[ 
https://issues.apache.org/jira/browse/SPARK-18568?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15822397#comment-15822397
 ] 

Andrew Ray commented on SPARK-18568:
------------------------------------

RDD's have the same problem for cached collections of mutable objects
{code}
import scala.collection.mutable.{Map => MMap}
val rdd = sc.parallelize(MMap(1->1)::Nil)
rdd.cache()
val rdd2 = rdd.map(_ += 2 ->2)

scala> rdd2.collect()
res21: Array[scala.collection.mutable.Map[Int,Int]] = Array(Map(2 -> 2, 1 -> 1))
scala> rdd.collect()
res22: Array[scala.collection.mutable.Map[Int,Int]] = Array(Map(2 -> 2, 1 -> 1))
{code}

So I think the moral is it use immutable objects.

> vertex attributes in the edge triplet not getting updated in super steps for 
> Pregel API
> ---------------------------------------------------------------------------------------
>
>                 Key: SPARK-18568
>                 URL: https://issues.apache.org/jira/browse/SPARK-18568
>             Project: Spark
>          Issue Type: Bug
>          Components: GraphX
>    Affects Versions: 2.0.2
>            Reporter: Rohit
>
> When running the Pregel API with vertex attribute as complex objects. The 
> vertex attributes are not getting updated in the triplet view. For example if 
> the vertex attributes changes in first superstep for vertex"a" the triplet 
> src attributes in the send msg program for the first super step gets the 
> latest attributes of the vertex "a" but on 2nd super step if the vertex 
> attributes changes in the vprog the edge triplets are not updated with this 
> new state of the vertex for all the edge triplets having the vertex "a" as 
> src or destination. if I re-create the graph using g = Graph(g.vertices, 
> g.edges) in the while loop before the next super step then its getting 
> updated. But this fix is not good performance wise. A detailed description of 
> the bug along with the code to recreate it is in the attached URL.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to