Github user tillrohrmann commented on the issue:
https://github.com/apache/flink/pull/4926
Thanks for the review @steveloughran, @StephanEwen and @aljoscha.
If I understood it correctly, then we shouldn't upload the configuration
files from the client and instead try to find the configurations on the
cluster. This means that we either have to properly set up the classpath to
rely on the the default resources or to also look for the `yarn-site.xml` in
`HadoopUtils.getHadoopConfiguration`. At the moment we do a mixture of both.
`yarn-site.xml` is supposed to be on the classpath while the `core-site.xml`
and `hdfs-site.xml` can also be found if that's not the case. Maybe we should
add the explicit loading of `yarn-site.xml` to
`HadoopUtils.getHadoopConfiguration` to make it consistent.
---
