1
0
mirror of https://github.com/moparisthebest/SickRage synced 2024-11-14 21:35:03 -05:00
SickRage/sickbeard/scene_exceptions.py

321 lines
12 KiB
Python
Raw Normal View History

# Author: Nic Wolfe <nic@wolfeden.ca>
# URL: http://code.google.com/p/sickbeard/
#
2014-05-23 08:37:22 -04:00
# This file is part of SickRage.
#
2014-05-23 08:37:22 -04:00
# SickRage is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
2014-05-23 08:37:22 -04:00
# SickRage 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 General Public License for more details.
#
# You should have received a copy of the GNU General Public License
2014-05-23 08:37:22 -04:00
# along with SickRage. If not, see <http://www.gnu.org/licenses/>.
import re
import time
import threading
import datetime
import sickbeard
import adba
from sickbeard import helpers
from sickbeard import name_cache
from sickbeard import logger
from sickbeard import db
from sickbeard import encodingKludge as ek
exception_dict = {}
anidb_exception_dict = {}
xem_exception_dict = {}
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
exceptionsCache = {}
exceptionsSeasonCache = {}
exceptionLock = threading.Lock()
def shouldRefresh(list):
2014-06-30 06:20:49 -04:00
MAX_REFRESH_AGE_SECS = 86400 # 1 day
myDB = db.DBConnection('cache.db')
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
rows = myDB.select("SELECT last_refreshed FROM scene_exceptions_refresh WHERE list = ?", [list])
if rows:
lastRefresh = int(rows[0]['last_refreshed'])
return int(time.mktime(datetime.datetime.today().timetuple())) > lastRefresh + MAX_REFRESH_AGE_SECS
else:
return True
def setLastRefresh(list):
myDB = db.DBConnection('cache.db')
myDB.upsert("scene_exceptions_refresh",
{'last_refreshed': int(time.mktime(datetime.datetime.today().timetuple()))},
{'list': list})
def get_scene_exceptions(indexer_id, season=-1):
"""
Given a indexer_id, return a list of all the scene exceptions.
"""
global exceptionsCache
2014-06-30 09:36:20 -04:00
exceptionsList = []
if indexer_id not in exceptionsCache or season not in exceptionsCache[indexer_id]:
myDB = db.DBConnection('cache.db')
exceptions = myDB.select("SELECT show_name FROM scene_exceptions WHERE indexer_id = ? and season = ?",
[indexer_id, season])
if exceptions:
exceptionsList = list(set([cur_exception["show_name"] for cur_exception in exceptions]))
if not indexer_id in exceptionsCache:
exceptionsCache[indexer_id] = {}
exceptionsCache[indexer_id][season] = exceptionsList
else:
exceptionsList = exceptionsCache[indexer_id][season]
2014-06-30 09:36:20 -04:00
if season == 1: # if we where looking for season 1 we can add generic names
exceptionsList += get_scene_exceptions(indexer_id, season=-1)
2014-06-30 09:36:20 -04:00
return exceptionsList
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
def get_all_scene_exceptions(indexer_id):
2014-06-30 09:36:20 -04:00
exceptionsDict = {}
myDB = db.DBConnection('cache.db')
exceptions = myDB.select("SELECT show_name,season FROM scene_exceptions WHERE indexer_id = ?", [indexer_id])
if exceptions:
for cur_exception in exceptions:
2014-06-30 09:36:20 -04:00
if not cur_exception["season"] in exceptionsDict:
exceptionsDict[cur_exception["season"]] = []
exceptionsDict[cur_exception["season"]].append(cur_exception["show_name"])
2014-06-30 09:36:20 -04:00
return exceptionsDict
def get_scene_seasons(indexer_id):
"""
return a list of season numbers that have scene exceptions
"""
global exceptionsSeasonCache
exceptionsSeasonList = []
if indexer_id not in exceptionsSeasonCache:
myDB = db.DBConnection('cache.db')
sqlResults = myDB.select("SELECT DISTINCT(season) as season FROM scene_exceptions WHERE indexer_id = ?",
[indexer_id])
if sqlResults:
exceptionsSeasonList = list(set([int(x["season"]) for x in sqlResults]))
if not indexer_id in exceptionsSeasonCache:
exceptionsSeasonCache[indexer_id] = {}
exceptionsSeasonCache[indexer_id] = exceptionsSeasonList
else:
exceptionsSeasonList = exceptionsSeasonCache[indexer_id]
return exceptionsSeasonList
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
def get_scene_exception_by_name(show_name):
return get_scene_exception_by_name_multiple(show_name)[0]
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
def get_scene_exception_by_name_multiple(show_name):
"""
Given a show name, return the indexerid of the exception, None if no exception
is present.
"""
# try the obvious case first
myDB = db.DBConnection('cache.db')
exception_result = myDB.select(
"SELECT indexer_id, season FROM scene_exceptions WHERE LOWER(show_name) = ? ORDER BY season ASC",
[show_name.lower()])
if exception_result:
return [(int(x["indexer_id"]), int(x["season"])) for x in exception_result]
out = []
all_exception_results = myDB.select("SELECT show_name, indexer_id, season FROM scene_exceptions")
for cur_exception in all_exception_results:
cur_exception_name = cur_exception["show_name"]
cur_indexer_id = int(cur_exception["indexer_id"])
cur_season = int(cur_exception["season"])
if show_name.lower() in (
cur_exception_name.lower(),
sickbeard.helpers.sanitizeSceneName(cur_exception_name).lower().replace('.', ' ')):
logger.log(u"Scene exception lookup got indexer id " + str(cur_indexer_id) + u", using that", logger.DEBUG)
out.append((cur_indexer_id, cur_season))
if out:
return out
2014-11-22 10:13:48 -05:00
return [(None, None)]
def retrieve_exceptions():
"""
Looks up the exceptions on github, parses them into a dict, and inserts them into the
scene_exceptions table in cache.db. Also clears the scene name cache.
"""
global exception_dict, anidb_exception_dict, xem_exception_dict
# exceptions are stored on github pages
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
for indexer in sickbeard.indexerApi().indexers:
if shouldRefresh(sickbeard.indexerApi(indexer).name):
logger.log(u"Checking for scene exception updates for " + sickbeard.indexerApi(indexer).name + "")
url = sickbeard.indexerApi(indexer).config['scene_url']
url_data = helpers.getURL(url)
if url_data is None:
# When urlData is None, trouble connecting to github
logger.log(u"Check scene exceptions update failed. Unable to get URL: " + url, logger.ERROR)
continue
setLastRefresh(sickbeard.indexerApi(indexer).name)
# each exception is on one line with the format indexer_id: 'show name 1', 'show name 2', etc
for cur_line in url_data.splitlines():
indexer_id, sep, aliases = cur_line.partition(':') # @UnusedVariable
if not aliases:
continue
indexer_id = int(indexer_id)
# regex out the list of shows, taking \' into account
# alias_list = [re.sub(r'\\(.)', r'\1', x) for x in re.findall(r"'(.*?)(?<!\\)',?", aliases)]
alias_list = [{re.sub(r'\\(.)', r'\1', x): -1} for x in re.findall(r"'(.*?)(?<!\\)',?", aliases)]
exception_dict[indexer_id] = alias_list
del alias_list
# cleanup
del url_data
# XEM scene exceptions
_xem_exceptions_fetcher()
for xem_ex in xem_exception_dict:
if xem_ex in exception_dict:
exception_dict[xem_ex] = exception_dict[xem_ex] + xem_exception_dict[xem_ex]
else:
exception_dict[xem_ex] = xem_exception_dict[xem_ex]
# AniDB scene exceptions
_anidb_exceptions_fetcher()
for anidb_ex in anidb_exception_dict:
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
if anidb_ex in exception_dict:
exception_dict[anidb_ex] = exception_dict[anidb_ex] + anidb_exception_dict[anidb_ex]
else:
exception_dict[anidb_ex] = anidb_exception_dict[anidb_ex]
changed_exceptions = False
# write all the exceptions we got off the net into the database
myDB = db.DBConnection('cache.db')
for cur_indexer_id in exception_dict:
# get a list of the existing exceptions for this ID
existing_exceptions = [x["show_name"] for x in
myDB.select("SELECT * FROM scene_exceptions WHERE indexer_id = ?", [cur_indexer_id])]
if not cur_indexer_id in exception_dict:
continue
for cur_exception_dict in exception_dict[cur_indexer_id]:
cur_exception, curSeason = cur_exception_dict.items()[0]
# if this exception isn't already in the DB then add it
if cur_exception not in existing_exceptions:
myDB.action("INSERT INTO scene_exceptions (indexer_id, show_name, season) VALUES (?,?,?)",
[cur_indexer_id, cur_exception, curSeason])
changed_exceptions = True
# since this could invalidate the results of the cache we clear it out after updating
if changed_exceptions:
logger.log(u"Updated scene exceptions")
else:
logger.log(u"No scene exceptions update needed")
# cleanup
exception_dict.clear()
anidb_exception_dict.clear()
xem_exception_dict.clear()
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
def update_scene_exceptions(indexer_id, scene_exceptions, season=-1):
"""
Given a indexer_id, and a list of all show scene exceptions, update the db.
"""
global exceptionsCache
myDB = db.DBConnection('cache.db')
myDB.action('DELETE FROM scene_exceptions WHERE indexer_id=? and season=?', [indexer_id, season])
logger.log(u"Updating scene exceptions", logger.INFO)
# A change has been made to the scene exception list. Let's clear the cache, to make this visible
if indexer_id in exceptionsCache:
exceptionsCache[indexer_id] = {}
exceptionsCache[indexer_id][season] = scene_exceptions
for cur_exception in scene_exceptions:
myDB.action("INSERT INTO scene_exceptions (indexer_id, show_name, season) VALUES (?,?,?)",
[indexer_id, cur_exception, season])
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
def _anidb_exceptions_fetcher():
global anidb_exception_dict
if shouldRefresh('anidb'):
logger.log(u"Checking for scene exception updates for AniDB")
for show in sickbeard.showList:
if show.is_anime and show.indexer == 1:
try:
anime = adba.Anime(None, name=show.name, tvdbid=show.indexerid, autoCorrectName=True)
except:
continue
else:
if anime.name and anime.name != show.name:
anidb_exception_dict[show.indexerid] = [{anime.name: -1}]
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
setLastRefresh('anidb')
return anidb_exception_dict
def _xem_exceptions_fetcher():
global xem_exception_dict
if shouldRefresh('xem'):
for indexer in sickbeard.indexerApi().indexers:
logger.log(u"Checking for XEM scene exception updates for " + sickbeard.indexerApi(indexer).name)
url = "http://thexem.de/map/allNames?origin=%s&seasonNumbers=1" % sickbeard.indexerApi(indexer).config[
'xem_origin']
parsedJSON = helpers.getURL(url, json=True)
if not parsedJSON:
logger.log(u"Check scene exceptions update failed for " + sickbeard.indexerApi(
indexer).name + ", Unable to get URL: " + url, logger.ERROR)
continue
if parsedJSON['result'] == 'failure':
continue
for indexerid, names in parsedJSON['data'].items():
xem_exception_dict[int(indexerid)] = names
Fixed issues with editing/saving custom scene exceptions. Fixed charmap issues for anime show names. Fixed issues with display show page and epCat key errors. Fixed duplicate log messages for clearing provider caches. Fixed issues with email notifier ep names not properly being encoded to UTF-8. TVDB<->TVRAGE Indexer ID mapping is now performed on demand to be used when needed such as newznab providers can be searched with tvrage_id's and some will return tvrage_id's that later can be used to create show objects from for faster and more accurate name parsing, mapping is done via Trakt API calls. Added stop event signals to schedualed tasks, SR now waits indefinate till task has been fully stopped before completing a restart or shutdown event. NameParserCache is now persistent and stores 200 parsed results at any given time for quicker lookups and better performance, this helps maintain results between updates or shutdown/startup events. Black and White lists for anime now only get used for anime shows as intended, performance gain for non-anime shows that dont need to load these lists. Internal name cache now builds it self on demand when needed per show request plus checks if show is already in cache and if true exits routine to save time. Schedualer and QueueItems classes are now a sub-class of threading.Thread and a stop threading event signal has been added to each. If I forgot to list something it doesn't mean its not fixed so please test and report back if anything is wrong or has been corrected by this new release.
2014-07-14 22:00:53 -04:00
setLastRefresh('xem')
return xem_exception_dict
def getSceneSeasons(indexer_id):
"""get a list of season numbers that have scene exceptions
"""
myDB = db.DBConnection('cache.db')
seasons = myDB.select("SELECT DISTINCT season FROM scene_exceptions WHERE indexer_id = ?", [indexer_id])
return [cur_exception["season"] for cur_exception in seasons]