It is an interesting requirement. Usually, we suggest our users to add these dependencies in lib directory or ship via --yarnship cli option.
If you really do not want to do this, I think you could ship an updated yarn-site.xml, set "yarn.application.classpath" including the mapreduce jars. Then all the TaskManagers will have them in the classpath. Best, Yang Richard Moorhead <richard.moorh...@gmail.com> 于2020年9月26日周六 上午4:56写道： > We're submitting jobs to a flink session running on YARN. Our > yarn-site.xml does not have our mapreduce libs in yarn.application.classpath. > Our application code has some provided dependencies from > hadoop-mapreduce-client-core. Without editing our yarn-site, or adding > those libraries to the flink lib dir, how can we extend our session's > classpath? >
