Currently the users need to use useDatasets in the configure phase to mention all the datasets that they will be using in the Worker/MR. This is not required anymore and thus these methods need to be deprecated.
Can we do this for all program types? Flows. Spark, Handlers, whatnot.
Yes we should. I will open a JIRA for that. Basically one should be able get and use datasets in runtime without specifying it during compile time. Right?
Right. But it will require code changes for Flows, definitely. I also am not sure about Mappers/Reducers. I know in the beforeSubmit() and onFinish() we can now dynamically instantiate, but the Mappers/Reducers might still rely on what was declared with useDataset().