[ 
https://issues.apache.org/jira/browse/BEAM-17?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15509658#comment-15509658
 ] 

Amit Sela commented on BEAM-17:
-------------------------------

Same for streaming, with UnboundedSource, but the challenge there is greater 
since a "checkpointing" mechanism needs to be available on the workers.

> Add support for new Beam Source API
> -----------------------------------
>
>                 Key: BEAM-17
>                 URL: https://issues.apache.org/jira/browse/BEAM-17
>             Project: Beam
>          Issue Type: Improvement
>          Components: runner-spark
>            Reporter: Amit Sela
>            Assignee: Amit Sela
>
> The API is discussed in 
> https://cloud.google.com/dataflow/model/sources-and-sinks#creating-sources
> To implement this, we need to add support for 
> com.google.cloud.dataflow.sdk.io.Read in TransformTranslator. This can be 
> done by creating a new SourceInputFormat class that translates from a DF 
> Source to a Hadoop InputFormat. The two concepts are pretty-well aligned 
> since they both have the concept of splits and readers.
> Note that when there's a native HadoopSource in DF, it will need 
> special-casing in the code for Read since we'll be able to use the underlying 
> InputFormat directly.
> This could be tested using XmlSource from the SDK.



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

Reply via email to