Improve traversal query performance

63 views

Skip to first unread message

Yingjie Li

unread,

Aug 15, 2022, 6:29:10 AM8/15/22

to Gremlin-users

Hello All,

I would like to start from a particular source node ( id = 01F546'), traverse both direction for x number (4 in the below example) of hops , and list the properties of the first 200 destination nodes meeting certain creteria ( 'type' = output' in the belwo sample). I have set up the timeLimt to make sure the query return before time out.

I have created composite/mixed indexes on 'id', 'type' each.

For a graph of 250k nodes and 400k edges, the above query takes about ~7 second via gremline query console. What can be done to speed up the performance?

Thank you,

Yingjie

Query: g.V().has('id',eq('01F546')).emit().repeat(bothE().otherV().timeLimit(300000)).times(4).has('type',eq('output')).map(properties().group().by(key()).by(value())).dedup().limit(200).toList()

The output of the profile is:

HasStep([type.eq(output)]) 17 17 627.249 90.79

TraversalMapStep([JanusGraphMultiQueryStep, Jan... 17 17 3.129 0.45

JanusGraphMultiQueryStep 17 17 0.207

JanusGraphPropertiesStep(property) 180 180 1.520

\_condition=(PROPERTY AND visibility:normal)

\_orders=[]

\_isFitted=true

\_isOrdered=true

\_query=SliceQuery[0x40,0x60)

\_multi=true

\_vertices=1

optimization 0.001

GroupStep(key,[PropertyValueStep]) 17 17 1.204

PropertyValueStep 180 180 0.279

DedupGlobalStep(null,null) 11 11 0.202 0.03

RangeGlobalStep(0,20) 11 11 0.151 0.02

>TOTAL - - 690.901 -

optimization 0.001

optimization 0.000

optimization 0.001

optimization 0.000

optimization 0.001

optimization 0.000

optimization 0.001

optimization 0.000

optimization 0.001

optimization 0.000

GroupStep(key,[PropertyValueStep]) 109 109 6.813

PropertyValueStep 1083 1083 1.854

DedupGlobalStep(null,null) 76 76 0.610 0.01

>TOTAL - - 6721.808 -

HadoopMarc

unread,

Aug 18, 2022, 1:50:19 AM8/18/22

to Gremlin-users

Questions specific to JanusGraph are best posted on:

https://lists.lfaidata.foundation/g/janusgraph-users/topics

or https://github.com/JanusGraph/janusgraph/discussions

If you post on one of these please also add:

what storage backend was used
the profile output seems like a combination of two queries. What does the first one with a duration of 0.690 sec belong to
link to the SO question you posted

Best wishes, Marc

Op maandag 15 augustus 2022 om 12:29:10 UTC+2 schreef yingj...@gmail.com:

Yingjie Li

unread,

Aug 19, 2022, 8:24:19 AM8/19/22

to gremli...@googlegroups.com

Hello Marc,

Thanks for your reply, I posted here as I am looking for a better optimized gremlin query. I will post this to the Janusgraph group channel as you suggested. The profile is from one single query.

I am also looking for an enhanced query which gives me all the paths. e., edges and as well as the nodes (with nodes properties) of the traversal until it stops at nodes with certain criteria.

The below query

g.V().has('id',eq('01F546')).emit().repeat(bothE().otherV()).times(4).has('type',eq('output')).dedup().path().by(valueMap())

does produce the path with edges and nodes with properties, but as I am submitting gremlin command via gremlin-python, the returned result is text and needs to be customer parsed. Is there an alternative way to make the result more consumable ?

Yingjie

--
You received this message because you are subscribed to a topic in the Google Groups "Gremlin-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/gremlin-users/DCpNNTYMxVY/unsubscribe.
To unsubscribe from this group and all its topics, send an email to gremlin-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gremlin-users/83d27db4-d299-4dd4-b9a8-56cae588f874n%40googlegroups.com.

Reply all

Reply to author

Forward

0 new messages