Scott Wegner created BEAM-692:
---------------------------------
Summary: DisplayData integration tests don't exercise branches in
runner pipeline translation
Key: BEAM-692
URL: https://issues.apache.org/jira/browse/BEAM-692
Project: Beam
Issue Type: Bug
Reporter: Scott Wegner
Assignee: Scott Wegner
Display Data has various integration tests (for example
[AvroIOTest.testPrimitiveReadDisplayData()|https://github.com/apache/incubator-beam/blob/master/sdks/java/core/src/test/java/org/apache/beam/sdk/io/AvroIOTest.java#L287])
which test display data under a runner as follows:
# Build a PTransform instance to test
# Attach the transform to a pipeline and translate the pipeline with a runner
and a default set of PipelineOptions
# Aggregate display data from the translated pipeline primitive transforms and
verify it includes expected items.
This validates that runners retain a core set of display data through any
PTransform replacements. However, this falls short because pipeline translation
can vary per-runner based on conditions set in PipelineOptions (for example,
[StreamingOptions.isStreaming()|https://github.com/apache/incubator-beam/blob/master/runners/google-cloud-dataflow-java/src/main/java/org/apache/beam/runners/dataflow/DataflowRunner.java#L325]).
The tests are runner-agnostic and thus unaware of runner-specific translation
branching. As a result, it's easy to implement pipeline translation logic with
untested display data.
[BEAM-691|https://issues.apache.org/jira/browse/BEAM-691] is one such instance.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
