[jira] [Commented] (BEAM-853) Force streaming execution on batch pipelines for testing.

["Amit Sela (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/BEAM-853?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15679706#comment-15679706 ] 

Amit Sela commented on BEAM-853:
--------------------------------

Streaming tests require forcing streaming mode on {{BoundedFromUnbounded}}, and {{PAssert}} requires proper {{GroupByKey}} implementation, so those two go together.

> Force streaming execution on batch pipelines for testing.
> ---------------------------------------------------------
>
>                 Key: BEAM-853
>                 URL: https://issues.apache.org/jira/browse/BEAM-853
>             Project: Beam
>          Issue Type: Bug
>          Components: runner-spark
>            Reporter: Amit Sela
>            Assignee: Amit Sela
>
> Beam's {{RunnableOnService}} tests for runners are written with bounded reads so the SparkRunner should force a streaming pipeline on an "all-batch" pipeline. 
> Currently the runner will decide the translation (batch/streaming) according to the input PCollection boundness (in case of input, according to output).
> One way to overcome this would be to override this behaviour in {{TestSparkRunner}}.
> Another challenge is the implementation itself - we could clearly read the RDD and create a single-RDD-DStream of it (following RDDs will be empty) but this will miss the point of testing streaming pipelines both for testing the execution of {{Read.Unbounded}} and for testing across-microbatches (tests state management).
> Finally, we have to consider how this is going to play nicely with {{SplittableDoFn}}.



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