[jira] [Commented] (SPARK-6974) Possible error in TwitterPopularTags

["Nicola Ferraro (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-6974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14499484#comment-14499484 ] 

Nicola Ferraro commented on SPARK-6974:
---------------------------------------

Sorry, I will post the problem to user@.

Just to point out what I mean.

1) topCounts60 is a windowed DStream, consisting of multiple RDDs, say 3 RDDs
2) topCounts60.foreachRDD(rdd => { ... function_body ... }): function body is executed 3 times during the micro-batch
3) function_body prints the top k tags of the RDD received as argument, not of the DStream

So, if the source DStream in step 1 is composed of 3 RDDs, you get 3 prints of 3 different top-k tags in latest step.

Maybe I am wrong, but this example works just because topCounts60 is made of just 1 RDD.

> Possible error in TwitterPopularTags
> ------------------------------------
>
>                 Key: SPARK-6974
>                 URL: https://issues.apache.org/jira/browse/SPARK-6974
>             Project: Spark
>          Issue Type: Bug
>            Reporter: Nicola Ferraro
>            Priority: Minor
>
> Looking at the example for Twitter popular tags in spark streaming (https://github.com/apache/spark/blob/master/examples/src/main/scala/org/apache/spark/examples/streaming/TwitterPopularTags.scala), it seems that the algorithm can have issues in some cases.
> Top k tags are computed using the following function on a DStream:
> topCounts60.foreachRDD(rdd => { ... print ... })
> But the function passed to "foreachRDD" is called multiple times when your DStream is composed of multiple RDDs, once per RDD in the DStream, resulting in multiple Top-k charts.
> Probably this scenario is unlikely to happen, because a previous transformation on the DStream (reduceByKeyAndWindow) collapses all RDDs of the stream into a single one.
> The problem is that this behavior is not stated in the documentation and can be changed in future versions.
> Moreover, computing correctly the topK chart in streaming seems impossible if you rely on the documentation only. But it is the base algorithm for many RT dashboard use cases.
> I have also tried to get some reply on stackoverflow (http://stackoverflow.com/questions/29539655/how-to-compute-the-top-k-words).



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org