Python: CombineGlobally() needs to be able to produce expected output when processing multiple windows simultaneously

[BjornPrime]

What would you like to happen?

CombineGlobally() is supposed to reduce a PCollection to a single value. When given multiple windows to process, it would previously fail due to an assert statement that limited the max length of its output to one. I replaced this assert statement with a more descriptive error statement here (#24435), but a comprehensive solution for situations when the transform handles multiple windows, and thus multiple PCollections, at once, is still needed, since this situation is expected to arise from fairly run-of-the-mill use of the transform in streaming pipelines.

Issue Priority

Priority: 2 (default / most feature requests should be filed as P2)

Issue Components

• Component: Python SDK
• Component: Java SDK
• Component: Go SDK
• Component: Typescript SDK
• Component: IO connector
• Component: Beam examples
• Component: Beam playground
• Component: Beam katas
• Component: Website
• Component: Spark Runner
• Component: Flink Runner
• Component: Samza Runner
• Component: Twister2 Runner
• Component: Hazelcast Jet Runner
• Component: Google Cloud Dataflow Runner

[liferoad]

.take-issue

[robertwb]

We would need to define what it means to combine globally with defaults for non-trivial windowing. (Presumably that'd be a empty value for every empty window, so we'd need some way to enumerate windows.) If one wants to see combined values only when there is data, without_defaults should be advised and used.

[liferoad]

fixed by #26922