[ https://issues.apache.org/jira/browse/BEAM-7420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16859306#comment-16859306 ]
Mike Kaplinskiy commented on BEAM-7420: --------------------------------------- Sorry for the late response. I manually excluded the flink-runner bits (I'm on a somewhat unorthodox build system (Buck) so I simply removed the dependencies). When the job runs in a flink cluster, the cluster includes all the necessary flink jars (as well as some others - e.g. to write checkpoints/savepoints to S3). I'm curious - would it make sense to mark the flink dependencies in beam as {{provided}}? I'm not 100% sure it's possible to run a flink cluster from just the java class files - you probably need the rest of the flink distribution anyway, which definitely comes with the flink jars? > Including the Flink runner causes exceptions unless running in a flink > environment > ---------------------------------------------------------------------------------- > > Key: BEAM-7420 > URL: https://issues.apache.org/jira/browse/BEAM-7420 > Project: Beam > Issue Type: Bug > Components: runner-flink > Reporter: Mike Kaplinskiy > Assignee: Maximilian Michels > Priority: Major > > The {{FlinkPipelineOptions}} imports various Flink enums which make it > impossible to e.g. run the direct runner with the same classpath but without > the flink runtime. The fix is potentially easy - make the arguments strings > and convert them to enums at the callsites. -- This message was sent by Atlassian JIRA (v7.6.3#76005)