if x, y are both RooRealVar, this is exactly how you should do it. Maybe print the contents of the dataset to see that the values it contains make sense?
One thing that’s different if you have a 2D dataset is that a 2D integral is required for normalisation. Make sure that the range of the variable y is chosen such that the integral can actually converge. I don’t remember the details of the argus distribution, but I could imagine that y cannot be negative or something like this.