dask.dataframe.read_parquet accepts a list of paths as input. It uses natural_sort_key under the hood to sort the paths in several places. In my case parquet files are not in the natural order and I'd like to sort them outside and keep the same order inside the read_parquet, so I do monkey patching like this to make it work:

with (mock.patch.object(dask.dataframe.io.parquet.core, 'natural_sort_key', new=lambda p: p),
      mock.patch.object(dask.dataframe.io.parquet.utils, 'natural_sort_key', new=lambda p: p)):    
    return dd.read_parquet(paths, index='index', calculate_divisions=True)

It would be great if there was a parameter to control the sort key (disable natural_sort_key or provide my own).