Why are decoded coordinates dask arrays?

[wachsylon]

Hi all,

let's say I open some CMIP6 data in a way:

urls=['http://aims3.llnl.gov/thredds/dodsC/css03_data/CMIP6/CMIP/NCAR/CESM2/historical/r10i1p1f1/Amon/tas/gn/v20190313/tas_Amon_CESM2_historical_r10i1p1f1_gn_185001-189912.nc',
 'http://aims3.llnl.gov/thredds/dodsC/css03_data/CMIP6/CMIP/NCAR/CESM2/historical/r10i1p1f1/Amon/tas/gn/v20190313/tas_Amon_CESM2_historical_r10i1p1f1_gn_190001-194912.nc']
import xarray as xr
mf_dset = xr.open_mfdataset(urls,
                                decode_cf=True,
                                use_cftime=True,
                                data_vars='minimal', 
                                coords='minimal', 
                                compat='override',
                                combine_attrs="drop_conflicts",
                                **{"decode_coords":"all"})

The resulting mf_dset has coordinates given with dask arrays. They have at least two dimensions. Wouldn't it be good to have them loaded as numpy arrays? When do coordinates come as dask arrays? Is there a rule?
Thanks in advance.

[wachsylon]

This becomes important when using

        mf_dset.to_zarr(store,
                   compute=False,
                   consolidated=True,
                   mode='w')

because dask arrays are not saved in this initialization. However, when writing zarr per region, you do not want to save coordinates as they probably not vary over the chunk dimension. I was not able to configure mf_dset in a way that lat_bnds and lon_bnds were written with this command.