GH created SPARK-34517:
--------------------------
Summary: Unexpected result when using 'aggregateMessages' with
'outerJoinVertices'
Key: SPARK-34517
URL: https://issues.apache.org/jira/browse/SPARK-34517
Project: Spark
Issue Type: Bug
Components: GraphX
Affects Versions: 3.0.2
Environment: OpenJDK 1.8.0_275
Scala 2.12.10
Hadoop 3.3.0
Spark 3.0.2
Reporter: GH
Using 'Graph.aggregateMessages' and 'Graph.outerJoinVertices' together when
implementing an iterative graph-algorithm which is very similar to
'Graph.staticParallelPersonalizedPageRank'.
There will be a bug if 'outerJoinVertices' does not change the data-type of
vertices. More specifically, 'aggregateMessages' gives the same results for
some iteration, while the value of vertices was still changed and the attribute
of a message is fully depend on the value of its source vertex.
I temporarily solve the problem by changing the data-type when using
'outerJoinVertices' and then change it back with 'mapVertices'. It showed that
the problem is probably due to the incorrect updating of activeSet when
'outerJoinVertices' keeps the data-type of vertices.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
