Order of data in variable of a dataset when using open_mfdataset

[PythonSchlumpf]

Dear xarray-users,

I understand now that with xr.open_mfdataset I can open multiple hdf5 files and store the data in a dataset. My hdf5 files don't have labeled dimensions. I get a dataset with

Dimensions:  (phony_dim_0: 21, phony_dim_1: 2167, phony_dim_2: 2070)
Dimensions without coordinates: phony_dim_0, phony_dim_1, phony_dim_2
Data variables:
    data     (phony_dim_0, phony_dim_1, phony_dim_2) uint32 dask.array<chunksize=(10, 2167, 2070), meta=np.ndarray

I used:

ds = xr.open_mfdataset("path/to/data_*.h5", combine='nested', concat_dim='phony_dim_0', group=sub_group)

to receive the output at the top.

When I open the hdf5 files, I have 10,10,1 2D numpy arrays, hence (phony_dim_0: 21). But how can I be sure, the 21 2D arrays are in the right order, please? How does xr.open_mfdataset keeps the correct order when putting all arrays in one big one together, please?

And my second question would be, is there a way to collect the attributes of the different hdf5 wile using open_mfdataset and store them as well with the dataset? I have per hdf5 file one attribute with two values.

Thank you.
PS: Sorry, if these questions are too easy.

[TomNicholas]

How does xr.open_mfdataset keeps the correct order when putting all arrays in one big one together, please?

With combine='nested', open_mfdataset will arrange the files in the order you supply them. Try it with a list of just a couple of files.

And my second question would be, is there a way to collect the attributes of the different hdf5 wile using open_mfdataset and store them as well with the dataset?

Not really - consistent rules for combining attributes from multiple datasets is a difficult problem in general, and at the moment open_mfdataset just lets you choose one "master" file to extract the attributes from.

I have per hdf5 file one attribute with two values.

What I suggest you try is using the preprocess function to promote these attributes to data variables, so that combine_nested within open_mfdataset will concatenate them all together along with the other data in the dataset.

[keewis]

Not really

that's not exactly true, open_mfdataset has a combine_attrs option that allows customizing what it does with the attrs (you can even pass your own function if you're not content with the built-in strategies). It seems I forgot to add that to the docstring in #4971, though.

[PythonSchlumpf]

Keewis and TomNicholas, many thanks. Do you have by any chance some code snippets as an example, please?