Chained or-clauses return incorrect result set

Ben Sheppard

unread,

Oct 7, 2020, 10:24:15 AM10/7/20

to JanusGraph users

If I execute a gremlin query for two or-clauses, containing two conditions each, the resulting query JanusGraph executes a single or-clause with four conditions. This is problematic as the result set is a superset of what it should actually be.

A basic setup using TinkerGraph or JanusGraph+inmemory does not seem to exhibit this issue.

To demonstrate, Ive set up a simple example which uses 4 boolean flags below. In the result query, the third element should not have been returned as it fails the second or-clause. My test setup is a simple Cassandra/Elasticsearch backend (JanusGraph 0.5.2):

gremlin> graph = JanusGraphFactory.open('test.conf')
==>standardjanusgraph[cql:[172.30.30.221]]
gremlin> mgmt = graph.openManagement()
==>org.janusgraph.graphdb.database.management.ManagementSystem.643ecfef
gremlin> sample = mgmt.makeVertexLabel('sample').make()
==>sample
gremlin> prop_a = mgmt.makePropertyKey('a').dataType(Boolean.class).make()
==>a
gremlin> prop_b = mgmt.makePropertyKey('b').dataType(Boolean.class).make()
==>b
gremlin> prop_c = mgmt.makePropertyKey('c').dataType(Boolean.class).make()
==>c
gremlin> prop_d = mgmt.makePropertyKey('d').dataType(Boolean.class).make()
==>d
gremlin> mgmt.addProperties(sample, prop_a, prop_b, prop_c, prop_d)
==>sample
gremlin> mgmt.buildIndex('propTest', Vertex.class).addKey(prop_a).addKey(prop_b).addKey(prop_c).addKey(prop_d).buildMixedIndex('search')
==>propTest
gremlin> mgmt.commit()
==>null
gremlin> g = graph.traversal()
==>graphtraversalsource[standardjanusgraph[cql:[172.30.30.221]], standard]
gremlin> g.addV('test').property('a', true).property('b', true).property('c', true).property('d', true).
......1> addV('test').property('a', true).property('b', false).property('c', true).property('d', false).
......2> addV('test').property('a', false).property('b', true).property('c', false).property('d', true).
......3> addV('test').property('a', false).property('b', false).property('c', true).property('d', false)
==>v[4280]
gremlin> g.tx().commit()
==>null
gremlin>
gremlin>
gremlin> g.V().
......1> or(
......2> has('a', true),
......3> has('b', true)
......4> ).
......5> or(
......6> has('c', false),
......7> has('d', true)
......8> ).propertyMap()
==>[a:[vp[a->false]],b:[vp[b->true]],c:[vp[c->false]],d:[vp[d->true]]]
==>[a:[vp[a->true]],b:[vp[b->true]],c:[vp[c->true]],d:[vp[d->true]]]
==>[a:[vp[a->true]],b:[vp[b->false]],c:[vp[c->true]],d:[vp[d->false]]]
gremlin> g.V().
......1> or(
......2> has('a', true),
......3> has('b', true)
......4> ).
......5> or(
......6> has('c', false),
......7> has('d', true)
......8> ).explain()
==>Traversal Explanation

===========================================================================================================================================================================================

Original Traversal [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

ConnectiveStrategy [D] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

MatchPredicateStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

FilterRankingStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

InlineFilterStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

CountStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

IncidentToAdjacentStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

RepeatUnrollStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

PathRetractionStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

EarlyLimitStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

AdjacentToIncidentStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

LazyBarrierStrategy [O] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

AdjacentVertexFilterOptimizerStrategy [P] [GraphStep(vertex,[]), OrStep([[HasStep([a.eq(true)])], [HasStep([b.eq(true)])]]), OrStep([[HasStep([c.eq(false)])], [HasStep([d.eq(true)])]])]

JanusGraphStepStrategy [P] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

AdjacentVertexHasIdOptimizerStrategy [P] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

AdjacentVertexIsOptimizerStrategy [P] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

JanusGraphLocalQueryOptimizerStrategy [P] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

JanusGraphIoRegistrationStrategy [P] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

ProfileStrategy [F] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

StandardVerificationStrategy [V] [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

Final Traversal [Or(JanusGraphStep([],[a.eq(true)]),JanusGraphStep([],[b.eq(true)]),JanusGraphStep([],[c.eq(false)]),JanusGraphStep([],[d.eq(true)]))]

Ive also tried wrapping the or-clauses in an and-clause, but it gets stripped out by strategies. Ive also inspected the query being sent to ElasticSearch, which still just shows 4 individual bool.must clauses wrapped inside a single bool.should.

Ben Sheppard

unread,

Oct 7, 2020, 4:24:44 PM10/7/20

to JanusGraph users

Ive retested JanusGraph+inmemory and found it to be susceptible as well (my first test was in the same shell as the TinkerGraph):

gremlin> graph = JanusGraphFactory.build().set('storage.backend', 'inmemory').open()
==>standardjanusgraph[inmemory:[127.0.0.1]]
gremlin> mgmt = graph.openManagement()
==>org.janusgraph.graphdb.database.management.ManagementSystem.798dad3d

gremlin> sample = mgmt.makeVertexLabel('sample').make()
==>sample
gremlin> prop_a = mgmt.makePropertyKey('a').dataType(Boolean.class).make()
==>a
gremlin> prop_b = mgmt.makePropertyKey('b').dataType(Boolean.class).make()
==>b
gremlin> prop_c = mgmt.makePropertyKey('c').dataType(Boolean.class).make()
==>c
gremlin> prop_d = mgmt.makePropertyKey('d').dataType(Boolean.class).make()
==>d
gremlin> mgmt.addProperties(sample, prop_a, prop_b, prop_c, prop_d)
==>sample

gremlin> mgmt.commit()
==>null
gremlin> g = graph.traversal()

==>graphtraversalsource[standardjanusgraph[inmemory:[127.0.0.1]], standard]

gremlin> g.addV('test').property('a', true).property('b', true).property('c', true).property('d', true).
......1> addV('test').property('a', true).property('b', false).property('c', true).property('d', false).
......2> addV('test').property('a', false).property('b', true).property('c', false).property('d', true).
......3> addV('test').property('a', false).property('b', false).property('c', true).property('d', false)

==>v[4176]
gremlin> g.tx().commit()
==>null

gremlin> g.V().
......1> or(
......2> has('a', true),
......3> has('b', true)
......4> ).
......5> or(
......6> has('c', false),
......7> has('d', true)
......8> )

==>[a:[vp[a->true]],b:[vp[b->true]],c:[vp[c->true]],d:[vp[d->true]]]
==>[a:[vp[a->true]],b:[vp[b->false]],c:[vp[c->true]],d:[vp[d->false]]]
==>[a:[vp[a->false]],b:[vp[b->true]],c:[vp[c->false]],d:[vp[d->true]]]

Li, Boxuan

unread,

Oct 7, 2020, 9:08:31 PM10/7/20

to janusgra...@googlegroups.com

Thanks Ben, I can reproduce it. Looks like a bug with Or support. Would you mind opening an issue on JanusGraph GitHub? Otherwise I can do so. Thanks!

Best regards,

Boxuan

--
You received this message because you are subscribed to the Google Groups "JanusGraph users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to janusgraph-use...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/janusgraph-users/9bd160d5-4ed7-4c9a-90cf-4bbb316203d8n%40googlegroups.com.

Ben Sheppard

unread,

Oct 7, 2020, 9:31:10 PM10/7/20

to JanusGraph users

Thanks for the quick response! Issue submitted: https://github.com/JanusGraph/janusgraph/issues/2231

Reply all

Reply to author

Forward