mirror of
https://github.com/Tautulli/Tautulli.git
synced 2025-07-16 02:02:58 -07:00
Add cloudinary v1.11.0
This commit is contained in:
parent
5710bcb43c
commit
136260a822
27 changed files with 10855 additions and 0 deletions
325
lib/cloudinary/uploader.py
Normal file
325
lib/cloudinary/uploader.py
Normal file
|
@ -0,0 +1,325 @@
|
|||
# Copyright Cloudinary
|
||||
import json
|
||||
import re
|
||||
import socket
|
||||
from os.path import getsize
|
||||
|
||||
import cloudinary
|
||||
import urllib3
|
||||
import certifi
|
||||
from cloudinary import utils
|
||||
from cloudinary.api import Error
|
||||
from cloudinary.compat import string_types
|
||||
from urllib3.exceptions import HTTPError
|
||||
from urllib3 import PoolManager
|
||||
|
||||
try:
|
||||
from urllib3.contrib.appengine import AppEngineManager, is_appengine_sandbox
|
||||
except Exception:
|
||||
def is_appengine_sandbox():
|
||||
return False
|
||||
|
||||
try: # Python 2.7+
|
||||
from collections import OrderedDict
|
||||
except ImportError:
|
||||
from urllib3.packages.ordered_dict import OrderedDict
|
||||
|
||||
if is_appengine_sandbox():
|
||||
# AppEngineManager uses AppEngine's URLFetch API behind the scenes
|
||||
_http = AppEngineManager()
|
||||
else:
|
||||
# PoolManager uses a socket-level API behind the scenes
|
||||
_http = PoolManager(
|
||||
cert_reqs='CERT_REQUIRED',
|
||||
ca_certs=certifi.where()
|
||||
)
|
||||
|
||||
|
||||
def upload(file, **options):
|
||||
params = utils.build_upload_params(**options)
|
||||
return call_api("upload", params, file=file, **options)
|
||||
|
||||
|
||||
def unsigned_upload(file, upload_preset, **options):
|
||||
return upload(file, upload_preset=upload_preset, unsigned=True, **options)
|
||||
|
||||
|
||||
def upload_image(file, **options):
|
||||
result = upload(file, **options)
|
||||
return cloudinary.CloudinaryImage(
|
||||
result["public_id"], version=str(result["version"]),
|
||||
format=result.get("format"), metadata=result)
|
||||
|
||||
|
||||
def upload_resource(file, **options):
|
||||
result = upload(file, **options)
|
||||
return cloudinary.CloudinaryResource(
|
||||
result["public_id"], version=str(result["version"]),
|
||||
format=result.get("format"), type=result["type"], resource_type=result["resource_type"], metadata=result)
|
||||
|
||||
|
||||
def upload_large(file, **options):
|
||||
""" Upload large files. """
|
||||
upload_id = utils.random_public_id()
|
||||
with open(file, 'rb') as file_io:
|
||||
results = None
|
||||
current_loc = 0
|
||||
chunk_size = options.get("chunk_size", 20000000)
|
||||
file_size = getsize(file)
|
||||
chunk = file_io.read(chunk_size)
|
||||
while chunk:
|
||||
range = "bytes {0}-{1}/{2}".format(current_loc, current_loc + len(chunk) - 1, file_size)
|
||||
current_loc += len(chunk)
|
||||
|
||||
results = upload_large_part((file, chunk),
|
||||
http_headers={"Content-Range": range, "X-Unique-Upload-Id": upload_id},
|
||||
**options)
|
||||
options["public_id"] = results.get("public_id")
|
||||
chunk = file_io.read(chunk_size)
|
||||
return results
|
||||
|
||||
|
||||
def upload_large_part(file, **options):
|
||||
""" Upload large files. """
|
||||
params = utils.build_upload_params(**options)
|
||||
if 'resource_type' not in options: options['resource_type'] = "raw"
|
||||
return call_api("upload", params, file=file, **options)
|
||||
|
||||
|
||||
def destroy(public_id, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"type": options.get("type"),
|
||||
"invalidate": options.get("invalidate"),
|
||||
"public_id": public_id
|
||||
}
|
||||
return call_api("destroy", params, **options)
|
||||
|
||||
|
||||
def rename(from_public_id, to_public_id, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"type": options.get("type"),
|
||||
"overwrite": options.get("overwrite"),
|
||||
"invalidate": options.get("invalidate"),
|
||||
"from_public_id": from_public_id,
|
||||
"to_public_id": to_public_id
|
||||
}
|
||||
return call_api("rename", params, **options)
|
||||
|
||||
|
||||
def explicit(public_id, **options):
|
||||
params = utils.build_upload_params(**options)
|
||||
params["public_id"] = public_id
|
||||
return call_api("explicit", params, **options)
|
||||
|
||||
|
||||
def create_archive(**options):
|
||||
params = utils.archive_params(**options)
|
||||
if options.get("target_format") is not None:
|
||||
params["target_format"] = options.get("target_format")
|
||||
return call_api("generate_archive", params, **options)
|
||||
|
||||
|
||||
def create_zip(**options):
|
||||
return create_archive(target_format="zip", **options)
|
||||
|
||||
|
||||
def generate_sprite(tag, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"tag": tag,
|
||||
"async": options.get("async"),
|
||||
"notification_url": options.get("notification_url"),
|
||||
"transformation": utils.generate_transformation_string(fetch_format=options.get("format"), **options)[0]
|
||||
}
|
||||
return call_api("sprite", params, **options)
|
||||
|
||||
|
||||
def multi(tag, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"tag": tag,
|
||||
"format": options.get("format"),
|
||||
"async": options.get("async"),
|
||||
"notification_url": options.get("notification_url"),
|
||||
"transformation": utils.generate_transformation_string(**options)[0]
|
||||
}
|
||||
return call_api("multi", params, **options)
|
||||
|
||||
|
||||
def explode(public_id, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"public_id": public_id,
|
||||
"format": options.get("format"),
|
||||
"notification_url": options.get("notification_url"),
|
||||
"transformation": utils.generate_transformation_string(**options)[0]
|
||||
}
|
||||
return call_api("explode", params, **options)
|
||||
|
||||
|
||||
# options may include 'exclusive' (boolean) which causes clearing this tag from all other resources
|
||||
def add_tag(tag, public_ids=None, **options):
|
||||
exclusive = options.pop("exclusive", None)
|
||||
command = "set_exclusive" if exclusive else "add"
|
||||
return call_tags_api(tag, command, public_ids, **options)
|
||||
|
||||
|
||||
def remove_tag(tag, public_ids=None, **options):
|
||||
return call_tags_api(tag, "remove", public_ids, **options)
|
||||
|
||||
|
||||
def replace_tag(tag, public_ids=None, **options):
|
||||
return call_tags_api(tag, "replace", public_ids, **options)
|
||||
|
||||
|
||||
def remove_all_tags(public_ids, **options):
|
||||
"""
|
||||
Remove all tags from the specified public IDs.
|
||||
:param public_ids: the public IDs of the resources to update
|
||||
:param options: additional options passed to the request
|
||||
:return: dictionary with a list of public IDs that were updated
|
||||
"""
|
||||
return call_tags_api(None, "remove_all", public_ids, **options)
|
||||
|
||||
|
||||
def add_context(context, public_ids, **options):
|
||||
"""
|
||||
Add a context keys and values. If a particular key already exists, the value associated with the key is updated.
|
||||
:param context: dictionary of context
|
||||
:param public_ids: the public IDs of the resources to update
|
||||
:param options: additional options passed to the request
|
||||
:return: dictionary with a list of public IDs that were updated
|
||||
"""
|
||||
return call_context_api(context, "add", public_ids, **options)
|
||||
|
||||
|
||||
def remove_all_context(public_ids, **options):
|
||||
"""
|
||||
Remove all custom context from the specified public IDs.
|
||||
:param public_ids: the public IDs of the resources to update
|
||||
:param options: additional options passed to the request
|
||||
:return: dictionary with a list of public IDs that were updated
|
||||
"""
|
||||
return call_context_api(None, "remove_all", public_ids, **options)
|
||||
|
||||
|
||||
def call_tags_api(tag, command, public_ids=None, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"tag": tag,
|
||||
"public_ids": utils.build_array(public_ids),
|
||||
"command": command,
|
||||
"type": options.get("type")
|
||||
}
|
||||
return call_api("tags", params, **options)
|
||||
|
||||
|
||||
def call_context_api(context, command, public_ids=None, **options):
|
||||
params = {
|
||||
"timestamp": utils.now(),
|
||||
"context": utils.encode_context(context),
|
||||
"public_ids": utils.build_array(public_ids),
|
||||
"command": command,
|
||||
"type": options.get("type")
|
||||
}
|
||||
return call_api("context", params, **options)
|
||||
|
||||
|
||||
TEXT_PARAMS = ["public_id",
|
||||
"font_family",
|
||||
"font_size",
|
||||
"font_color",
|
||||
"text_align",
|
||||
"font_weight",
|
||||
"font_style",
|
||||
"background",
|
||||
"opacity",
|
||||
"text_decoration"
|
||||
]
|
||||
|
||||
|
||||
def text(text, **options):
|
||||
params = {"timestamp": utils.now(), "text": text}
|
||||
for key in TEXT_PARAMS:
|
||||
params[key] = options.get(key)
|
||||
return call_api("text", params, **options)
|
||||
|
||||
|
||||
def call_api(action, params, http_headers=None, return_error=False, unsigned=False, file=None, timeout=None, **options):
|
||||
if http_headers is None:
|
||||
http_headers = {}
|
||||
file_io = None
|
||||
try:
|
||||
if unsigned:
|
||||
params = utils.cleanup_params(params)
|
||||
else:
|
||||
params = utils.sign_request(params, options)
|
||||
|
||||
param_list = OrderedDict()
|
||||
for k, v in params.items():
|
||||
if isinstance(v, list):
|
||||
for i in range(len(v)):
|
||||
param_list["{0}[{1}]".format(k, i)] = v[i]
|
||||
elif v:
|
||||
param_list[k] = v
|
||||
|
||||
api_url = utils.cloudinary_api_url(action, **options)
|
||||
if file:
|
||||
if isinstance(file, string_types):
|
||||
if re.match(r'ftp:|https?:|s3:|data:[^;]*;base64,([a-zA-Z0-9\/+\n=]+)$', file):
|
||||
# URL
|
||||
name = None
|
||||
data = file
|
||||
else:
|
||||
# file path
|
||||
name = file
|
||||
with open(file, "rb") as opened:
|
||||
data = opened.read()
|
||||
elif hasattr(file, 'read') and callable(file.read):
|
||||
# stream
|
||||
data = file.read()
|
||||
name = file.name if hasattr(file, 'name') and isinstance(file.name, str) else "stream"
|
||||
elif isinstance(file, tuple):
|
||||
name = None
|
||||
data = file
|
||||
else:
|
||||
# Not a string, not a stream
|
||||
name = "file"
|
||||
data = file
|
||||
|
||||
param_list["file"] = (name, data) if name else data
|
||||
|
||||
headers = {"User-Agent": cloudinary.get_user_agent()}
|
||||
headers.update(http_headers)
|
||||
|
||||
kw = {}
|
||||
if timeout is not None:
|
||||
kw['timeout'] = timeout
|
||||
|
||||
code = 200
|
||||
try:
|
||||
response = _http.request("POST", api_url, param_list, headers, **kw)
|
||||
except HTTPError as e:
|
||||
raise Error("Unexpected error - {0!r}".format(e))
|
||||
except socket.error as e:
|
||||
raise Error("Socket error: {0!r}".format(e))
|
||||
|
||||
try:
|
||||
result = json.loads(response.data.decode('utf-8'))
|
||||
except Exception as e:
|
||||
# Error is parsing json
|
||||
raise Error("Error parsing server response (%d) - %s. Got - %s", response.status, response, e)
|
||||
|
||||
if "error" in result:
|
||||
if response.status not in [200, 400, 401, 403, 404, 500]:
|
||||
code = response.status
|
||||
if return_error:
|
||||
result["error"]["http_code"] = code
|
||||
else:
|
||||
raise Error(result["error"]["message"])
|
||||
|
||||
return result
|
||||
finally:
|
||||
if file_io: file_io.close()
|
Loading…
Add table
Add a link
Reference in a new issue