[ https://issues.apache.org/jira/browse/FLINK-17500?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17099208#comment-17099208 ]
Kostas Kloudas commented on FLINK-17500: ---------------------------------------- [~uce] and [~trohrmann] with the introduction of the application mode, I believe that there is no reason for the {{ClassPathJobGraphRetriever}}. Actually currently on the master the {{ClassPathJobGraphRetriever}} is replaced by the {{ClassPathPackagedProgramRetriever}}. The reason for this is because the {{ClassPathJobGraphRetriever}} was executing the user's {{main()}} on the cluster, on a special environment, and only to extract the {{JobGraph}} which it was then submitting for execution. Now, with the application mode (see [FLIP-85|https://cwiki.apache.org/confluence/display/FLINK/FLIP-85+Flink+Application+Mode]), the user's {{main()}} is properly executed on the cluster, so there is no need for hijacking the {{execute()}}. Allowing the users though to execute a job graph that they have extracted by other means on the client may make sense. So adding the option to use the {{FileJobGraphRetriever}} on a standalone deployment can be a valid addition. > Deploy JobGraph from file in StandaloneClusterEntrypoint > -------------------------------------------------------- > > Key: FLINK-17500 > URL: https://issues.apache.org/jira/browse/FLINK-17500 > Project: Flink > Issue Type: Wish > Components: Deployment / Docker > Reporter: Ufuk Celebi > Priority: Minor > > We have a requirement to deploy a pre-generated {{JobGraph}} from a file in > {{StandaloneClusterEntrypoint}}. > Currently, {{StandaloneClusterEntrypoint}} only supports deployment of a > Flink job from the class path using {{ClassPathPackagedProgramRetriever}}. > Our desired behaviour would be as follows: > If {{internal.jobgraph-path}} is set, prepare a {{PackagedProgram}} from a > local {{JobGraph}} file using {{FileJobGraphRetriever}}. Otherwise, deploy > using {{ClassPathPackagedProgramRetriever}} (current behavior). > --- > I understand that this requirement is pretty niche, but wanted to get > feedback whether the Flink community would be open to supporting this > nonetheless. > > -- This message was sent by Atlassian Jira (v8.3.4#803005)