[
https://issues.apache.org/jira/browse/FLINK-21445?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17288900#comment-17288900
]
Matthias commented on FLINK-21445:
----------------------------------
Thanks for creating this issue. Your finding makes sense after looking into the
code. I assume that the configuration would be necessary in all three cases
which are currently handled by {{ClassPathPackagedProgramRetriever}}. Taking
this into account I am wondering whether we should refactor
{{ClassPathPackagedProgramRetriever}} a bit to have separate classes for the
Python-based {{PackagedProgram}} and the {{PackagedProgram}} with and without a
jar. This way, we would avoid redundant test implementation. Right now, there
is no testing for the Python-based implementation as far as I can see.
[~fly_in_gis] does that make sense?
> Application mode does not set the configuration when building PackagedProgram
> -----------------------------------------------------------------------------
>
> Key: FLINK-21445
> URL: https://issues.apache.org/jira/browse/FLINK-21445
> Project: Flink
> Issue Type: Bug
> Components: Deployment / Kubernetes, Deployment / Scripts,
> Deployment / YARN
> Affects Versions: 1.11.3, 1.12.1, 1.13.0
> Reporter: Yang Wang
> Priority: Critical
> Fix For: 1.11.4, 1.13.0, 1.12.3
>
>
> Application mode uses {{ClassPathPackagedProgramRetriever}} to create the
> {{PackagedProgram}}. However, it does not set the configuration. This will
> cause some client configurations not take effect. For example,
> {{classloader.resolve-order}}.
> I think we just forget to do this since we have done the similar thing in
> {{CliFrontend}}.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
