ValueError: unrecognized chunk manager dask

[forestbat]

I have a list which contains paths of netcdf files：

['era5_xaj/era5_datas_201801.nc', 'era5_xaj/era5_datas_201802.nc', 'era5_xaj/era5_datas_201803.nc', ……]

and I use xr.open_mfdataset() to open them as a dataset:
total_era5_ds = xr.open_mfdataset(abs_era5_files, engine='netcdf4', combine='nested')

However it crashed with error:

test\test_read_rain_stas.py:209 (test_localize_by_era)
def test_localize_by_era():
        era_path = os.path.join(definitions.ROOT_DIR, 'example/era5_xaj/')
        abs_era5_files = []
        for dir_name, sub_dir, files in os.walk(era_path):
            for file in files:
                era5_file = os.path.join(era_path, file)
                abs_era5_files.append(era5_file)
>       total_era5_ds = xr.open_mfdataset(abs_era5_files, engine='netcdf4', combine='nested')

test_read_rain_stas.py:218: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
..\venv\lib\site-packages\xarray\backends\api.py:1012: in open_mfdataset
    datasets = [open_(p, **open_kwargs) for p in paths]
..\venv\lib\site-packages\xarray\backends\api.py:1012: in <listcomp>
    datasets = [open_(p, **open_kwargs) for p in paths]
..\venv\lib\site-packages\xarray\backends\api.py:576: in open_dataset
    ds = _dataset_from_backend_dataset(
..\venv\lib\site-packages\xarray\backends\api.py:371: in _dataset_from_backend_dataset
    ds = _chunk_ds(
..\venv\lib\site-packages\xarray\backends\api.py:319: in _chunk_ds
    chunkmanager = guess_chunkmanager(chunked_array_type)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

manager = 'dask'

    def guess_chunkmanager(
        manager: str | ChunkManagerEntrypoint | None,
    ) -> ChunkManagerEntrypoint:
        """
        Get namespace of chunk-handling methods, guessing from what's available.
    
        If the name of a specific ChunkManager is given (e.g. "dask"), then use that.
        Else use whatever is installed, defaulting to dask if there are multiple options.
        """
    
        chunkmanagers = list_chunkmanagers()
    
        if manager is None:
            if len(chunkmanagers) == 1:
                # use the only option available
                manager = next(iter(chunkmanagers.keys()))
            else:
                # default to trying to use dask
                manager = "dask"
    
        if isinstance(manager, str):
            if manager not in chunkmanagers:
>               raise ValueError(
                    f"unrecognized chunk manager {manager} - must be one of: {list(chunkmanagers)}"
                )
E               ValueError: unrecognized chunk manager dask - must be one of: []

..\venv\lib\site-packages\xarray\core\parallelcompat.py:93: ValueError

So where's my problem? How to solve it?

[keewis]

open_mfdataset requires a chunked array library to be installed (we don't support concatenation in our lazy array wrappers). By default, this is dask. If you install that, the call should succeed.

@TomNicholas, should we raise a different error if the number of available chunk managers is 0?

[TomNicholas]

I started a PR to improve this in #7963, but it stalled after we weren't sure exactly what the message should be. To be honest, given there are only 2 known chunkmanagers at the moment, maybe we should just special case this error message.