Try partial_decompress=True

[JackKelly]

Especially for NWPs, where we often only want a single value per chunk.

Also try in combination with uncompressed chunks

See zarr-developers/zarr-python#667

[JackKelly]

Tried it. Doesn't seem to speed things up much (if at all). But I'll leave it in the code, just in case it speeds things up a bit!

Using consolidated metadata seems to significantly (double?) the read speed though.

Code:

# Pretty sure this sets partial_decompress to True:
import zarr
array = zarr.open_array(
    FILENAME, path='stacked_eumetsat_data', partial_decompress=False, mode='r'
    storage_options=dict(consolidated=True))

%%time
data = array[:12, :128, :128, 0]  # takes about 35 ms with or without partial_decompress.

This code might set partial_decompress=True, but I'm not certain!

ds = xr.open_dataset(
    FILENAME, 
    engine='zarr',
    consolidated=True,
    storage_options=dict(partial_decompress=True)
)