What is the best way to update a zarr archive

[Boorhin]

I am producing some zarr archive of xarray datasets and I would like to compute new variables and add them to the same archive without modifying what is already existing. What is the best practice?

ds=xr.open_zarr('archive.zarr', mode='a')
ds['new_var']=ds['var1']-ds['var2']

=> will the 'new_var' saved automatically or do I have to force the writing of it with to_zarr?
ds.to_zarr('archive.zarr')
Thanks

[shoyer]

Xarray datasets are never directly linked to data on disk. To modify an existing archive, you need to explicitly call to_zarr() or to_netcdf().