Pei He created BEAM-2681: ---------------------------- Summary: TransformHierarchy.Node getInputs() TupleTags doesn't match previous ParDo.MultiOutput's getOutputs(). Key: BEAM-2681 URL: https://issues.apache.org/jira/browse/BEAM-2681 Project: Beam Issue Type: Bug Components: sdk-java-core Reporter: Pei He Assignee: Thomas Groh Priority: Minor
For example, ParDo.MultiOutput ->PCollectionTuple [tag1 -> pc1 (tag_pc1), tag2 -> pc2 (tag_pc2)] pc1 (tag_pc1) -> transform_a pc2 (tag_pc2) -> transform_b During translation, calling getOutputs() for Node ParDo.MultiOutput will returns map keyed by tag1 and tag2, and calling getInputs() for Node transform_a and transform_b will returns map keyed by tag_pc1 and tag_pc2 respectively. Every runner will need to add special case for ParDo translation in order to handle this. Is this a bug or intended? (If it is intended, should we document it somewhere for runner implementors?) -- This message was sent by Atlassian JIRA (v6.4.14#64029)