Hi there, Currently, YARN Application mode only supports running a jar job. And as Lijie said, the main method is executed on JM, so if I understand correctly, your previous way of running execute() method on client side to submit a job to YARN cluster may not work. A quick workaround is to create such a job jar for submitting your SQL jobs and you can run this job jar using Application Mode. Thanks to Gyula's work, you can refer to https://github.com/apache/flink-kubernetes-operator/blob/main/examples/flink-sql-runner-example/src/main/java/org/apache/flink/examples/SqlRunner.java for more a minimal example.
Best, Biao Geng Lijie Wang <wangdachui9...@gmail.com> 于2022年8月2日周二 09:49写道： > Hi, > I think the difference between ApplicationMode and PerJob is just where > the main method is executed (ApplicationMode executes on JM, PerJob > executes on client side). So I think your original job code should work > well under ApplicationMode. Did you encounter any problems? > You can get more details about ApplicationMode in [1]. > > [1] > https://nightlies.apache.org/flink/flink-docs-master/docs/deployment/resource-providers/yarn/#application-mode > > Best, > Lijie > > Tamas Kiss <tak...@cloudera.com> 于2022年8月1日周一 19:30写道： > >> Hi Experts, >> >> We are planning to migrate our flink jobs from per-job mode to >> application mode in our platform. Our jobs are basically SQL scripts so we >> have some custom Java code to leverage Flink's SQL and Table API to build >> the execution environment and execute the jobs on Yarn. We would like to >> keep the current flow in our platform so we are looking for a way to run >> Flink SQL in application mode via some Java code. Does this seem to be a >> good approach to follow or should we look for a better solution? If this is >> a good choice, is there any API to support this? >> >> Thanks >> Tamas >> >
