Hi all, Recently, I've been working on adding CombineFn.setup and CombineFn.teardown to Python SDK [1] (Java SDK is also on the list, but for now, it's only Python). Here's my implementation: https://github.com/apache/beam/pull/13048. Would someone be willing to take a look?
The functionality is based on `Operation.start` and `Operation.finish`. It works on all runners I tested: Flink, Spark and both direct runners. The only configuration that makes problems is the batch Dataflow worker without `--experiments=beam_fn_api` being set. I wonder how big the problem is though, since new pipelines are going to use the new Dataflow Runner V2 in the near future. Thank you, Kamil [1] https://issues.apache.org/jira/browse/BEAM-3736