[
https://issues.apache.org/jira/browse/BEAM-14421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17536236#comment-17536236
]
Yichi Zhang commented on BEAM-14421:
------------------------------------
The issue seems to be that dataflow service needs a FNAPI_BATCH or
FNAPI_STREAMING job type when turn on the runner v2, but service side doesn't
change the job type from 'STREAMING' or 'JAVA_BATCH_AUTOSCALING'
([https://github.com/apache/beam/blob/356dcc81832a1046458f7d86e3f2924eb7a883ca/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java#L1471)]
to FNAPI job type when seeing –dataflowServiceOptions=use_runner_v2
> --dataflowServiceOptions=use_runner_v2 is broken
> ------------------------------------------------
>
> Key: BEAM-14421
> URL: https://issues.apache.org/jira/browse/BEAM-14421
> Project: Beam
> Issue Type: Bug
> Components: runner-dataflow, sdk-java-core
> Affects Versions: 2.38.0
> Reporter: Robert Bradshaw
> Priority: P1
>
> Per [https://lists.apache.org/list.html?u...@beam.apache.org] passing
> `–dataflowServiceOptions=use_runner_v2` alone triggers pipeline errors. This
> is surprising as it should be an alias for experiments, and furthermore no
> action should be required on the SDK side to produce a runner v2 compatible
> graph.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
