Merge remote branch 'remotes/jwandborg/f587-split_storage_into_submodules'
This commit is contained in:
commit
9122a9d047
@ -43,10 +43,12 @@ allow_attachments = boolean(default=False)
|
||||
|
||||
|
||||
[storage:publicstore]
|
||||
storage_class = string(default="mediagoblin.storage.filestorage:BasicFileStorage")
|
||||
base_dir = string(default="%(here)s/user_dev/media/public")
|
||||
base_url = string(default="/mgoblin_media/")
|
||||
|
||||
[storage:queuestore]
|
||||
storage_class = string(default="mediagoblin.storage.filestorage:BasicFileStorage")
|
||||
base_dir = string(default="%(here)s/user_dev/media/queue")
|
||||
|
||||
|
||||
|
@ -17,7 +17,7 @@
|
||||
from mediagoblin import mg_globals
|
||||
from mediagoblin.db.open import setup_connection_and_db_from_config
|
||||
from mediagoblin.init.config import read_mediagoblin_config
|
||||
from mediagoblin.storage import BasicFileStorage
|
||||
from mediagoblin.storage.filestorage import BasicFileStorage
|
||||
from mediagoblin.init import setup_storage, setup_global_and_app_config
|
||||
|
||||
import shutil
|
||||
|
@ -1,568 +0,0 @@
|
||||
# GNU MediaGoblin -- federated, autonomous media hosting
|
||||
# Copyright (C) 2011 MediaGoblin contributors. See AUTHORS.
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
#
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU Affero General Public License for more details.
|
||||
#
|
||||
# You should have received a copy of the GNU Affero General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
import os
|
||||
import shutil
|
||||
import urlparse
|
||||
import uuid
|
||||
import cloudfiles
|
||||
import mimetypes
|
||||
import tempfile
|
||||
|
||||
from werkzeug.utils import secure_filename
|
||||
|
||||
from mediagoblin import util
|
||||
|
||||
########
|
||||
# Errors
|
||||
########
|
||||
|
||||
|
||||
class Error(Exception):
|
||||
pass
|
||||
|
||||
|
||||
class InvalidFilepath(Error):
|
||||
pass
|
||||
|
||||
|
||||
class NoWebServing(Error):
|
||||
pass
|
||||
|
||||
|
||||
class NotImplementedError(Error):
|
||||
pass
|
||||
|
||||
|
||||
###############################################
|
||||
# Storage interface & basic file implementation
|
||||
###############################################
|
||||
|
||||
class StorageInterface(object):
|
||||
"""
|
||||
Interface for the storage API.
|
||||
|
||||
This interface doesn't actually provide behavior, but it defines
|
||||
what kind of storage patterns subclasses should provide.
|
||||
|
||||
It is important to note that the storage API idea of a "filepath"
|
||||
is actually like ['dir1', 'dir2', 'file.jpg'], so keep that in
|
||||
mind while reading method documentation.
|
||||
|
||||
You should set up your __init__ method with whatever keyword
|
||||
arguments are appropriate to your storage system, but you should
|
||||
also passively accept all extraneous keyword arguments like:
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
pass
|
||||
|
||||
See BasicFileStorage as a simple implementation of the
|
||||
StorageInterface.
|
||||
"""
|
||||
|
||||
# Whether this file store is on the local filesystem.
|
||||
local_storage = False
|
||||
|
||||
def __raise_not_implemented(self):
|
||||
"""
|
||||
Raise a warning about some component not implemented by a
|
||||
subclass of this interface.
|
||||
"""
|
||||
raise NotImplementedError(
|
||||
"This feature not implemented in this storage API implementation.")
|
||||
|
||||
def file_exists(self, filepath):
|
||||
"""
|
||||
Return a boolean asserting whether or not file at filepath
|
||||
exists in our storage system.
|
||||
|
||||
Returns:
|
||||
True / False depending on whether file exists or not.
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def get_file(self, filepath, mode='r'):
|
||||
"""
|
||||
Return a file-like object for reading/writing from this filepath.
|
||||
|
||||
Should create directories, buckets, whatever, as necessary.
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def delete_file(self, filepath):
|
||||
"""
|
||||
Delete or dereference the file at filepath.
|
||||
|
||||
This might need to delete directories, buckets, whatever, for
|
||||
cleanliness. (Be sure to avoid race conditions on that though)
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def file_url(self, filepath):
|
||||
"""
|
||||
Get the URL for this file. This assumes our storage has been
|
||||
mounted with some kind of URL which makes this possible.
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def get_unique_filepath(self, filepath):
|
||||
"""
|
||||
If a filename at filepath already exists, generate a new name.
|
||||
|
||||
Eg, if the filename doesn't exist:
|
||||
>>> storage_handler.get_unique_filename(['dir1', 'dir2', 'fname.jpg'])
|
||||
[u'dir1', u'dir2', u'fname.jpg']
|
||||
|
||||
But if a file does exist, let's get one back with at uuid tacked on:
|
||||
>>> storage_handler.get_unique_filename(['dir1', 'dir2', 'fname.jpg'])
|
||||
[u'dir1', u'dir2', u'd02c3571-dd62-4479-9d62-9e3012dada29-fname.jpg']
|
||||
"""
|
||||
# Make sure we have a clean filepath to start with, since
|
||||
# we'll be possibly tacking on stuff to the filename.
|
||||
filepath = clean_listy_filepath(filepath)
|
||||
|
||||
if self.file_exists(filepath):
|
||||
return filepath[:-1] + ["%s-%s" % (uuid.uuid4(), filepath[-1])]
|
||||
else:
|
||||
return filepath
|
||||
|
||||
def get_local_path(self, filepath):
|
||||
"""
|
||||
If this is a local_storage implementation, give us a link to
|
||||
the local filesystem reference to this file.
|
||||
|
||||
>>> storage_handler.get_local_path(['foo', 'bar', 'baz.jpg'])
|
||||
u'/path/to/mounting/foo/bar/baz.jpg'
|
||||
"""
|
||||
# Subclasses should override this method, if applicable.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def copy_locally(self, filepath, dest_path):
|
||||
"""
|
||||
Copy this file locally.
|
||||
|
||||
A basic working method for this is provided that should
|
||||
function both for local_storage systems and remote storge
|
||||
systems, but if more efficient systems for copying locally
|
||||
apply to your system, override this method with something more
|
||||
appropriate.
|
||||
"""
|
||||
if self.local_storage:
|
||||
shutil.copy(
|
||||
self.get_local_path(filepath), dest_path)
|
||||
else:
|
||||
with self.get_file(filepath, 'rb') as source_file:
|
||||
with file(dest_path, 'wb') as dest_file:
|
||||
dest_file.write(source_file.read())
|
||||
|
||||
|
||||
class BasicFileStorage(StorageInterface):
|
||||
"""
|
||||
Basic local filesystem implementation of storage API
|
||||
"""
|
||||
|
||||
local_storage = True
|
||||
|
||||
def __init__(self, base_dir, base_url=None, **kwargs):
|
||||
"""
|
||||
Keyword arguments:
|
||||
- base_dir: Base directory things will be served out of. MUST
|
||||
be an absolute path.
|
||||
- base_url: URL files will be served from
|
||||
"""
|
||||
self.base_dir = base_dir
|
||||
self.base_url = base_url
|
||||
|
||||
def _resolve_filepath(self, filepath):
|
||||
"""
|
||||
Transform the given filepath into a local filesystem filepath.
|
||||
"""
|
||||
return os.path.join(
|
||||
self.base_dir, *clean_listy_filepath(filepath))
|
||||
|
||||
def file_exists(self, filepath):
|
||||
return os.path.exists(self._resolve_filepath(filepath))
|
||||
|
||||
def get_file(self, filepath, mode='r'):
|
||||
# Make directories if necessary
|
||||
if len(filepath) > 1:
|
||||
directory = self._resolve_filepath(filepath[:-1])
|
||||
if not os.path.exists(directory):
|
||||
os.makedirs(directory)
|
||||
|
||||
# Grab and return the file in the mode specified
|
||||
return open(self._resolve_filepath(filepath), mode)
|
||||
|
||||
def delete_file(self, filepath):
|
||||
# TODO: Also delete unused directories if empty (safely, with
|
||||
# checks to avoid race conditions).
|
||||
os.remove(self._resolve_filepath(filepath))
|
||||
|
||||
def file_url(self, filepath):
|
||||
if not self.base_url:
|
||||
raise NoWebServing(
|
||||
"base_url not set, cannot provide file urls")
|
||||
|
||||
return urlparse.urljoin(
|
||||
self.base_url,
|
||||
'/'.join(clean_listy_filepath(filepath)))
|
||||
|
||||
def get_local_path(self, filepath):
|
||||
return self._resolve_filepath(filepath)
|
||||
|
||||
|
||||
# ----------------------------------------------------
|
||||
# OpenStack/Rackspace Cloud's Swift/CloudFiles support
|
||||
# ----------------------------------------------------
|
||||
|
||||
class CloudFilesStorage(StorageInterface):
|
||||
def __init__(self, **kwargs):
|
||||
self.param_container = kwargs.get('cloudfiles_container')
|
||||
self.param_user = kwargs.get('cloudfiles_user')
|
||||
self.param_api_key = kwargs.get('cloudfiles_api_key')
|
||||
self.param_host = kwargs.get('cloudfiles_host')
|
||||
self.param_use_servicenet = kwargs.get('cloudfiles_use_servicenet')
|
||||
|
||||
if not self.param_host:
|
||||
print('No CloudFiles host URL specified, '
|
||||
'defaulting to Rackspace US')
|
||||
|
||||
self.connection = cloudfiles.get_connection(
|
||||
username=self.param_user,
|
||||
api_key=self.param_api_key,
|
||||
servicenet=True if self.param_use_servicenet == 'true' or \
|
||||
self.param_use_servicenet == True else False)
|
||||
|
||||
if not self.param_container == \
|
||||
self.connection.get_container(self.param_container):
|
||||
self.container = self.connection.create_container(
|
||||
self.param_container)
|
||||
self.container.make_public(
|
||||
ttl=60 * 60 * 2)
|
||||
else:
|
||||
self.container = self.connection.get_container(
|
||||
self.param_container)
|
||||
|
||||
self.container_uri = self.container.public_uri()
|
||||
|
||||
def _resolve_filepath(self, filepath):
|
||||
return '/'.join(
|
||||
clean_listy_filepath(filepath))
|
||||
|
||||
def file_exists(self, filepath):
|
||||
try:
|
||||
object = self.container.get_object(
|
||||
self._resolve_filepath(filepath))
|
||||
return True
|
||||
except cloudfiles.errors.NoSuchObject:
|
||||
return False
|
||||
|
||||
def get_file(self, filepath, *args, **kwargs):
|
||||
"""
|
||||
- Doesn't care about the "mode" argument
|
||||
"""
|
||||
try:
|
||||
obj = self.container.get_object(
|
||||
self._resolve_filepath(filepath))
|
||||
except cloudfiles.errors.NoSuchObject:
|
||||
obj = self.container.create_object(
|
||||
self._resolve_filepath(filepath))
|
||||
|
||||
mimetype = mimetypes.guess_type(
|
||||
filepath[-1])
|
||||
|
||||
if mimetype:
|
||||
obj.content_type = mimetype[0]
|
||||
|
||||
return CloudFilesStorageObjectWrapper(obj, *args, **kwargs)
|
||||
|
||||
def delete_file(self, filepath):
|
||||
# TODO: Also delete unused directories if empty (safely, with
|
||||
# checks to avoid race conditions).
|
||||
self.container.delete_object(
|
||||
self._resolve_filepath(filepath))
|
||||
|
||||
def file_url(self, filepath):
|
||||
return '/'.join([
|
||||
self.container_uri,
|
||||
self._resolve_filepath(filepath)])
|
||||
|
||||
|
||||
class CloudFilesStorageObjectWrapper():
|
||||
"""
|
||||
Wrapper for python-cloudfiles's cloudfiles.storage_object.Object
|
||||
used to circumvent the mystic `medium.jpg` corruption issue, where
|
||||
we had both python-cloudfiles and PIL doing buffering on both
|
||||
ends and that breaking things.
|
||||
|
||||
This wrapper currently meets mediagoblin's needs for a public_store
|
||||
file-like object.
|
||||
"""
|
||||
def __init__(self, storage_object, *args, **kwargs):
|
||||
self.storage_object = storage_object
|
||||
|
||||
def read(self, *args, **kwargs):
|
||||
return self.storage_object.read(*args, **kwargs)
|
||||
|
||||
def write(self, data, *args, **kwargs):
|
||||
"""
|
||||
write data to the cloudfiles storage object
|
||||
|
||||
The original motivation for this wrapper is to ensure
|
||||
that buffered writing to a cloudfiles storage object does not overwrite
|
||||
any preexisting data.
|
||||
|
||||
Currently this method does not support any write modes except "append".
|
||||
However if we should need it it would be easy implement.
|
||||
"""
|
||||
if self.storage_object.size and type(data) == str:
|
||||
data = self.read() + data
|
||||
|
||||
self.storage_object.write(data, *args, **kwargs)
|
||||
|
||||
def close(self):
|
||||
pass
|
||||
|
||||
def __enter__(self):
|
||||
"""
|
||||
Context Manager API implementation
|
||||
http://docs.python.org/library/stdtypes.html#context-manager-types
|
||||
"""
|
||||
return self
|
||||
|
||||
def __exit__(self, *exc_info):
|
||||
"""
|
||||
Context Manger API implementation
|
||||
see self.__enter__()
|
||||
"""
|
||||
self.close()
|
||||
|
||||
|
||||
# ------------
|
||||
# MountStorage
|
||||
# ------------
|
||||
|
||||
class MountStorage(StorageInterface):
|
||||
"""
|
||||
Experimental "Mount" virtual Storage Interface
|
||||
|
||||
This isn't an interface to some real storage, instead it's a
|
||||
redirecting interface, that redirects requests to other
|
||||
"StorageInterface"s.
|
||||
|
||||
For example, say you have the paths:
|
||||
|
||||
1. ['user_data', 'cwebber', 'avatar.jpg']
|
||||
2. ['user_data', 'elrond', 'avatar.jpg']
|
||||
3. ['media_entries', '34352f304c3f4d0ad8ad0f043522b6f2', 'thumb.jpg']
|
||||
|
||||
You could mount media_entries under CloudFileStorage and user_data
|
||||
under BasicFileStorage. Then 1 would be passed to
|
||||
BasicFileStorage under the path ['cwebber', 'avatar.jpg'] and 3
|
||||
would be passed to CloudFileStorage under
|
||||
['34352f304c3f4d0ad8ad0f043522b6f2', 'thumb.jpg'].
|
||||
|
||||
In other words, this is kind of like mounting /home/ and /etc/
|
||||
under different filesystems on your operating system... but with
|
||||
mediagoblin filestorages :)
|
||||
|
||||
To set this up, you currently need to call the mount() method with
|
||||
the target path and a backend, that shall be available under that
|
||||
target path. You have to mount things in a sensible order,
|
||||
especially you can't mount ["a", "b"] before ["a"].
|
||||
"""
|
||||
def __init__(self, **kwargs):
|
||||
self.mounttab = {}
|
||||
|
||||
def mount(self, dirpath, backend):
|
||||
"""
|
||||
Mount a new backend under dirpath
|
||||
"""
|
||||
new_ent = clean_listy_filepath(dirpath)
|
||||
|
||||
print "Mounting:", repr(new_ent)
|
||||
already, rem_1, table, rem_2 = self._resolve_to_backend(new_ent, True)
|
||||
print "===", repr(already), repr(rem_1), repr(rem_2), len(table)
|
||||
|
||||
assert (len(rem_2) > 0) or (None not in table), \
|
||||
"That path is already mounted"
|
||||
assert (len(rem_2) > 0) or (len(table)==0), \
|
||||
"A longer path is already mounted here"
|
||||
|
||||
for part in rem_2:
|
||||
table[part] = {}
|
||||
table = table[part]
|
||||
table[None] = backend
|
||||
|
||||
def _resolve_to_backend(self, filepath, extra_info = False):
|
||||
"""
|
||||
extra_info = True is for internal use!
|
||||
|
||||
Normally, returns the backend and the filepath inside that backend.
|
||||
|
||||
With extra_info = True it returns the last directory node and the
|
||||
remaining filepath from there in addition.
|
||||
"""
|
||||
table = self.mounttab
|
||||
filepath = filepath[:]
|
||||
res_fp = None
|
||||
while True:
|
||||
new_be = table.get(None)
|
||||
if (new_be is not None) or res_fp is None:
|
||||
res_be = new_be
|
||||
res_fp = filepath[:]
|
||||
res_extra = (table, filepath[:])
|
||||
# print "... New res: %r, %r, %r" % (res_be, res_fp, res_extra)
|
||||
if len(filepath) == 0:
|
||||
break
|
||||
query = filepath.pop(0)
|
||||
entry = table.get(query)
|
||||
if entry is not None:
|
||||
table = entry
|
||||
res_extra = (table, filepath[:])
|
||||
else:
|
||||
break
|
||||
if extra_info:
|
||||
return (res_be, res_fp) + res_extra
|
||||
else:
|
||||
return (res_be, res_fp)
|
||||
|
||||
def resolve_to_backend(self, filepath):
|
||||
backend, filepath = self._resolve_to_backend(filepath)
|
||||
if backend is None:
|
||||
raise Error("Path not mounted")
|
||||
return backend, filepath
|
||||
|
||||
def __repr__(self, table = None, indent = []):
|
||||
res = []
|
||||
if table is None:
|
||||
res.append("MountStorage<")
|
||||
table = self.mounttab
|
||||
v = table.get(None)
|
||||
if v:
|
||||
res.append(" " * len(indent) + repr(indent) + ": " + repr(v))
|
||||
for k, v in table.iteritems():
|
||||
if k == None:
|
||||
continue
|
||||
res.append(" " * len(indent) + repr(k) + ":")
|
||||
res += self.__repr__(v, indent + [k])
|
||||
if table is self.mounttab:
|
||||
res.append(">")
|
||||
return "\n".join(res)
|
||||
else:
|
||||
return res
|
||||
|
||||
def file_exists(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.file_exists(filepath)
|
||||
|
||||
def get_file(self, filepath, mode='r'):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.get_file(filepath, mode)
|
||||
|
||||
def delete_file(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.delete_file(filepath)
|
||||
|
||||
def file_url(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.file_url(filepath)
|
||||
|
||||
def get_local_path(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.get_local_path(filepath)
|
||||
|
||||
def copy_locally(self, filepath, dest_path):
|
||||
"""
|
||||
Need to override copy_locally, because the local_storage
|
||||
attribute is not correct.
|
||||
"""
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
backend.copy_locally(filepath, dest_path)
|
||||
|
||||
|
||||
###########
|
||||
# Utilities
|
||||
###########
|
||||
|
||||
def clean_listy_filepath(listy_filepath):
|
||||
"""
|
||||
Take a listy filepath (like ['dir1', 'dir2', 'filename.jpg']) and
|
||||
clean out any nastiness from it.
|
||||
|
||||
|
||||
>>> clean_listy_filepath([u'/dir1/', u'foo/../nasty', u'linooks.jpg'])
|
||||
[u'dir1', u'foo_.._nasty', u'linooks.jpg']
|
||||
|
||||
Args:
|
||||
- listy_filepath: a list of filepath components, mediagoblin
|
||||
storage API style.
|
||||
|
||||
Returns:
|
||||
A cleaned list of unicode objects.
|
||||
"""
|
||||
cleaned_filepath = [
|
||||
unicode(secure_filename(filepath))
|
||||
for filepath in listy_filepath]
|
||||
|
||||
if u'' in cleaned_filepath:
|
||||
raise InvalidFilepath(
|
||||
"A filename component could not be resolved into a usable name.")
|
||||
|
||||
return cleaned_filepath
|
||||
|
||||
|
||||
def storage_system_from_config(config_section):
|
||||
"""
|
||||
Utility for setting up a storage system from a config section.
|
||||
|
||||
Note that a special argument may be passed in to
|
||||
the config_section which is "storage_class" which will provide an
|
||||
import path to a storage system. This defaults to
|
||||
"mediagoblin.storage:BasicFileStorage" if otherwise undefined.
|
||||
|
||||
Arguments:
|
||||
- config_section: dictionary of config parameters
|
||||
|
||||
Returns:
|
||||
An instantiated storage system.
|
||||
|
||||
Example:
|
||||
storage_system_from_config(
|
||||
{'base_url': '/media/',
|
||||
'base_dir': '/var/whatever/media/'})
|
||||
|
||||
Will return:
|
||||
BasicFileStorage(
|
||||
base_url='/media/',
|
||||
base_dir='/var/whatever/media')
|
||||
"""
|
||||
# This construct is needed, because dict(config) does
|
||||
# not replace the variables in the config items.
|
||||
config_params = dict(config_section.iteritems())
|
||||
|
||||
if 'storage_class' in config_params:
|
||||
storage_class = config_params['storage_class']
|
||||
config_params.pop('storage_class')
|
||||
else:
|
||||
storage_class = "mediagoblin.storage:BasicFileStorage"
|
||||
|
||||
storage_class = util.import_component(storage_class)
|
||||
return storage_class(**config_params)
|
240
mediagoblin/storage/__init__.py
Normal file
240
mediagoblin/storage/__init__.py
Normal file
@ -0,0 +1,240 @@
|
||||
# GNU MediaGoblin -- federated, autonomous media hosting
|
||||
# Copyright (C) 2011 MediaGoblin contributors. See AUTHORS.
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
#
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU Affero General Public License for more details.
|
||||
#
|
||||
# You should have received a copy of the GNU Affero General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
import os
|
||||
import shutil
|
||||
import urlparse
|
||||
import uuid
|
||||
|
||||
from werkzeug.utils import secure_filename
|
||||
|
||||
from mediagoblin import util
|
||||
|
||||
########
|
||||
# Errors
|
||||
########
|
||||
|
||||
|
||||
class Error(Exception):
|
||||
pass
|
||||
|
||||
|
||||
class InvalidFilepath(Error):
|
||||
pass
|
||||
|
||||
|
||||
class NoWebServing(Error):
|
||||
pass
|
||||
|
||||
|
||||
class NotImplementedError(Error):
|
||||
pass
|
||||
|
||||
|
||||
###############################################
|
||||
# Storage interface & basic file implementation
|
||||
###############################################
|
||||
|
||||
class StorageInterface(object):
|
||||
"""
|
||||
Interface for the storage API.
|
||||
|
||||
This interface doesn't actually provide behavior, but it defines
|
||||
what kind of storage patterns subclasses should provide.
|
||||
|
||||
It is important to note that the storage API idea of a "filepath"
|
||||
is actually like ['dir1', 'dir2', 'file.jpg'], so keep that in
|
||||
mind while reading method documentation.
|
||||
|
||||
You should set up your __init__ method with whatever keyword
|
||||
arguments are appropriate to your storage system, but you should
|
||||
also passively accept all extraneous keyword arguments like:
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
pass
|
||||
|
||||
See BasicFileStorage as a simple implementation of the
|
||||
StorageInterface.
|
||||
"""
|
||||
|
||||
# Whether this file store is on the local filesystem.
|
||||
local_storage = False
|
||||
|
||||
def __raise_not_implemented(self):
|
||||
"""
|
||||
Raise a warning about some component not implemented by a
|
||||
subclass of this interface.
|
||||
"""
|
||||
raise NotImplementedError(
|
||||
"This feature not implemented in this storage API implementation.")
|
||||
|
||||
def file_exists(self, filepath):
|
||||
"""
|
||||
Return a boolean asserting whether or not file at filepath
|
||||
exists in our storage system.
|
||||
|
||||
Returns:
|
||||
True / False depending on whether file exists or not.
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def get_file(self, filepath, mode='r'):
|
||||
"""
|
||||
Return a file-like object for reading/writing from this filepath.
|
||||
|
||||
Should create directories, buckets, whatever, as necessary.
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def delete_file(self, filepath):
|
||||
"""
|
||||
Delete or dereference the file at filepath.
|
||||
|
||||
This might need to delete directories, buckets, whatever, for
|
||||
cleanliness. (Be sure to avoid race conditions on that though)
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def file_url(self, filepath):
|
||||
"""
|
||||
Get the URL for this file. This assumes our storage has been
|
||||
mounted with some kind of URL which makes this possible.
|
||||
"""
|
||||
# Subclasses should override this method.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def get_unique_filepath(self, filepath):
|
||||
"""
|
||||
If a filename at filepath already exists, generate a new name.
|
||||
|
||||
Eg, if the filename doesn't exist:
|
||||
>>> storage_handler.get_unique_filename(['dir1', 'dir2', 'fname.jpg'])
|
||||
[u'dir1', u'dir2', u'fname.jpg']
|
||||
|
||||
But if a file does exist, let's get one back with at uuid tacked on:
|
||||
>>> storage_handler.get_unique_filename(['dir1', 'dir2', 'fname.jpg'])
|
||||
[u'dir1', u'dir2', u'd02c3571-dd62-4479-9d62-9e3012dada29-fname.jpg']
|
||||
"""
|
||||
# Make sure we have a clean filepath to start with, since
|
||||
# we'll be possibly tacking on stuff to the filename.
|
||||
filepath = clean_listy_filepath(filepath)
|
||||
|
||||
if self.file_exists(filepath):
|
||||
return filepath[:-1] + ["%s-%s" % (uuid.uuid4(), filepath[-1])]
|
||||
else:
|
||||
return filepath
|
||||
|
||||
def get_local_path(self, filepath):
|
||||
"""
|
||||
If this is a local_storage implementation, give us a link to
|
||||
the local filesystem reference to this file.
|
||||
|
||||
>>> storage_handler.get_local_path(['foo', 'bar', 'baz.jpg'])
|
||||
u'/path/to/mounting/foo/bar/baz.jpg'
|
||||
"""
|
||||
# Subclasses should override this method, if applicable.
|
||||
self.__raise_not_implemented()
|
||||
|
||||
def copy_locally(self, filepath, dest_path):
|
||||
"""
|
||||
Copy this file locally.
|
||||
|
||||
A basic working method for this is provided that should
|
||||
function both for local_storage systems and remote storge
|
||||
systems, but if more efficient systems for copying locally
|
||||
apply to your system, override this method with something more
|
||||
appropriate.
|
||||
"""
|
||||
if self.local_storage:
|
||||
shutil.copy(
|
||||
self.get_local_path(filepath), dest_path)
|
||||
else:
|
||||
with self.get_file(filepath, 'rb') as source_file:
|
||||
with file(dest_path, 'wb') as dest_file:
|
||||
dest_file.write(source_file.read())
|
||||
|
||||
|
||||
###########
|
||||
# Utilities
|
||||
###########
|
||||
|
||||
def clean_listy_filepath(listy_filepath):
|
||||
"""
|
||||
Take a listy filepath (like ['dir1', 'dir2', 'filename.jpg']) and
|
||||
clean out any nastiness from it.
|
||||
|
||||
|
||||
>>> clean_listy_filepath([u'/dir1/', u'foo/../nasty', u'linooks.jpg'])
|
||||
[u'dir1', u'foo_.._nasty', u'linooks.jpg']
|
||||
|
||||
Args:
|
||||
- listy_filepath: a list of filepath components, mediagoblin
|
||||
storage API style.
|
||||
|
||||
Returns:
|
||||
A cleaned list of unicode objects.
|
||||
"""
|
||||
cleaned_filepath = [
|
||||
unicode(secure_filename(filepath))
|
||||
for filepath in listy_filepath]
|
||||
|
||||
if u'' in cleaned_filepath:
|
||||
raise InvalidFilepath(
|
||||
"A filename component could not be resolved into a usable name.")
|
||||
|
||||
return cleaned_filepath
|
||||
|
||||
|
||||
def storage_system_from_config(config_section):
|
||||
"""
|
||||
Utility for setting up a storage system from a config section.
|
||||
|
||||
Note that a special argument may be passed in to
|
||||
the config_section which is "storage_class" which will provide an
|
||||
import path to a storage system. This defaults to
|
||||
"mediagoblin.storage:BasicFileStorage" if otherwise undefined.
|
||||
|
||||
Arguments:
|
||||
- config_section: dictionary of config parameters
|
||||
|
||||
Returns:
|
||||
An instantiated storage system.
|
||||
|
||||
Example:
|
||||
storage_system_from_config(
|
||||
{'base_url': '/media/',
|
||||
'base_dir': '/var/whatever/media/'})
|
||||
|
||||
Will return:
|
||||
BasicFileStorage(
|
||||
base_url='/media/',
|
||||
base_dir='/var/whatever/media')
|
||||
"""
|
||||
# This construct is needed, because dict(config) does
|
||||
# not replace the variables in the config items.
|
||||
config_params = dict(config_section.iteritems())
|
||||
|
||||
if 'storage_class' in config_params:
|
||||
storage_class = config_params['storage_class']
|
||||
config_params.pop('storage_class')
|
||||
else:
|
||||
storage_class = 'mediagoblin.storage.filestorage:BasicFileStorage'
|
||||
|
||||
storage_class = util.import_component(storage_class)
|
||||
return storage_class(**config_params)
|
156
mediagoblin/storage/cloudfiles.py
Normal file
156
mediagoblin/storage/cloudfiles.py
Normal file
@ -0,0 +1,156 @@
|
||||
# GNU MediaGoblin -- federated, autonomous media hosting
|
||||
# Copyright (C) 2011 MediaGoblin contributors. See AUTHORS.
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
#
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU Affero General Public License for more details.
|
||||
#
|
||||
# You should have received a copy of the GNU Affero General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
'''
|
||||
Make it so that ``import cloudfiles`` does not pick THIS file, but the
|
||||
python-cloudfiles one.
|
||||
|
||||
http://docs.python.org/whatsnew/2.5.html#pep-328-absolute-and-relative-imports
|
||||
'''
|
||||
from __future__ import absolute_import
|
||||
|
||||
from mediagoblin.storage import StorageInterface, clean_listy_filepath
|
||||
|
||||
import cloudfiles
|
||||
import mimetypes
|
||||
|
||||
class CloudFilesStorage(StorageInterface):
|
||||
'''
|
||||
OpenStack/Rackspace Cloud's Swift/CloudFiles support
|
||||
'''
|
||||
|
||||
local_storage = False
|
||||
|
||||
def __init__(self, **kwargs):
|
||||
self.param_container = kwargs.get('cloudfiles_container')
|
||||
self.param_user = kwargs.get('cloudfiles_user')
|
||||
self.param_api_key = kwargs.get('cloudfiles_api_key')
|
||||
self.param_host = kwargs.get('cloudfiles_host')
|
||||
self.param_use_servicenet = kwargs.get('cloudfiles_use_servicenet')
|
||||
|
||||
if not self.param_host:
|
||||
print('No CloudFiles host URL specified, '
|
||||
'defaulting to Rackspace US')
|
||||
|
||||
self.connection = cloudfiles.get_connection(
|
||||
username=self.param_user,
|
||||
api_key=self.param_api_key,
|
||||
servicenet=True if self.param_use_servicenet == 'true' or \
|
||||
self.param_use_servicenet == True else False)
|
||||
|
||||
if not self.param_container == \
|
||||
self.connection.get_container(self.param_container):
|
||||
self.container = self.connection.create_container(
|
||||
self.param_container)
|
||||
self.container.make_public(
|
||||
ttl=60 * 60 * 2)
|
||||
else:
|
||||
self.container = self.connection.get_container(
|
||||
self.param_container)
|
||||
|
||||
self.container_uri = self.container.public_uri()
|
||||
|
||||
def _resolve_filepath(self, filepath):
|
||||
return '/'.join(
|
||||
clean_listy_filepath(filepath))
|
||||
|
||||
def file_exists(self, filepath):
|
||||
try:
|
||||
object = self.container.get_object(
|
||||
self._resolve_filepath(filepath))
|
||||
return True
|
||||
except cloudfiles.errors.NoSuchObject:
|
||||
return False
|
||||
|
||||
def get_file(self, filepath, *args, **kwargs):
|
||||
"""
|
||||
- Doesn't care about the "mode" argument
|
||||
"""
|
||||
try:
|
||||
obj = self.container.get_object(
|
||||
self._resolve_filepath(filepath))
|
||||
except cloudfiles.errors.NoSuchObject:
|
||||
obj = self.container.create_object(
|
||||
self._resolve_filepath(filepath))
|
||||
|
||||
mimetype = mimetypes.guess_type(
|
||||
filepath[-1])
|
||||
|
||||
if mimetype:
|
||||
obj.content_type = mimetype[0]
|
||||
|
||||
return CloudFilesStorageObjectWrapper(obj, *args, **kwargs)
|
||||
|
||||
def delete_file(self, filepath):
|
||||
# TODO: Also delete unused directories if empty (safely, with
|
||||
# checks to avoid race conditions).
|
||||
self.container.delete_object(
|
||||
self._resolve_filepath(filepath))
|
||||
|
||||
def file_url(self, filepath):
|
||||
return '/'.join([
|
||||
self.container_uri,
|
||||
self._resolve_filepath(filepath)])
|
||||
|
||||
|
||||
class CloudFilesStorageObjectWrapper():
|
||||
"""
|
||||
Wrapper for python-cloudfiles's cloudfiles.storage_object.Object
|
||||
used to circumvent the mystic `medium.jpg` corruption issue, where
|
||||
we had both python-cloudfiles and PIL doing buffering on both
|
||||
ends and that breaking things.
|
||||
|
||||
This wrapper currently meets mediagoblin's needs for a public_store
|
||||
file-like object.
|
||||
"""
|
||||
def __init__(self, storage_object, *args, **kwargs):
|
||||
self.storage_object = storage_object
|
||||
|
||||
def read(self, *args, **kwargs):
|
||||
return self.storage_object.read(*args, **kwargs)
|
||||
|
||||
def write(self, data, *args, **kwargs):
|
||||
"""
|
||||
write data to the cloudfiles storage object
|
||||
|
||||
The original motivation for this wrapper is to ensure
|
||||
that buffered writing to a cloudfiles storage object does not overwrite
|
||||
any preexisting data.
|
||||
|
||||
Currently this method does not support any write modes except "append".
|
||||
However if we should need it it would be easy implement.
|
||||
"""
|
||||
if self.storage_object.size and type(data) == str:
|
||||
data = self.read() + data
|
||||
|
||||
self.storage_object.write(data, *args, **kwargs)
|
||||
|
||||
def close(self):
|
||||
pass
|
||||
|
||||
def __enter__(self):
|
||||
"""
|
||||
Context Manager API implementation
|
||||
http://docs.python.org/library/stdtypes.html#context-manager-types
|
||||
"""
|
||||
return self
|
||||
|
||||
def __exit__(self, *exc_info):
|
||||
"""
|
||||
Context Manger API implementation
|
||||
see self.__enter__()
|
||||
"""
|
||||
self.close()
|
78
mediagoblin/storage/filestorage.py
Normal file
78
mediagoblin/storage/filestorage.py
Normal file
@ -0,0 +1,78 @@
|
||||
# GNU MediaGoblin -- federated, autonomous media hosting
|
||||
# Copyright (C) 2011 MediaGoblin contributors. See AUTHORS.
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
#
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU Affero General Public License for more details.
|
||||
#
|
||||
# You should have received a copy of the GNU Affero General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
from mediagoblin.storage import (
|
||||
StorageInterface,
|
||||
clean_listy_filepath,
|
||||
NoWebServing)
|
||||
|
||||
import os
|
||||
import urlparse
|
||||
|
||||
|
||||
class BasicFileStorage(StorageInterface):
|
||||
"""
|
||||
Basic local filesystem implementation of storage API
|
||||
"""
|
||||
|
||||
local_storage = True
|
||||
|
||||
def __init__(self, base_dir, base_url=None, **kwargs):
|
||||
"""
|
||||
Keyword arguments:
|
||||
- base_dir: Base directory things will be served out of. MUST
|
||||
be an absolute path.
|
||||
- base_url: URL files will be served from
|
||||
"""
|
||||
self.base_dir = base_dir
|
||||
self.base_url = base_url
|
||||
|
||||
def _resolve_filepath(self, filepath):
|
||||
"""
|
||||
Transform the given filepath into a local filesystem filepath.
|
||||
"""
|
||||
return os.path.join(
|
||||
self.base_dir, *clean_listy_filepath(filepath))
|
||||
|
||||
def file_exists(self, filepath):
|
||||
return os.path.exists(self._resolve_filepath(filepath))
|
||||
|
||||
def get_file(self, filepath, mode='r'):
|
||||
# Make directories if necessary
|
||||
if len(filepath) > 1:
|
||||
directory = self._resolve_filepath(filepath[:-1])
|
||||
if not os.path.exists(directory):
|
||||
os.makedirs(directory)
|
||||
|
||||
# Grab and return the file in the mode specified
|
||||
return open(self._resolve_filepath(filepath), mode)
|
||||
|
||||
def delete_file(self, filepath):
|
||||
# TODO: Also delete unused directories if empty (safely, with
|
||||
# checks to avoid race conditions).
|
||||
os.remove(self._resolve_filepath(filepath))
|
||||
|
||||
def file_url(self, filepath):
|
||||
if not self.base_url:
|
||||
raise NoWebServing(
|
||||
"base_url not set, cannot provide file urls")
|
||||
|
||||
return urlparse.urljoin(
|
||||
self.base_url,
|
||||
'/'.join(clean_listy_filepath(filepath)))
|
||||
|
||||
def get_local_path(self, filepath):
|
||||
return self._resolve_filepath(filepath)
|
156
mediagoblin/storage/mountstorage.py
Normal file
156
mediagoblin/storage/mountstorage.py
Normal file
@ -0,0 +1,156 @@
|
||||
# GNU MediaGoblin -- federated, autonomous media hosting
|
||||
# Copyright (C) 2011 MediaGoblin contributors. See AUTHORS.
|
||||
#
|
||||
# This program is free software: you can redistribute it and/or modify
|
||||
# it under the terms of the GNU Affero General Public License as published by
|
||||
# the Free Software Foundation, either version 3 of the License, or
|
||||
# (at your option) any later version.
|
||||
#
|
||||
# This program is distributed in the hope that it will be useful,
|
||||
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
# GNU Affero General Public License for more details.
|
||||
#
|
||||
# You should have received a copy of the GNU Affero General Public License
|
||||
# along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||||
|
||||
from medigoblin.storage import StorageInterface, clean_listy_filepath
|
||||
|
||||
|
||||
class MountStorage(StorageInterface):
|
||||
"""
|
||||
Experimental "Mount" virtual Storage Interface
|
||||
|
||||
This isn't an interface to some real storage, instead it's a
|
||||
redirecting interface, that redirects requests to other
|
||||
"StorageInterface"s.
|
||||
|
||||
For example, say you have the paths:
|
||||
|
||||
1. ['user_data', 'cwebber', 'avatar.jpg']
|
||||
2. ['user_data', 'elrond', 'avatar.jpg']
|
||||
3. ['media_entries', '34352f304c3f4d0ad8ad0f043522b6f2', 'thumb.jpg']
|
||||
|
||||
You could mount media_entries under CloudFileStorage and user_data
|
||||
under BasicFileStorage. Then 1 would be passed to
|
||||
BasicFileStorage under the path ['cwebber', 'avatar.jpg'] and 3
|
||||
would be passed to CloudFileStorage under
|
||||
['34352f304c3f4d0ad8ad0f043522b6f2', 'thumb.jpg'].
|
||||
|
||||
In other words, this is kind of like mounting /home/ and /etc/
|
||||
under different filesystems on your operating system... but with
|
||||
mediagoblin filestorages :)
|
||||
|
||||
To set this up, you currently need to call the mount() method with
|
||||
the target path and a backend, that shall be available under that
|
||||
target path. You have to mount things in a sensible order,
|
||||
especially you can't mount ["a", "b"] before ["a"].
|
||||
"""
|
||||
def __init__(self, **kwargs):
|
||||
self.mounttab = {}
|
||||
|
||||
def mount(self, dirpath, backend):
|
||||
"""
|
||||
Mount a new backend under dirpath
|
||||
"""
|
||||
new_ent = clean_listy_filepath(dirpath)
|
||||
|
||||
print "Mounting:", repr(new_ent)
|
||||
already, rem_1, table, rem_2 = self._resolve_to_backend(new_ent, True)
|
||||
print "===", repr(already), repr(rem_1), repr(rem_2), len(table)
|
||||
|
||||
assert (len(rem_2) > 0) or (None not in table), \
|
||||
"That path is already mounted"
|
||||
assert (len(rem_2) > 0) or (len(table) == 0), \
|
||||
"A longer path is already mounted here"
|
||||
|
||||
for part in rem_2:
|
||||
table[part] = {}
|
||||
table = table[part]
|
||||
table[None] = backend
|
||||
|
||||
def _resolve_to_backend(self, filepath, extra_info=False):
|
||||
"""
|
||||
extra_info = True is for internal use!
|
||||
|
||||
Normally, returns the backend and the filepath inside that backend.
|
||||
|
||||
With extra_info = True it returns the last directory node and the
|
||||
remaining filepath from there in addition.
|
||||
"""
|
||||
table = self.mounttab
|
||||
filepath = filepath[:]
|
||||
res_fp = None
|
||||
while True:
|
||||
new_be = table.get(None)
|
||||
if (new_be is not None) or res_fp is None:
|
||||
res_be = new_be
|
||||
res_fp = filepath[:]
|
||||
res_extra = (table, filepath[:])
|
||||
# print "... New res: %r, %r, %r" % (res_be, res_fp, res_extra)
|
||||
if len(filepath) == 0:
|
||||
break
|
||||
query = filepath.pop(0)
|
||||
entry = table.get(query)
|
||||
if entry is not None:
|
||||
table = entry
|
||||
res_extra = (table, filepath[:])
|
||||
else:
|
||||
break
|
||||
if extra_info:
|
||||
return (res_be, res_fp) + res_extra
|
||||
else:
|
||||
return (res_be, res_fp)
|
||||
|
||||
def resolve_to_backend(self, filepath):
|
||||
backend, filepath = self._resolve_to_backend(filepath)
|
||||
if backend is None:
|
||||
raise Error("Path not mounted")
|
||||
return backend, filepath
|
||||
|
||||
def __repr__(self, table=None, indent=[]):
|
||||
res = []
|
||||
if table is None:
|
||||
res.append("MountStorage<")
|
||||
table = self.mounttab
|
||||
v = table.get(None)
|
||||
if v:
|
||||
res.append(" " * len(indent) + repr(indent) + ": " + repr(v))
|
||||
for k, v in table.iteritems():
|
||||
if k == None:
|
||||
continue
|
||||
res.append(" " * len(indent) + repr(k) + ":")
|
||||
res += self.__repr__(v, indent + [k])
|
||||
if table is self.mounttab:
|
||||
res.append(">")
|
||||
return "\n".join(res)
|
||||
else:
|
||||
return res
|
||||
|
||||
def file_exists(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.file_exists(filepath)
|
||||
|
||||
def get_file(self, filepath, mode='r'):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.get_file(filepath, mode)
|
||||
|
||||
def delete_file(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.delete_file(filepath)
|
||||
|
||||
def file_url(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.file_url(filepath)
|
||||
|
||||
def get_local_path(self, filepath):
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
return backend.get_local_path(filepath)
|
||||
|
||||
def copy_locally(self, filepath, dest_path):
|
||||
"""
|
||||
Need to override copy_locally, because the local_storage
|
||||
attribute is not correct.
|
||||
"""
|
||||
backend, filepath = self.resolve_to_backend(filepath)
|
||||
backend.copy_locally(filepath, dest_path)
|
@ -52,7 +52,7 @@ class FakeStorageSystem():
|
||||
self.foobie = foobie
|
||||
self.blech = blech
|
||||
|
||||
class FakeRemoteStorage(storage.BasicFileStorage):
|
||||
class FakeRemoteStorage(storage.filestorage.BasicFileStorage):
|
||||
# Theoretically despite this, all the methods should work but it
|
||||
# should force copying to the workbench
|
||||
local_storage = False
|
||||
@ -66,7 +66,7 @@ def test_storage_system_from_config():
|
||||
'garbage_arg': 'trash'})
|
||||
assert this_storage.base_url == 'http://example.org/moodia/'
|
||||
assert this_storage.base_dir == '/tmp/'
|
||||
assert this_storage.__class__ is storage.BasicFileStorage
|
||||
assert this_storage.__class__ is storage.filestorage.BasicFileStorage
|
||||
|
||||
this_storage = storage.storage_system_from_config(
|
||||
{'foobie': 'eiboof',
|
||||
@ -88,7 +88,7 @@ def get_tmp_filestorage(mount_url=None, fake_remote=False):
|
||||
if fake_remote:
|
||||
this_storage = FakeRemoteStorage(tmpdir, mount_url)
|
||||
else:
|
||||
this_storage = storage.BasicFileStorage(tmpdir, mount_url)
|
||||
this_storage = storage.filestorage.BasicFileStorage(tmpdir, mount_url)
|
||||
return tmpdir, this_storage
|
||||
|
||||
|
||||
|
Loading…
x
Reference in New Issue
Block a user