[ 
https://issues.apache.org/jira/browse/BEAM-14403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17535090#comment-17535090
 ] 

Kenneth Knowles commented on BEAM-14403:
----------------------------------------

A while ago we had a similar problem with {{PTransform.expand}} methods doing 
different things based on the pipeline options, even though they were not 
supposed to. Many even checked what the runner was and built a totally 
different graph based on the runner, which meant they ended up not being truly 
portable, or not actually implemented in terms of the Beam model. We solved 
this problem by completely removing access to this information, so PTransforms 
had no choice but to have a portable expansion.

An analogous solution might be to have service options exist outside the 
pipeline options. Or to have pipeline options that are somehow not actually 
available in the code paths where the runner is submitting things.

I want to note that the issue here is in the {{DataflowRunner}} not the core 
SDK. I know many people do not care about that distinction. But the issue is in 
the logic that decides how to submit a Beam graph to Dataflow. So it is not 
really a generalized problem anyhow. It is a small enough code base that code 
review, and a general best practice of having as few options as possible, 
should suffice.

> Allow Prime to be used with Legacy workers
> ------------------------------------------
>
>                 Key: BEAM-14403
>                 URL: https://issues.apache.org/jira/browse/BEAM-14403
>             Project: Beam
>          Issue Type: Bug
>          Components: runner-dataflow
>    Affects Versions: 2.35.0, 2.36.0, 2.37.0, 2.38.0
>            Reporter: Robert Bradshaw
>            Assignee: Robert Bradshaw
>            Priority: P1
>             Fix For: 2.39.0
>
>          Time Spent: 50m
>  Remaining Estimate: 0h
>
> [https://github.com/apache/beam/blob/v2.37.0/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java#L2315]
>  assumes that enable_prime forces UW. This decision should be left up to the 
> service.



--
This message was sent by Atlassian Jira
(v8.20.7#820007)

Reply via email to