damccorm commented on issue #21721: URL: https://github.com/apache/beam/issues/21721#issuecomment-1243900254
I think my personal lean would be: 1) Leave the metrics namespace definition in the `ModelHandler` classes. These have the advantage of being model specific; for example, the `PytorchModelHandlerTensor` has a metrics namespace `RunInferencePytorch`. Getting that level of model-specificity is harder to do elsewhere. 2) Add a method (or probably just a class parameter) to override the namespace on the `RunInference` base class. This gives users full control. I'd actually also lean towards making this a postfix to the default metrics namespace instead of a full override since this makes usage tracking easier for runners that can do that (probably just dataflow at this point). It also guarantees that similar metrics are grouped together alphabetically. It does remove some user control though and I could be convinced this is a bad idea. > Anand, I think there are a few options how to do this. What do you think about outlining them in a short doc, and then we can pick one? Agreed - so far we've collectively proposed 3 different options :) it would be good to call out pros/cons side by side. -- 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]
