jrmccluskey commented on pull request #15400: URL: https://github.com/apache/beam/pull/15400#issuecomment-914518273
Yeah it seems to be that hard-coded expectation of a stream. That's what gave rise to the direct runner having to drop in a CoGBK ahead of the global combine, since the CoGBK outputs in that FullValue/ReStream pair. I'm a little surprised the Flink runner doesn't like it given that every other runner does, though. And adding logic to handle the nil ReStream case will likely not fix it, as that means it is outputting individual elements to process element instead of arranging them to be processed together. The next operation would only get the accumulation function output for a single element, not for all of the elements. Regardless, it makes sense that the expectation of getting a ReStream should be wrapped with some checking and a more helpful error message than just a generic index out of bounds panic. I added a quick check and we can go from there. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
