[BUG] open_virtual_dataset raises opaque error when called before login

[jjmcnelis]

Is this issue already tracked somewhere, or is this a new report?

• I've reviewed existing issues and couldn't find a duplicate for this problem.

Have you checked the status of Earthdata services?

• I've executed earthaccess.status() and both CMR and EDL returned 'OK'.

Current Behavior

Calling earthaccess.open_virtual_dataset before authenticating with earthaccess.login raises an opaque error. See traceback below.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[4], line 1
----> 1 vds = earthaccess.open_virtual_dataset(results[0], access="indirect", load=False)
      3 vds

File ~/opt/miniforge3/envs/vds312/lib/python3.12/site-packages/earthaccess/dmrpp_zarr.py:240, in open_virtual_dataset(granule, load, group, access)
    198 def open_virtual_dataset(
    199     granule: earthaccess.DataGranule,
    200     load: bool = True,
    201     group: str | None = None,
    202     access: str = "indirect",
    203 ) -> xr.Dataset:
    204     """Open a granule as a single virtual xarray Dataset.
    205 
    206     Uses NASA DMR++ metadata files to create a virtual xarray dataset with ManifestArrays. This virtual dataset can be used to create zarr reference files. See [https://virtualizarr.readthedocs.io](https://virtualizarr.readthedocs.io) for more information on virtual xarray datasets.
   (...)    238         ```
    239     """
--> 240     return open_virtual_mfdataset(
    241         granules=[granule],
    242         load=load,
    243         group=group,
    244         access=access,
    245         parallel=False,
    246         preprocess=None,
    247     )

File ~/opt/miniforge3/envs/vds312/lib/python3.12/site-packages/earthaccess/dmrpp_zarr.py:115, in open_virtual_mfdataset(granules, group, access, preprocess, parallel, load, reference_dir, reference_format, **xr_combine_nested_kwargs)
    112     raise ValueError("No granules provided. At least one granule is required.")
    114 parsed_url = urlparse(granules[0].data_links(access=access)[0])
--> 115 fs = earthaccess.get_fsspec_https_session()
    116 if len(granules):
    117     collection_id = granules[0]["meta"]["collection-concept-id"]

File ~/opt/miniforge3/envs/vds312/lib/python3.12/site-packages/earthaccess/api.py:553, in get_fsspec_https_session()
    537 def get_fsspec_https_session() -> AbstractFileSystem:
    538     """Returns a fsspec session that can be used to access datafiles across many different DAACs.
    539 
    540     Returns:
   (...)    551         ```
    552     """
--> 553     session = earthaccess.__store__.get_fsspec_session()
    554     return session

AttributeError: 'NoneType' object has no attribute 'get_fsspec_session'

The error indicates that earthaccess.__store__ is None. It's set when the user calls the earthaccess.login routine, as demonstrated below.

>>> import earthaccess
>>> earthaccess.__version__
'0.16.0'
>>> earthaccess.__store__ is None
True
>>> earthaccess.login("netrc")
<earthaccess.auth.Auth object at 0x7f7462df4fe0>
>>> earthaccess.__store__.get_fsspec_session
<bound method Store.get_fsspec_session of <earthaccess.store.Store object at 0x7f99442fb500>>

I'm on v0.16.0.

Note that this is in "indirect" access mode. I haven't tested in "direct" access mode.

Expected Behavior

Raise an appropriate exception and remind the user to call earthaccess.login before trying again.

Perhaps that could happen here (called by earthaccess.open_virtual_dataset here).

Sorry in advance if I missed an existing issue about this.

Steps To Reproduce

Import earthaccess and search for any granule. Do not call the login routine. Try open_virtual_dataset in "indirect" access mode. (I haven't tried it in "direct" access mode.)

Environment

- OS: Debian 13, behavior is not os-specific presumably
- Python: 3.12.13

Additional Context

No response

[betolink]

Hi @jjmcnelis thanks for reporting this, This change was documented behavior but I always keep forgetting too. I wonder if we should enforce the login() on search, or before executing I/O operations: open() download() and now virtualize().

BTW open_virtual_dataset, open_virtual_mfdataset and consolidate_metadata are now under virtualize() it should accept the same parameters you're using.