Are you talking about transform names? The main reason was because for runners that support updating pipelines in place, the only way to do so safely is if the runner can perfectly identify which transforms in the new graph match the ones in the old graph. There's no good way to auto generate names that will stay stable across updates - even small changes to the pipeline might change the order of nodes in the graph, which could result in a corrupted update.
However, if you don't care about update, Beam can auto generate these names for you! When you call PCollection.apply (if using BeamJava), simply omit the name parameter and Beam will auto generate a unique name for you. Reuven On Sat, Sep 30, 2023 at 11:54 AM Joey Tran <joey.t...@schrodinger.com> wrote: > After writing a few pipelines now, I keep getting tripped up from > accidentally have duplicate labels from using multiple of the same > transforms without labeling them. I figure this must be a common complaint, > so I was just curious, what the rationale behind this design was? My naive > thought off the top of my head is that it'd be more user friendly to just > auto increment duplicate transforms, but I figure I must be missing > something > > Cheers, > Joey >
