import os
import types
import numpy as np
import xarray as xr
import pandas as pd
import fsspec
import aiohttp
import shutil
import pickle # nosec B403 only used with internal files/assets
import json
import tempfile
import warnings
import logging
from packaging import version
import concurrent.futures
import multiprocessing
from ..options import OPTIONS
from ..errors import FileSystemHasNoCache, CacheFileNotFound, DataNotFound, \
InvalidMethod
from abc import ABC, abstractmethod
from ..utilities import Registry
log = logging.getLogger("argopy.stores")
try:
from tqdm import tqdm
except ModuleNotFoundError:
log.debug("argopy needs tqdm installed to display progress bars")
def tqdm(fct, **kw):
return fct
try:
import distributed
has_distributed = True
except ModuleNotFoundError:
log.debug("argopy needs distributed to use Dask cluster/client")
has_distributed = False
def new_fs(protocol: str = '', cache: bool = False, cachedir: str = OPTIONS['cachedir'], **kwargs):
""" Create a new fsspec file system
Parameters
----------
protocol: str (optional)
cache: bool (optional)
Use a filecache system on top of the protocol. Default: False
cachedir: str
Define path to cache directory.
**kwargs: (optional)
Other arguments passed to :class:`fsspec.filesystem`
"""
default_filesystem_kwargs = {'simple_links': True, "block_size": 0}
if protocol == 'http':
default_filesystem_kwargs = {**default_filesystem_kwargs,
**{"client_kwargs": {"trust_env": OPTIONS['trust_env']}}}
elif protocol == 'ftp':
default_filesystem_kwargs = {**default_filesystem_kwargs,
**{"block_size": 1000 * (2 ** 20)}}
filesystem_kwargs = {**default_filesystem_kwargs, **kwargs}
if not cache:
fs = fsspec.filesystem(protocol, **filesystem_kwargs)
cache_registry = None
log_msg = "Opening a fsspec [file] system for '%s' protocol with options: %s" % \
(protocol, str(filesystem_kwargs))
else:
# https://filesystem-spec.readthedocs.io/en/latest/_modules/fsspec/implementations/cached.html#WholeFileCacheFileSystem
fs = fsspec.filesystem("filecache",
target_protocol=protocol,
target_options={**filesystem_kwargs},
cache_storage=cachedir,
expiry_time=86400, cache_check=10)
# We use a refresh rate for cache of 1 day,
# since this is the update frequency of the Ifremer erddap
cache_registry = Registry(name='Cache') # Will hold uri cached by this store instance
log_msg = "Opening a fsspec [filecache, storage='%s'] system for '%s' protocol with options: %s" % \
(cachedir, protocol, str(filesystem_kwargs))
if protocol == 'file' and os.path.sep != fs.sep:
# For some reasons (see https://github.com/fsspec/filesystem_spec/issues/937), the property fs.sep is
# not '\' under Windows. So, using this dirty fix to overwrite it:
fs.sep = os.path.sep
# fsspec folks recommend to use posix internally. But I don't see how to handle this. So keeping this fix
# because it solves issues with failing tests under Windows. Enough at this time.
#todo: Revisit this choice in a while
# log_msg = "%s\n[sys sep=%s] vs [fs sep=%s]" % (log_msg, os.path.sep, fs.sep)
# log.warning(log_msg)
log.debug(log_msg)
return fs, cache_registry
class argo_store_proto(ABC):
""" Argo Abstract File System
Provide a prototype for Argo file systems
Should this class inherits from fsspec.spec.AbstractFileSystem ?
"""
protocol = ''
"""str: File system name, one in fsspec.registry.known_implementations"""
def __init__(self,
cache: bool = False,
cachedir: str = "",
**kwargs):
""" Create a file storage system for Argo data
Parameters
----------
cache: bool (False)
cachedir: str (from OPTIONS)
**kwargs: (optional)
Other arguments passed to :class:`fsspec.filesystem`
"""
self.cache = cache
self.cachedir = OPTIONS['cachedir'] if cachedir == '' else cachedir
self._filesystem_kwargs = {**kwargs}
self.fs, self.cache_registry = new_fs(self.protocol,
self.cache,
self.cachedir,
**self._filesystem_kwargs)
def open(self, path, *args, **kwargs):
self.register(path)
# log.debug("Opening path: %s" % path)
return self.fs.open(path, *args, **kwargs)
def glob(self, path, **kwargs):
return self.fs.glob(path, **kwargs)
def exists(self, path, *args):
return self.fs.exists(path, *args)
def expand_path(self, path):
if self.protocol != "http" and self.protocol != "https":
return self.fs.expand_path(path)
else:
return [path]
def store_path(self, uri):
path = uri
path = self.expand_path(path)[0]
if not path.startswith(self.fs.target_protocol) and version.parse(fsspec.__version__) <= version.parse("0.8.3"):
path = self.fs.target_protocol + "://" + path
return path
def register(self, uri):
""" Keep track of files open with this instance """
if self.cache:
path = self.store_path(uri)
if path not in self.cache_registry:
self.cache_registry.commit(path)
def cachepath(self, uri: str, errors: str = 'raise'):
""" Return path to cached file for a given URI """
if not self.cache:
if errors == 'raise':
raise FileSystemHasNoCache("%s has no cache system" % type(self.fs))
elif uri is not None:
store_path = self.store_path(uri)
self.fs.load_cache() # Read set of stored blocks from file and populate self.fs.cached_files
if store_path in self.fs.cached_files[-1]:
return os.path.sep.join([self.cachedir, self.fs.cached_files[-1][store_path]['fn']])
elif errors == 'raise':
raise CacheFileNotFound("No cached file found in %s for: \n%s" % (self.fs.storage[-1], uri))
else:
raise CacheFileNotFound("No cached file found in %s for: \n%s" % (self.fs.storage[-1], uri))
def _clear_cache_item(self, uri):
""" Open fsspec cache registry (pickle file) and remove entry for uri
"""
# See the "save_cache()" method in:
# https://filesystem-spec.readthedocs.io/en/latest/_modules/fsspec/implementations/cached.html#WholeFileCacheFileSystem
fn = os.path.join(self.fs.storage[-1], "cache")
self.fs.load_cache() # Read set of stored blocks from file and populate self.fs.cached_files
cache = self.fs.cached_files[-1]
if os.path.exists(fn):
with open(fn, "rb") as f:
cached_files = pickle.load(f) # nosec B301 because files controlled internally
else:
cached_files = cache
cache = {}
for k, v in cached_files.items():
if k != uri:
cache[k] = v.copy()
else:
os.remove(os.path.join(self.fs.storage[-1], v['fn']))
# log.debug("Removed %s -> %s" % (uri, v['fn']))
with tempfile.NamedTemporaryFile(mode="wb", delete=False) as f:
pickle.dump(cache, f)
shutil.move(f.name, fn)
def clear_cache(self):
""" Remove cache files and entry from uri open with this store instance """
if self.cache:
for uri in self.cache_registry:
# log.debug("Removing from cache %s" % uri)
self._clear_cache_item(uri)
self.cache_registry.clear() # Reset registry
@abstractmethod
def open_dataset(self, *args, **kwargs):
raise NotImplementedError("Not implemented")
@abstractmethod
def read_csv(self):
raise NotImplementedError("Not implemented")
[docs]class filestore(argo_store_proto):
"""Argo local file system
Relies on:
https://filesystem-spec.readthedocs.io/en/latest/api.html#fsspec.implementations.local.LocalFileSystem
"""
protocol = 'file'
[docs] def open_dataset(self, path, *args, **kwargs):
"""Return a xarray.dataset from a path.
Parameters
----------
path: str
Path to resources passed to xarray.open_dataset
*args, **kwargs:
Other arguments are passed to :func:`xarray.open_dataset`
Returns
-------
:class:`xarray.DataSet`
"""
with self.open(path) as of:
# log.debug("Opening dataset: '%s'" % path) # Redundant with fsspec logger
ds = xr.open_dataset(of, *args, **kwargs)
ds.load()
if "source" not in ds.encoding:
if isinstance(path, str):
ds.encoding["source"] = path
return ds.copy()
def _mfprocessor(self, url, preprocess=None, *args, **kwargs):
# Load data
ds = self.open_dataset(url, *args, **kwargs)
# Pre-process
if isinstance(preprocess, types.FunctionType) or isinstance(preprocess, types.MethodType):
ds = preprocess(ds)
return ds
[docs] def open_mfdataset(self, # noqa: C901
urls,
concat_dim='row',
max_workers: int = 112,
method: str = 'thread',
progress: bool = False,
concat: bool = True,
preprocess=None,
errors: str = 'ignore',
*args, **kwargs):
""" Open multiple urls as a single xarray dataset.
This is a version of the ``open_dataset`` method that is able to handle a list of urls/paths
sequentially or in parallel.
Use a Threads Pool by default for parallelization.
Parameters
----------
urls: list(str)
List of url/path to open
concat_dim: str
Name of the dimension to use to concatenate all datasets (passed to :class:`xarray.concat`)
max_workers: int
Maximum number of threads or processes
method: str
The parallelization method to execute calls asynchronously:
- ``thread`` (Default): use a pool of at most ``max_workers`` threads
- ``process``: use a pool of at most ``max_workers`` processes
- (XFAIL) a :class:`distributed.client.Client` object (:class:`distributed.client.Client`)
Use 'seq' to simply open data sequentially
progress: bool
Display a progress bar (True by default)
preprocess: callable (optional)
If provided, call this function on each dataset prior to concatenation
errors: str
Should it 'raise' or 'ignore' errors. Default: 'ignore'
Returns
-------
:class:`xarray.Dataset`
"""
if not isinstance(urls, list):
urls = [urls]
results = []
if method in ['thread', 'process']:
if method == 'thread':
ConcurrentExecutor = concurrent.futures.ThreadPoolExecutor(max_workers=max_workers)
else:
if max_workers == 112:
max_workers = multiprocessing.cpu_count()
ConcurrentExecutor = concurrent.futures.ProcessPoolExecutor(max_workers=max_workers)
with ConcurrentExecutor as executor:
future_to_url = {executor.submit(self._mfprocessor, url,
preprocess=preprocess, *args, **kwargs): url
for url in urls}
futures = concurrent.futures.as_completed(future_to_url)
if progress:
futures = tqdm(futures, total=len(urls))
for future in futures:
data = None
# url = future_to_url[future]
try:
data = future.result()
except Exception as e:
if errors == 'ignore':
log.debug(
"Ignored error with this file: %s\nException raised: %s"
% (future_to_url[future], str(e.args)))
pass
else:
raise
finally:
results.append(data)
elif has_distributed and isinstance(method, distributed.client.Client):
# Use a dask client:
if progress:
from dask.diagnostics import ProgressBar
with ProgressBar():
futures = method.map(self._mfprocessor, urls, preprocess=preprocess, *args, **kwargs)
results = method.gather(futures)
else:
futures = method.map(self._mfprocessor, urls, preprocess=preprocess, *args, **kwargs)
results = method.gather(futures)
elif method in ['seq', 'sequential']:
if progress:
urls = tqdm(urls, total=len(urls))
for url in urls:
data = None
try:
data = self._mfprocessor(url, preprocess=preprocess, *args, **kwargs)
except Exception as e:
if errors == 'ignore':
log.debug(
"Ignored error with this url: %s\nException raised: %s" % (url, str(e.args)))
pass
else:
raise
finally:
results.append(data)
else:
raise InvalidMethod(method)
# Post-process results
results = [r for r in results if r is not None] # Only keep non-empty results
if len(results) > 0:
if concat:
# ds = xr.concat(results, dim=concat_dim, data_vars='all', coords='all', compat='override')
ds = xr.concat(results, dim=concat_dim, data_vars='minimal', coords='minimal', compat='override')
return ds
else:
return results
else:
raise DataNotFound(urls)
[docs] def read_csv(self, url, **kwargs):
""" Return a pandas.dataframe from an url that is a csv resource
Parameters
----------
Path: str
Path to csv resources passed to pandas.read_csv
Returns
-------
:class:`pandas.DataFrame`
"""
log.debug("Reading csv: %s" % url)
with self.open(url) as of:
df = pd.read_csv(of, **kwargs)
return df
[docs]class httpstore(argo_store_proto):
"""Argo http file system
Relies on:
https://filesystem-spec.readthedocs.io/en/latest/api.html#fsspec.implementations.http.HTTPFileSystem
This store intends to make argopy: safer to failures from http requests and to provide higher levels methods to
work with our datasets
This store is primarily used by the Erddap/Argovis data/index fetchers
"""
protocol = "http"
[docs] def open_dataset(self, url, *args, **kwargs):
""" Open and decode a xarray dataset from an url
Parameters
----------
url: str
Returns
-------
:class:`xarray.Dataset`
"""
try:
data = self.fs.cat_file(url)
except aiohttp.ClientResponseError as e:
if e.status == 413:
warnings.warn("Server says payload Too Large ! Try to use 'parallel=True' or a smaller "
"chunk parameter with your fetcher")
log.debug("Error %i (Payload Too Large) raised with %s" % (e.status, url))
raise
ds = xr.open_dataset(data, *args, **kwargs)
if "source" not in ds.encoding:
if isinstance(url, str):
ds.encoding["source"] = url
self.register(url)
return ds
def _mfprocessor_dataset(self, url, preprocess=None, preprocess_opts={}, *args, **kwargs):
# Load data
ds = self.open_dataset(url, *args, **kwargs)
# Pre-process
if isinstance(preprocess, types.FunctionType) or isinstance(preprocess, types.MethodType):
ds = preprocess(ds, **preprocess_opts)
return ds
[docs] def open_mfdataset(self, # noqa: C901
urls,
max_workers: int = 112,
method: str = 'thread',
progress: bool = False,
concat: bool = True,
concat_dim='row',
preprocess=None,
preprocess_opts={},
errors: str = 'ignore',
*args, **kwargs):
""" Open multiple urls as a single xarray dataset.
This is a version of the :class:`argopy.stores.httpstore.open_dataset` method that is able to
handle a list of urls/paths sequentially or in parallel.
Use a Threads Pool by default for parallelization.
Parameters
----------
urls: list(str)
List of url/path to open
max_workers: int, default: 112
Maximum number of threads or processes
method: str, default: ``thread``
The parallelization method to execute calls asynchronously:
- ``thread`` (default): use a pool of at most ``max_workers`` threads
- ``process``: use a pool of at most ``max_workers`` processes
- :class:`distributed.client.Client`: Experimental, expect this method to fail !
- ``seq``: open data sequentially, no parallelization applied
progress: bool, default: False
Display a progress bar
concat: bool, default: True
Concatenate results in a single :class:`xarray.Dataset` or not (in this case, function will return a
list of :class:`xarray.Dataset`)
concat_dim: str, default: ``row``
Name of the dimension to use to concatenate all datasets (passed to :class:`xarray.concat`)
preprocess: callable (optional)
If provided, call this function on each dataset prior to concatenation
preprocess_opts: dict (optional)
If ``preprocess`` is provided, pass this as options
errors: str, default: ``ignore``
Define how to handle errors raised during data URIs fetching:
- ``raise``: Raise any error encountered
- ``ignore`` (default): Do not stop processing, simply issue a debug message in logging console
- ``silent``: Do not stop processing and do not issue log message
Other args and kwargs: other options passed to :class:`argopy.stores.httpstore.open_dataset`.
Returns
-------
output: :class:`xarray.Dataset` or list of :class:`xarray.Dataset`
"""
strUrl = lambda x: x.replace("https://", "").replace("http://", "") # noqa: E731
def drop_variables_not_in_all_datasets(ds_collection):
"""Drop variables that are not in all datasets (Lowest common denominator)"""
# List all possible data variables:
vlist = []
for res in ds_collection:
[vlist.append(v) for v in list(res.data_vars)]
vlist = np.unique(vlist)
# Check if all possible variables are in each datasets:
ishere = np.zeros((len(vlist), len(ds_collection)))
for ir, res in enumerate(ds_collection):
for iv, v in enumerate(res.data_vars):
for iu, u in enumerate(vlist):
if v == u:
ishere[iu, ir] = 1
# List of dataset with missing variables:
ir_missing = np.sum(ishere, axis=0) < len(vlist)
# List of variables missing in some dataset:
iv_missing = np.sum(ishere, axis=1) < len(ds_collection)
# List of variables to keep
iv_tokeep = np.sum(ishere, axis=1) == len(ds_collection)
for ir, res in enumerate(ds_collection):
# print("\n", res.attrs['Fetched_uri'])
v_to_drop = []
for iv, v in enumerate(res.data_vars):
if v not in vlist[iv_tokeep]:
v_to_drop.append(v)
ds_collection[ir] = ds_collection[ir].drop_vars(v_to_drop)
return ds_collection
if not isinstance(urls, list):
urls = [urls]
results = []
failed = []
if method in ['thread', 'process']:
if method == 'thread':
ConcurrentExecutor = concurrent.futures.ThreadPoolExecutor(max_workers=max_workers)
else:
if max_workers == 112:
max_workers = multiprocessing.cpu_count()
ConcurrentExecutor = concurrent.futures.ProcessPoolExecutor(max_workers=max_workers)
with ConcurrentExecutor as executor:
future_to_url = {executor.submit(self._mfprocessor_dataset,
url,
preprocess=preprocess,
preprocess_opts=preprocess_opts, *args, **kwargs): url
for url in urls}
futures = concurrent.futures.as_completed(future_to_url)
if progress:
futures = tqdm(futures, total=len(urls))
for future in futures:
data = None
try:
data = future.result()
except Exception:
failed.append(future_to_url[future])
if errors == 'ignore':
log.debug("Ignored error with this url: %s" % strUrl(future_to_url[future]))
# See fsspec.http logger for more
pass
elif errors == 'silent':
pass
else:
raise
finally:
results.append(data)
elif has_distributed and isinstance(method, distributed.client.Client):
# Use a dask client:
if progress:
from dask.diagnostics import ProgressBar
with ProgressBar():
futures = method.map(self._mfprocessor_dataset,
urls,
preprocess=preprocess, *args, **kwargs)
results = method.gather(futures)
else:
futures = method.map(self._mfprocessor_dataset,
urls,
preprocess=preprocess, *args, **kwargs)
results = method.gather(futures)
elif method in ['seq', 'sequential']:
if progress:
urls = tqdm(urls, total=len(urls))
for url in urls:
data = None
try:
data = self._mfprocessor_dataset(url,
preprocess=preprocess,
preprocess_opts=preprocess_opts, *args, **kwargs)
except Exception:
failed.append(url)
if errors == 'ignore':
log.debug("Ignored error with this url: %s" % strUrl(url)) # See fsspec.http logger for more
pass
elif errors == 'silent':
pass
else:
raise
finally:
results.append(data)
else:
raise InvalidMethod(method)
# Post-process results
results = [r for r in results if r is not None] # Only keep non-empty results
if len(results) > 0:
if concat:
# ds = xr.concat(results, dim=concat_dim, data_vars='all', coords='all', compat='override')
results = drop_variables_not_in_all_datasets(results)
ds = xr.concat(results,
dim=concat_dim,
data_vars='minimal',
coords='minimal',
compat='override')
return ds
else:
return results
elif len(failed) == len(urls):
raise ValueError("Errors happened with all URLs, this could be due to an internal impossibility to read returned content.")
else:
raise DataNotFound(urls)
[docs] def read_csv(self, url, **kwargs):
""" Read a comma-separated values (csv) url into Pandas DataFrame.
Parameters
----------
url: str
**kwargs: Arguments passed to :class:`pandas.read_csv`
Returns
-------
:class:`pandas.DataFrame`
"""
log.debug("Opening/reading csv from: %s" % url)
with self.open(url) as of:
df = pd.read_csv(of, **kwargs)
return df
[docs] def open_json(self, url, **kwargs):
""" Return a json from an url, or verbose errors
Parameters
----------
url: str
Returns
-------
json
"""
log.debug("Opening json from: %s" % url)
# try:
# with self.open(url) as of:
# js = json.load(of, **kwargs)
# return js
# except ClientResponseError:
# raise
# except json.JSONDecodeError:
# raise
data = self.fs.cat_file(url)
js = json.loads(data, **kwargs)
if len(js) == 0:
js = None
self.register(url)
return js
def _mfprocessor_json(self, url, preprocess=None, url_follow=False, *args, **kwargs):
# Load data
data = self.open_json(url, **kwargs)
# Pre-process
if data is None:
raise DataNotFound(url)
elif isinstance(preprocess, types.FunctionType) or isinstance(preprocess, types.MethodType):
if url_follow:
data = preprocess(data, url=url, **kwargs)
else:
data = preprocess(data)
return data
[docs] def open_mfjson(self, # noqa: C901
urls,
max_workers=112,
method: str = 'thread',
progress: bool = False,
preprocess=None,
url_follow=False,
errors: str = 'ignore',
*args, **kwargs):
""" Open multiple json urls
This is a parallelized version of ``open_json``.
Use a Threads Pool by default for parallelization.
Parameters
----------
urls: list(str)
max_workers: int
Maximum number of threads or processes.
method:
The parallelization method to execute calls asynchronously:
- 'thread' (Default): use a pool of at most ``max_workers`` threads
- 'process': use a pool of at most ``max_workers`` processes
- (XFAIL) Dask client object: use a Dask distributed client object
Use 'seq' to simply open data sequentially
progress: bool
Display a progress bar (True by default, not for dask client method)
preprocess: (callable, optional)
If provided, call this function on each json set
url_follow: bool, False
Follow the URL to the preprocess method as ``url`` argument.
Returns
-------
list()
"""
strUrl = lambda x: x.replace("https://", "").replace("http://", "") # noqa: E731
if not isinstance(urls, list):
urls = [urls]
results = []
failed = []
if method in ['thread', 'process']:
if method == 'thread':
ConcurrentExecutor = concurrent.futures.ThreadPoolExecutor(max_workers=max_workers)
else:
if max_workers == 112:
max_workers = multiprocessing.cpu_count()
ConcurrentExecutor = concurrent.futures.ProcessPoolExecutor(max_workers=max_workers)
with ConcurrentExecutor as executor:
future_to_url = {executor.submit(self._mfprocessor_json, url,
preprocess=preprocess, url_follow=url_follow, *args, **kwargs): url for url in urls}
futures = concurrent.futures.as_completed(future_to_url)
if progress:
futures = tqdm(futures, total=len(urls))
for future in futures:
data = None
try:
data = future.result()
except Exception:
failed.append(future_to_url[future])
if errors == 'ignore':
log.debug("Ignored error with this url: %s" % strUrl(future_to_url[future]))
# See fsspec.http logger for more
pass
elif errors == 'silent':
pass
else:
raise
finally:
results.append(data)
# elif type(method) == distributed.client.Client:
# # Use a dask client:
# futures = method.map(self._mfprocessor_json, urls, preprocess=preprocess, *args, **kwargs)
# results = method.gather(futures)
elif method in ['seq', 'sequential']:
if progress:
urls = tqdm(urls, total=len(urls))
for url in urls:
data = None
try:
data = self._mfprocessor_json(url, preprocess=preprocess, url_follow=url_follow, *args, **kwargs)
except Exception:
failed.append(url)
if errors == 'ignore':
log.debug("Ignored error with this url: %s" % strUrl(url))
# See fsspec.http logger for more
pass
elif errors == 'silent':
pass
else:
raise
finally:
results.append(data)
else:
raise InvalidMethod(method)
# Post-process results
results = [r for r in results if r is not None] # Only keep non-empty results
if len(results) > 0:
return results
else:
raise DataNotFound(urls)
[docs]class memorystore(filestore):
""" Argo in-memory file system (global)
Note that this inherits from filestore, not argo_store_proto
Relies on:
https://filesystem-spec.readthedocs.io/en/latest/api.html#fsspec.implementations.memory.MemoryFileSystem
"""
protocol = 'memory'
[docs] def exists(self, path, *args):
""" Check if path can be open or not
Special handling for memory store
The fsspec.exists() will return False even if the path is in cache.
Here we bypass this in order to return True if the path is in cache
This assumes that the goal of fs.exists is to determine if we can load the path or not
If the path is in cache, it can be loaded
"""
guess = self.fs.exists(path, *args)
if not guess:
try:
self.cachepath(path)
return True
except CacheFileNotFound:
pass
except FileSystemHasNoCache:
pass
return guess
[docs]class ftpstore(httpstore):
""" Argo ftp file system
Relies on:
https://filesystem-spec.readthedocs.io/en/latest/api.html#fsspec.implementations.ftp.FTPFileSystem
"""
protocol = 'ftp'
[docs] def open_dataset(self, url, *args, **kwargs):
""" Open and decode a xarray dataset from an ftp url
Parameters
----------
url: str
Returns
-------
:class:`xarray.Dataset`
"""
try:
this_url = self.fs._strip_protocol(url)
data = self.fs.cat_file(this_url)
except Exception:
log.debug("Error with: %s" % url)
# except aiohttp.ClientResponseError as e:
raise
ds = xr.open_dataset(data, *args, **kwargs)
if "source" not in ds.encoding:
if isinstance(url, str):
ds.encoding["source"] = url
self.register(this_url)
self.register(url)
return ds
def _mfprocessor_dataset(self, url, preprocess=None, preprocess_opts={}, *args, **kwargs):
# Load data
ds = self.open_dataset(url, *args, **kwargs)
# Pre-process
if isinstance(preprocess, types.FunctionType) or isinstance(preprocess, types.MethodType):
ds = preprocess(ds, **preprocess_opts)
return ds
[docs] def open_mfdataset(self, # noqa: C901
urls,
max_workers: int = 112,
method: str = 'thread',
progress: bool = False,
concat: bool = True,
concat_dim='row',
preprocess=None,
preprocess_opts={},
errors: str = 'ignore',
*args, **kwargs):
""" Open multiple ftp urls as a single xarray dataset.
This is a version of the :class:`argopy.stores.ftpstore.open_dataset` method that is able
to handle a list of urls/paths sequentially or in parallel.
Use a Threads Pool by default for parallelization.
Parameters
----------
urls: list(str)
List of url/path to open
max_workers: int, default: 112
Maximum number of threads or processes
method: str, default: ``thread``
The parallelization method to execute calls asynchronously:
- ``thread`` (default): use a pool of at most ``max_workers`` threads
- ``process``: use a pool of at most ``max_workers`` processes
- :class:`distributed.client.Client`: Experimental, expect this method to fail !
- ``seq``: open data sequentially, no parallelization applied
progress: bool, default: False
Display a progress bar
concat: bool, default: True
Concatenate results in a single :class:`xarray.Dataset` or not (in this case, function will return a
list of :class:`xarray.Dataset`)
concat_dim: str, default: ``row``
Name of the dimension to use to concatenate all datasets (passed to :class:`xarray.concat`)
preprocess: callable (optional)
If provided, call this function on each dataset prior to concatenation
preprocess_opts: dict (optional)
If ``preprocess`` is provided, pass this as options
errors: str, default: ``ignore``
Define how to handle errors raised during data URIs fetching:
- ``raise``: Raise any error encountered
- ``ignore`` (default): Do not stop processing, simply issue a debug message in logging console
- ``silent``: Do not stop processing and do not issue log message
Other args and kwargs: other options passed to :class:`argopy.stores.httpstore.open_dataset`.
Returns
-------
output: :class:`xarray.Dataset` or list of :class:`xarray.Dataset`
"""
strUrl = lambda x: x.replace("ftps://", "").replace("ftp://", "") # noqa: E731
def drop_variables_not_in_all_datasets(ds_collection):
"""Drop variables that are not in all datasets (Lowest common denominator)"""
# List all possible data variables:
vlist = []
for res in ds_collection:
[vlist.append(v) for v in list(res.data_vars)]
vlist = np.unique(vlist)
# Check if all possible variables are in each datasets:
ishere = np.zeros((len(vlist), len(ds_collection)))
for ir, res in enumerate(ds_collection):
for iv, v in enumerate(res.data_vars):
for iu, u in enumerate(vlist):
if v == u:
ishere[iu, ir] = 1
# List of dataset with missing variables:
ir_missing = np.sum(ishere, axis=0) < len(vlist)
# List of variables missing in some dataset:
iv_missing = np.sum(ishere, axis=1) < len(ds_collection)
# List of variables to keep
iv_tokeep = np.sum(ishere, axis=1) == len(ds_collection)
for ir, res in enumerate(ds_collection):
# print("\n", res.attrs['Fetched_uri'])
v_to_drop = []
for iv, v in enumerate(res.data_vars):
if v not in vlist[iv_tokeep]:
v_to_drop.append(v)
ds_collection[ir] = ds_collection[ir].drop_vars(v_to_drop)
return ds_collection
if not isinstance(urls, list):
urls = [urls]
results = []
failed = []
if method in ['thread', 'process']:
if method == 'thread':
ConcurrentExecutor = concurrent.futures.ThreadPoolExecutor(max_workers=max_workers)
else:
if max_workers == 112:
max_workers = multiprocessing.cpu_count()
ConcurrentExecutor = concurrent.futures.ProcessPoolExecutor(max_workers=max_workers)
with ConcurrentExecutor as executor:
future_to_url = {executor.submit(self._mfprocessor_dataset, url,
preprocess=preprocess,
preprocess_opts=preprocess_opts, *args, **kwargs): url for url in urls}
futures = concurrent.futures.as_completed(future_to_url)
if progress:
futures = tqdm(futures, total=len(urls))
for future in futures:
data = None
try:
data = future.result()
except Exception:
failed.append(future_to_url[future])
if errors == 'ignore':
log.debug("Ignored error with this url: %s" % strUrl(future_to_url[future]))
# See fsspec.http logger for more
pass
elif errors == 'silent':
pass
else:
raise
finally:
results.append(data)
elif has_distributed and isinstance(method, distributed.client.Client):
# Use a dask client:
if progress:
from dask.diagnostics import ProgressBar
with ProgressBar():
futures = method.map(self._mfprocessor_dataset,
urls,
preprocess=preprocess,
*args, **kwargs)
results = method.gather(futures)
else:
futures = method.map(self._mfprocessor_dataset,
urls,
preprocess=preprocess,
*args, **kwargs)
results = method.gather(futures)
elif method in ['seq', 'sequential']:
if progress:
urls = tqdm(urls, total=len(urls))
for url in urls:
data = None
try:
data = self._mfprocessor_dataset(url,
preprocess=preprocess,
preprocess_opts=preprocess_opts,
*args, **kwargs)
except Exception:
failed.append(url)
if errors == 'ignore':
log.debug("Ignored error with this url: %s" % strUrl(url)) # See fsspec.http logger for more
pass
elif errors == 'silent':
pass
else:
raise
finally:
results.append(data)
else:
raise InvalidMethod(method)
# Post-process results
results = [r for r in results if r is not None] # Only keep non-empty results
if len(results) > 0:
if concat:
# ds = xr.concat(results, dim=concat_dim, data_vars='all', coords='all', compat='override')
results = drop_variables_not_in_all_datasets(results)
ds = xr.concat(results,
dim=concat_dim,
data_vars='minimal',
coords='minimal',
compat='override')
return ds
else:
return results
else:
raise DataNotFound(urls)
