Added vip and daily download limit to addic7ed and try to slow down a little to prevent being ip blocked by the provider.

This commit is contained in:
morpheus65535 2021-12-18 09:41:48 -05:00
parent 6192df6019
commit 5d1dccde99
4 changed files with 62 additions and 12 deletions

View file

@ -132,7 +132,8 @@ defaults = {
},
'addic7ed': {
'username': '',
'password': ''
'password': '',
'vip': 'False'
},
'podnapisi': {
'verify_ssl': 'True'

View file

@ -127,6 +127,7 @@ def get_providers_auth():
'addic7ed' : {
'username': settings.addic7ed.username,
'password': settings.addic7ed.password,
'is_vip': settings.addic7ed.getboolean('vip'),
},
'opensubtitles' : {
'username' : settings.opensubtitles.username,

View file

@ -21,6 +21,10 @@ export const ProviderList: Readonly<ProviderInfo[]> = [
defaultKey: {
username: "",
password: "",
vip: false,
},
keyNameOverride: {
vip: "VIP",
},
},
{ key: "argenteam", description: "LATAM Spanish Subtitles Provider" },

View file

@ -23,6 +23,8 @@ from subzero.language import Language
logger = logging.getLogger(__name__)
show_cells_re = re.compile(b'<td class="(?:version|vr)">.*?</td>', re.DOTALL)
#: Series header parsing regex
series_year_re = re.compile(r'^(?P<series>[ \w\'.:(),*&!?-]+?)(?: \((?P<year>\d{4})\))?$')
@ -68,6 +70,7 @@ class Addic7edProvider(_Addic7edProvider):
languages.update(set(Language.rebuild(l, hi=True) for l in languages))
video_types = (Episode, Movie)
vip = False
USE_ADDICTED_RANDOM_AGENTS = False
hearing_impaired_verifiable = True
subtitle_class = Addic7edSubtitle
@ -76,9 +79,10 @@ class Addic7edProvider(_Addic7edProvider):
sanitize_characters = {'-', ':', '(', ')', '.', '/'}
last_show_ids_fetch_key = "addic7ed_last_id_fetch"
def __init__(self, username=None, password=None, use_random_agents=False):
def __init__(self, username=None, password=None, use_random_agents=False, is_vip=False):
super(Addic7edProvider, self).__init__(username=username, password=password)
self.USE_ADDICTED_RANDOM_AGENTS = use_random_agents
self.vip = is_vip
if not all((username, password)):
raise ConfigurationError('Username and password must be specified')
@ -95,7 +99,7 @@ class Addic7edProvider(_Addic7edProvider):
# login
if self.username and self.password:
def check_verification(cache_region):
rr = self.session.get(self.server_url + 'panel.php', allow_redirects=False, timeout=60,
rr = self.session.get(self.server_url + 'panel.php', allow_redirects=False, timeout=10,
headers={"Referer": self.server_url})
if rr.status_code == 302:
logger.info('Addic7ed: Login expired')
@ -115,7 +119,7 @@ class Addic7edProvider(_Addic7edProvider):
tries = 0
while tries <= 3:
tries += 1
r = self.session.get(self.server_url + 'login.php', timeout=60, headers={"Referer": self.server_url})
r = self.session.get(self.server_url + 'login.php', timeout=10, headers={"Referer": self.server_url})
if "g-recaptcha" in r.text or "grecaptcha" in r.text:
logger.info('Addic7ed: Solving captcha. This might take a couple of minutes, but should only '
'happen once every so often')
@ -139,11 +143,13 @@ class Addic7edProvider(_Addic7edProvider):
if tries >= 3:
raise Exception("Addic7ed: Couldn't solve captcha!")
logger.info("Addic7ed: Couldn't solve captcha! Retrying")
time.sleep(4)
continue
data[g] = result
r = self.session.post(self.server_url + 'dologin.php', data, allow_redirects=False, timeout=60,
time.sleep(1)
r = self.session.post(self.server_url + 'dologin.php', data, allow_redirects=False, timeout=10,
headers={"Referer": self.server_url + "login.php"})
if "relax, slow down" in r.text:
@ -157,6 +163,7 @@ class Addic7edProvider(_Addic7edProvider):
logger.error("Addic7ed: Something went wrong when logging in")
raise AuthenticationError(self.username)
logger.info("Addic7ed: Something went wrong when logging in; retrying")
time.sleep(4)
continue
break
@ -165,6 +172,8 @@ class Addic7edProvider(_Addic7edProvider):
logger.debug('Addic7ed: Logged in')
self.logged_in = True
time.sleep(2)
def terminate(self):
self.session.close()
@ -238,7 +247,7 @@ class Addic7edProvider(_Addic7edProvider):
# get the movie id
logger.info('Getting movie id')
r = self.session.get(self.server_url + 'search.php?search=' + quote_plus(movie), timeout=60)
r = self.session.get(self.server_url + 'search.php?search=' + quote_plus(movie), timeout=10)
r.raise_for_status()
soup = ParserBeautifulSoup(r.content.decode('utf-8', 'ignore'), ['lxml', 'html.parser'])
@ -285,10 +294,18 @@ class Addic7edProvider(_Addic7edProvider):
logger.info('Getting show ids')
region.set(self.last_show_ids_fetch_key, datetime.datetime.now())
r = self.session.get(self.server_url + 'shows.php', timeout=60)
r = self.session.get(self.server_url + 'shows.php', timeout=10)
r.raise_for_status()
soup = ParserBeautifulSoup(r.content.decode('utf-8', 'ignore'), ['lxml', 'html.parser'])
# LXML parser seems to fail when parsing Addic7ed.com HTML markup.
# Last known version to work properly is 3.6.4 (next version, 3.7.0, fails)
# Assuming the site's markup is bad, and stripping it down to only contain what's needed.
show_cells = re.findall(show_cells_re, r.content)
if show_cells:
soup = ParserBeautifulSoup(''.join(show_cells).decode('utf-8', 'ignore'), ['lxml', 'html.parser'])
else:
# If RegEx fails, fall back to original r.content and use 'html.parser'
soup = ParserBeautifulSoup(r.content, ['html.parser'])
# populate the show ids
show_ids = {}
@ -345,7 +362,7 @@ class Addic7edProvider(_Addic7edProvider):
headers = {
"referer": self.server_url + "srch.php"
}
r = self.session.get(self.server_url + endpoint, params=params, timeout=60, headers=headers)
r = self.session.get(self.server_url + endpoint, params=params, timeout=10, headers=headers)
r.raise_for_status()
if r.text and "Sorry, your search" not in r.text:
@ -386,7 +403,7 @@ class Addic7edProvider(_Addic7edProvider):
logger.info('Getting the page of show id %d, season %d', show_id, season)
r = self.session.get(self.server_url + 'ajax_loadShow.php',
params={'show': show_id, 'season': season},
timeout=60,
timeout=10,
headers={
"referer": "%sshow/%s" % (self.server_url, show_id),
"X-Requested-With": "XMLHttpRequest"
@ -446,7 +463,7 @@ class Addic7edProvider(_Addic7edProvider):
# get the page of the movie
logger.info('Getting the page of movie id %d', movie_id)
r = self.session.get(self.server_url + 'movie/' + movie_id,
timeout=60,
timeout=10,
headers={
"referer": self.server_url,
"X-Requested-With": "XMLHttpRequest"
@ -550,9 +567,30 @@ class Addic7edProvider(_Addic7edProvider):
return []
def download_subtitle(self, subtitle):
last_dls = region.get("addic7ed_dls")
now = datetime.datetime.now()
one_day = datetime.timedelta(hours=24)
def raise_limit():
logger.info("Addic7ed: Downloads per day exceeded (%s)", cap)
raise DownloadLimitPerDayExceeded
if type(last_dls) is not list:
last_dls = []
else:
# filter all non-expired DLs
last_dls = list(filter(lambda t: t + one_day > now, last_dls))
region.set("addic7ed_dls", last_dls)
cap = self.vip and 80 or 40
amount = len(last_dls)
if amount >= cap:
raise_limit()
# download the subtitle
r = self.session.get(self.server_url + subtitle.download_link, headers={'Referer': subtitle.page_link},
timeout=60)
timeout=10)
r.raise_for_status()
if r.status_code == 304:
@ -569,3 +607,9 @@ class Addic7edProvider(_Addic7edProvider):
raise DownloadLimitExceeded
subtitle.content = fix_line_ending(r.content)
last_dls.append(datetime.datetime.now())
region.set("addic7ed_dls", last_dls)
logger.info("Addic7ed: Used %s/%s downloads", amount + 1, cap)
if amount + 1 >= cap:
raise_limit()