pyspark high performance rolling/window aggregations on timeseries data

Question

Basic Question

I have a dataset with ~10 billion rows. I\'m looking for the most performant way to calculate rolling/windowed aggregates/metrics (sum, mean, min, ma