I say that this is about as close as you can get with CosmosDB is because I don't see that CosmosDB supports the sum() step here. If it did then:

gremlin> g.V().
......1>   group().
......2>     by('id').
......3>     by(emit(has('cost')).
......4>        repeat(out()).
......5>        values('cost').
......6>        sum())
==>[1:6,2:4,3:2,4:1,5:1,6:2,7:1,8:1]

I guess you will have to do that final computation on the returned result yourself.

For others (or when CosmosDB supports sack() in the future) you can do:

gremlin> g.V().has('cost').
......1>   sack(assign).
......2>     by('cost').
......3>   emit().
......4>     repeat(__.in('link')).
......5>   group().
......6>     by('id').
......7>     by(sack().sum())
==>[1:6,2:4,3:2,4:1,5:1,6:2,7:1,8:1]

Courtesy of the Gremlin Guru.

Thanks a lot! Once cosmos releases support for sacks, would there be a way to do it using those? Because (correct me if I'm wrong), the current way repeats a lot of computation with repeat(out()) at each node? — Raghav Anand, Jul 05 '18 at 18:23
Also, just an fyi for anyone who sees this later, cosmosDB does support sum! — Raghav Anand, Jul 05 '18 at 18:29

Value Aggregation along Traversal Gremlin Cosmos DB

1 Answers1