import typing import dask import fsspec import pandas as pd import pydantic import xarray as xr from intake.source.base import DataSource, Schema from .cat import Aggregation, DataFormat from .utils import OPTIONS class ESMDataSourceError(Exception): pass def _get_xarray_open_kwargs(data_format, xarray_open_kwargs=None, storage_options=None): xarray_open_kwargs = (xarray_open_kwargs or {}).copy() _default_open_kwargs = { 'engine': 'zarr' if data_format in {'zarr', 'reference'} else 'netcdf4', 'chunks': {}, 'backend_kwargs': {}, } xarray_open_kwargs = ( {**_default_open_kwargs, **xarray_open_kwargs} if xarray_open_kwargs else _default_open_kwargs ) if ( xarray_open_kwargs['engine'] == 'zarr' and 'storage_options' not in xarray_open_kwargs['backend_kwargs'] ): xarray_open_kwargs['backend_kwargs']['storage_options'] = {} or storage_options return xarray_open_kwargs @dask.delayed def _open_dataset( urlpath, varname, *, xarray_open_kwargs=None, preprocess=None, requested_variables=None, additional_attrs=None, expand_dims=None, data_format=None, ): storage_options = xarray_open_kwargs.get('backend_kwargs', {}).get('storage_options', {}) # Support kerchunk datasets, setting the file object (fo) and urlpath if data_format == 'reference': xarray_open_kwargs['backend_kwargs']['storage_options']['fo'] = urlpath xarray_open_kwargs['backend_kwargs']['consolidated'] = False urlpath = 'reference://' if xarray_open_kwargs['engine'] == 'zarr': url = urlpath elif fsspec.utils.can_be_local(urlpath): url = fsspec.open_local(urlpath, **storage_options) else: url = fsspec.open(urlpath, **storage_options).open() # Handle multi-file datasets with `xr.open_mfdataset()` if '*' in url or isinstance(url, list): # How should we handle concat_dim, and other xr.open_mfdataset kwargs? xarray_open_kwargs.update(preprocess=preprocess) xarray_open_kwargs.update(parallel=True) ds = xr.open_mfdataset(url, **xarray_open_kwargs) else: ds = xr.open_dataset(url, **xarray_open_kwargs) if preprocess is not None: ds = preprocess(ds) if varname and isinstance(varname, str): varname = [varname] if requested_variables: if isinstance(requested_variables, str): requested_variables = [requested_variables] variable_intersection = set(requested_variables).intersection(set(varname)) variables = [variable for variable in variable_intersection if variable in ds.data_vars] scalar_variables = [v for v in ds.data_vars if len(ds[v].dims) == 0] ds = ds.set_coords(scalar_variables) ds = ds[variables] ds.attrs[OPTIONS['vars_key']] = variables else: ds.attrs[OPTIONS['vars_key']] = varname ds = _expand_dims(expand_dims, ds) ds = _update_attrs(additional_attrs=additional_attrs, ds=ds) return ds def _update_attrs(*, additional_attrs, ds): additional_attrs = additional_attrs or {} if additional_attrs: additional_attrs = { f"{OPTIONS['attrs_prefix']}:{key}": f'{value}' if isinstance(value, str) or not hasattr(value, '__iter__') else ','.join(value) for key, value in additional_attrs.items() } ds.attrs = {**ds.attrs, **additional_attrs} return ds def _expand_dims(expand_dims, ds): if expand_dims: for variable in ds.attrs[OPTIONS['vars_key']]: ds[variable] = ds[variable].expand_dims(**expand_dims) return ds class ESMDataSource(DataSource): version = '1.0' container = 'xarray' name = 'esm_datasource' partition_access = True @pydantic.validate_arguments def __init__( self, key: pydantic.StrictStr, records: typing.List[typing.Dict[str, typing.Any]], variable_column_name: pydantic.StrictStr, path_column_name: pydantic.StrictStr, data_format: typing.Optional[DataFormat], format_column_name: typing.Optional[pydantic.StrictStr], *, aggregations: typing.Optional[typing.List[Aggregation]] = None, requested_variables: typing.List[str] = None, preprocess: typing.Callable = None, storage_options: typing.Dict[str, typing.Any] = None, xarray_open_kwargs: typing.Dict[str, typing.Any] = None, xarray_combine_by_coords_kwargs: typing.Dict[str, typing.Any] = None, intake_kwargs: typing.Dict[str, typing.Any] = None, ): """An intake compatible Data Source for ESM data. Parameters ---------- key: str The key of the data source. records: list of dict A list of records, each of which is a dictionary mapping column names to values. variable_column_name: str The column name of the variable name. path_column_name: str The column name of the path. data_format: DataFormat The data format of the data. aggregations: list of Aggregation, optional A list of aggregations to apply to the data. requested_variables: list of str, optional A list of variables to load. preprocess: callable, optional A preprocessing function to apply to the data. storage_options: dict, optional fsspec parameters passed to the backend file-system such as Google Cloud Storage, Amazon Web Service S3. xarray_open_kwargs: dict, optional Keyword arguments to pass to :py:func:`~xarray.open_dataset` function. xarray_combine_by_coords_kwargs: dict, optional Keyword arguments to pass to :py:func:`~xarray.combine_by_coords` function. intake_kwargs: dict, optional Additional keyword arguments are passed through to the :py:class:`~intake.source.base.DataSource` base class. """ intake_kwargs = intake_kwargs or {} super().__init__(**intake_kwargs) self.key = key self.storage_options = storage_options or {} self.preprocess = preprocess self.requested_variables = requested_variables or [] self.path_column_name = path_column_name self.variable_column_name = variable_column_name self.aggregations = aggregations self.df = pd.DataFrame.from_records(records) self.xarray_open_kwargs = xarray_open_kwargs self.xarray_combine_by_coords_kwargs = dict(combine_attrs='drop_conflicts') if xarray_combine_by_coords_kwargs is None: xarray_combine_by_coords_kwargs = {} self.xarray_combine_by_coords_kwargs = { **self.xarray_combine_by_coords_kwargs, **xarray_combine_by_coords_kwargs, } self._ds = None if data_format is not None: self.df['_data_format_'] = data_format.value else: self.df = self.df.rename(columns={format_column_name: '_data_format_'}) def __repr__(self) -> str: return f'<{type(self).__name__} (name: {self.key}, asset(s): {len(self.df)})>' def _get_schema(self) -> Schema: if self._ds is None: self._open_dataset() metadata = {'dims': {}, 'data_vars': {}, 'coords': ()} self._schema = Schema( datashape=None, dtype=None, shape=None, npartitions=None, extra_metadata=metadata, ) return self._schema def _open_dataset(self): """Open dataset with xarray""" try: datasets = [ _open_dataset( record[self.path_column_name], record[self.variable_column_name], xarray_open_kwargs=_get_xarray_open_kwargs( record['_data_format_'], self.xarray_open_kwargs, self.storage_options ), preprocess=self.preprocess, expand_dims={ agg.attribute_name: [record[agg.attribute_name]] for agg in self.aggregations if agg.type.value == 'join_new' }, requested_variables=self.requested_variables, data_format=record['_data_format_'], additional_attrs=record[~record.isnull()].to_dict(), ) for _, record in self.df.iterrows() ] datasets = dask.compute(*datasets) if len(datasets) == 1: self._ds = datasets[0] else: datasets = sorted( datasets, key=lambda ds: tuple( f"{OPTIONS['attrs_prefix']}/{agg.attribute_name}" for agg in self.aggregations ), ) with dask.config.set( {'scheduler': 'single-threaded', 'array.slicing.split_large_chunks': True} ): # Use single-threaded scheduler datasets = [ ds.set_coords(set(ds.variables) - set(ds.attrs[OPTIONS['vars_key']])) for ds in datasets ] self._ds = xr.combine_by_coords( datasets, **self.xarray_combine_by_coords_kwargs ) self._ds.attrs[OPTIONS['dataset_key']] = self.key except Exception as exc: raise ESMDataSourceError( f"""Failed to load dataset with key='{self.key}' You can use `cat['{self.key}'].df` to inspect the assets/files for this key. """ ) from exc def to_dask(self): """Return xarray object (which will have chunks)""" self._load_metadata() return self._ds def close(self): """Delete open files from memory""" self._ds = None self._schema = None