[
https://issues.apache.org/jira/browse/TINKERPOP-3164?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17956465#comment-17956465
]
Yang Xia edited comment on TINKERPOP-3164 at 6/5/25 10:54 PM:
--------------------------------------------------------------
Hi, I'm trying to replicate this in TinkerGraph with the Gremlin Console,
however, it looks like everything is as expected. As I see that you are using
JanusGraph, I wonder if this is an issue on the JanusGraph side?
For TinkerGraph (via Gremlin Console 3.7.2), by default the step by() filters
out unproductive traversers
([https://tinkerpop.apache.org/docs/3.7.3/reference/#by-step|https://tinkerpop.apache.org/docs/3.7.3/reference/#by-step)]),
it is actually expected that both queries omit the "Honda" vertex:
{code:java}
gremlin> g.V().hasLabel("Car").order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:0,label:Car,weight:5000,model:F-150,type:TRUCK,brand:Ford]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
gremlin> g.V().hasLabel("Car").choose(__.values("type")).option("TRUCK",
__.has("weight", P.lte(4500))).option("CROSSOVER",
__.identity()).order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
{code}
Now for TinkerGraph, you can allow unproductive by()'s to be passed through
using the "ProductiveByStrategy", for which you'll see that the "Honda" vertex
appears in the result of both queries:
{code:java}
gremlin>
g.withStrategies(ProductiveByStrategy).V().hasLabel("Car").order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:0,label:Car,weight:5000,model:F-150,type:TRUCK,brand:Ford]
==>[id:23,label:Car,model:CR-V,type:CROSSOVER,brand:Honda]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
gremlin>
g.withStrategies(ProductiveByStrategy).V().hasLabel("Car").choose(__.values("type")).option("TRUCK",
__.has("weight", P.lte(4500))).option("CROSSOVER",
__.identity()).order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:23,label:Car,model:CR-V,type:CROSSOVER,brand:Honda]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
{code}
Looking at your results with JanusGraph, it looks as if the
`ProductiveByStrategy` was enabled for the first query but disabled in the
second one, so I'm wondering if it has to do with some configuration in
JanusGraph that's worth a check.
was (Author: xiazcy):
Hi, I'm trying to replicate this in TinkerGraph with the Gremlin Console,
however, it looks like everything is as expected. As I see that you are using
JanusGraph, I wonder if this is an issue on the JanusGraph side?
For TinkerGraph (via Gremlin Console 3.7.2), by default the step by() filters
out unproductive traversers
([https://tinkerpop.apache.org/docs/3.7.3/reference/#by-step|https://tinkerpop.apache.org/docs/3.7.3/reference/#by-step)]),
it is actually expected that both queries omit the "Honda" vertex:
{code:java}
gremlin> g.V().hasLabel("Car").order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:0,label:Car,weight:5000,model:F-150,type:TRUCK,brand:Ford]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
gremlin> g.V().hasLabel("Car").choose(__.values("type")).option("TRUCK",
__.has("weight", P.lte(4500))).option("CROSSOVER",
__.identity()).order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
{code}
Now for TinkerGraph, you can allow unproductive by()'s to be passed through
using the "ProductiveByStrategy", for which you'll see that the "Honda" vertex
appears in the result of both queries:
{code:java}
gremlin>
g.withStrategies(ProductiveByStrategy).V().hasLabel("Car").order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:0,label:Car,weight:5000,model:F-150,type:TRUCK,brand:Ford]
==>[id:23,label:Car,model:CR-V,type:CROSSOVER,brand:Honda]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
gremlin>
g.withStrategies(ProductiveByStrategy).V().hasLabel("Car").choose(__.values("type")).option("TRUCK",
__.has("weight", P.lte(4500))).option("CROSSOVER",
__.identity()).order().by("brand").by("weight").elementMap()
==>[id:18,label:Car,weight:4900,model:Q8,type:CROSSOVER,brand:Audi]
==>[id:23,label:Car,model:CR-V,type:CROSSOVER,brand:Honda]
==>[id:13,label:Car,weight:4500,model:Frontier,type:TRUCK,brand:Nissan]
{code}
Looking at your results with JanusGraph, it looks as if the
`ProductiveByStrategy` was enabled for the first query but disabled in the
second one, so I'm wondering if it has to do with some configuration in
JanusGraph that's worth a check.
> Sorting doesn't work after the 'choose' step (if the property doesn't exist)
> ----------------------------------------------------------------------------
>
> Key: TINKERPOP-3164
> URL: https://issues.apache.org/jira/browse/TINKERPOP-3164
> Project: TinkerPop
> Issue Type: Bug
> Components: process
> Affects Versions: 3.7.2
> Reporter: Vladislav
> Priority: Major
>
> Elements are lost when sorting by property (after choose step) if the
> property is missing in a vertex
> Example graph:
> {code:java}
> gremlin> graph.traversal().addV("Car").property("brand",
> "Ford").property("model", "F-150").property("type",
> "TRUCK").property("weight", 5000).iterate()
> gremlin> graph.traversal().addV("Car").property("brand",
> "Nissan").property("model", "Frontier").property("type",
> "TRUCK").property("weight", 4500).iterate()
> gremlin> graph.traversal().addV("Car").property("brand",
> "Audi").property("model", "Q8").property("type",
> "CROSSOVER").property("weight", 4900).iterate()
> gremlin> graph.traversal().addV("Car").property("brand",
> "Honda").property("model", "CR-V").property("type", "CROSSOVER").iterate()
> {code}
> Base sort (without choose step)
> {code:java}
> gremlin>
> graph.traversal().V().hasLabel("Car").order().by("brand").by("weight").elementMap()
> 15:42:10 WARN
> org.janusgraph.graphdb.transaction.StandardJanusGraphTx$3.execute - Query
> requires iterating over all vertices [[~label = Car]]. For better
> performance, use indexes
> ==>[weight:4900,model:Q8,type:CROSSOVER,brand:Audi,id:8408,label:Car]
> ==>[weight:5000,model:F-150,type:TRUCK,brand:Ford,id:4136,label:Car]
> ==>[model:CR-V,type:CROSSOVER,brand:Honda,id:4328,label:Car]
> ==>[weight:4500,model:Frontier,type:TRUCK,brand:Nissan,id:4312,label:Car]
> {code}
> Sort with choose step (Honda CR-V is missing):
> {code:java}
> gremlin>
> graph.traversal().V().hasLabel("Car").choose(__.values("type")).option("TRUCK",
> __.has("weight", P.lte(4500))).option("CROSSOVER",
> __.identity()).order().by("brand").by("weight").elementMap()
> 15:43:53 WARN
> org.janusgraph.graphdb.transaction.StandardJanusGraphTx$3.execute - Query
> requires iterating over all vertices [[~label = Car]]. For better
> performance, use indexes
> ==>[weight:4900,model:Q8,type:CROSSOVER,brand:Audi,id:8408,label:Car]
> ==>[weight:4500,model:Frontier,type:TRUCK,brand:Nissan,id:4312,label:Car]
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
