[ https://issues.apache.org/jira/browse/BEAM-1325?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15839821#comment-15839821 ]
Luke Cwik commented on BEAM-1325: --------------------------------- Its not just that the Class-Path should be understood and added to filesToStage, the JVM specification says that "The manifest can define a JAR-class-path, which further extends the class path (but only while loading classes from that JAR)." This would change the way how a runner launches a worker as well since `filesToStage != files to put on classpath` which is something that Google Cloud Dataflow assumes today. See: http://docs.oracle.com/javase/7/docs/technotes/tools/findingclasses.html#jarclass > DataflowRunner support for Class-Path jars > ------------------------------------------ > > Key: BEAM-1325 > URL: https://issues.apache.org/jira/browse/BEAM-1325 > Project: Beam > Issue Type: New Feature > Components: runner-dataflow > Reporter: Daniel Halperin > Assignee: Davor Bonaci > > See https://github.com/GoogleCloudPlatform/DataflowJavaSDK/issues/538 > The code that infers {{--filesToStage}} from the classpath does not seem to > support a classpath jar. which uses a {{Class-Path}} manifest to indicate > which other jars are needed at runtime. This is apparently a standard, > approved way to pass the classpath. > We need to add this support to > {{DataflowRunner#detectClasspathResourcesToStage}}. > (As {{FlinkRunner}} has copied this code, we should probably either pull the > common utility somewhere public or update the FlinkRunner too?) -- This message was sent by Atlassian JIRA (v6.3.4#6332)