|
- from threading import Thread
- import requests
- from time import sleep
- from os import mkdir, rmdir, listdir, system, environ
- from os.path import isdir, isfile, getsize
- from json import dumps, loads
-
- import signal
-
- from youtube_dl.utils import DownloadError
-
- import tracker
-
- from youtube_dl import YoutubeDL
-
- from shutil import make_archive, rmtree
-
- from queue import Queue
-
- from gc import collect
-
- from discovery import getmetadata
- from export import subprrun
-
- batchcontent = []
-
- def batchfunc():
- ydl = YoutubeDL({"extract_flat": "in_playlist", "simulate": True, "skip_download": True, "quiet": True, "cookiefile": "cookies.txt", "source_address": "0.0.0.0", "call_home": False})
- while jobs.qsize() < 501:
- desit = tracker.request_item_from_tracker()
- if desit:
- if desit.split(":", 1)[0] == "video":
- jobs.put(desit.split(":", 1)[1])
- elif desit.split(":", 1)[0] == "channel":
- y = ydl.extract_info("https://www.youtube.com/channel/"+desit.split(":", 1)[1], download=False)
- for itemyv in y["entries"]:
- tracker.add_item_to_tracker(tracker.ItemType.Video, itemyv["id"])
- elif desit.split(":", 1)[0] == "playlist":
- y = ydl.extract_info("https://www.youtube.com/playlist?list="+desit.split(":", 1)[1], download=False)
- for itemyvp in y["entries"]:
- tracker.add_item_to_tracker(tracker.ItemType.Video, itemyvp["id"])
- else:
- print("Ignoring item for now", desit)
- else:
- print("Ignoring item for now", desit)
-
- batchcontent.append(desit.split(":", 1)[1])
-
- def submitfunc(submitqueue):
- while not submitqueue.empty():
- itype, ival = submitqueue.get()
- tracker.add_item_to_tracker(itype, ival)
-
- langs = ['ab', 'aa', 'af', 'sq', 'ase', 'am', 'ar', 'arc', 'hy', 'as', 'ay', 'az', 'bn', 'ba', 'eu', 'be', 'bh', 'bi', 'bs', 'br',
- 'bg', 'yue', 'yue-HK', 'ca', 'chr', 'zh-CN', 'zh-HK', 'zh-Hans', 'zh-SG', 'zh-TW', 'zh-Hant', 'cho', 'co', 'hr', 'cs', 'da', 'nl',
- 'nl-BE', 'nl-NL', 'dz', 'en', 'en-CA', 'en-IN', 'en-IE', 'en-GB', 'en-US', 'eo', 'et', 'fo', 'fj', 'fil', 'fi', 'fr', 'fr-BE',
- 'fr-CA', 'fr-FR', 'fr-CH', 'ff', 'gl', 'ka', 'de', 'de-AT', 'de-DE', 'de-CH', 'el', 'kl', 'gn', 'gu', 'ht', 'hak', 'hak-TW', 'ha',
- 'iw', 'hi', 'hi-Latn', 'ho', 'hu', 'is', 'ig', 'id', 'ia', 'ie', 'iu', 'ik', 'ga', 'it', 'ja', 'jv', 'kn', 'ks', 'kk', 'km', 'rw',
- 'tlh', 'ko', 'ku', 'ky', 'lo', 'la', 'lv', 'ln', 'lt', 'lb', 'mk', 'mg', 'ms', 'ml', 'mt', 'mni', 'mi', 'mr', 'mas', 'nan',
- 'nan-TW', 'lus', 'mo', 'mn', 'my', 'na', 'nv', 'ne', 'no', 'oc', 'or', 'om', 'ps', 'fa', 'fa-AF', 'fa-IR', 'pl', 'pt', 'pt-BR',
- 'pt-PT', 'pa', 'qu', 'ro', 'rm', 'rn', 'ru', 'ru-Latn', 'sm', 'sg', 'sa', 'sc', 'gd', 'sr', 'sr-Cyrl', 'sr-Latn', 'sh', 'sdp', 'sn',
- 'scn', 'sd', 'si', 'sk', 'sl', 'so', 'st', 'es', 'es-419', 'es-MX', 'es-ES', 'es-US', 'su', 'sw', 'ss', 'sv', 'tl', 'tg', 'ta',
- 'tt', 'te', 'th', 'bo', 'ti', 'tpi', 'to', 'ts', 'tn', 'tr', 'tk', 'tw', 'uk', 'ur', 'uz', 'vi', 'vo', 'vor', 'cy', 'fy', 'wo',
- 'xh', 'yi', 'yo', 'zu']
-
- #useful Queue example: https://stackoverflow.com/a/54658363
- jobs = Queue()
-
- ccenabledl = []
-
- recvids = set()
- recchans = set()
- recmixes = set()
- recplayl = set()
-
- #HSID, SSID, SID cookies required
- if "HSID" in environ.keys() and "SSID" in environ.keys() and "SID" in environ.keys():
- cookies = {"HSID": environ["HSID"], "SSID": environ["SSID"], "SID": environ["SID"]}
- elif isfile("config.json"):
- cookies = loads(open("config.json").read())
- else:
- print("HSID, SSID, and SID cookies from youtube.com are required. Specify in config.json or as environment variables.")
- assert False
- if not (cookies["HSID"] and cookies["SSID"] and cookies["SID"]):
- print("HSID, SSID, and SID cookies from youtube.com are required. Specify in config.json or as environment variables.")
- assert False
-
- mysession = requests.session()
- mysession.headers.update({"cookie": "HSID="+cookies["HSID"]+"; SSID="+cookies["SSID"]+"; SID="+cookies["SID"], "Accept-Language": "en-US",})
-
- validationtest = mysession.get("https://www.youtube.com/timedtext_editor?action_mde_edit_form=1&v=1iNTtHUwvq4&lang=en&bl=vmp&ui=hd&ref=player&tab=captions&o=U")
-
- assert not "accounts.google.com" in validationtest.url, "Please ensure you have correctly specified account cookies."
- assert """<button class="yt-uix-button yt-uix-button-size-default yt-uix-button-default yt-uix-button-has-icon" type="button" onclick=";return false;" id="yt-picker-language-button" data-button-action="yt.www.picker.load" data-button-menu-id="arrow-display" data-picker-key="language" data-picker-position="footer" data-button-toggle="true"><span class="yt-uix-button-icon-wrapper"><span class="yt-uix-button-icon yt-uix-button-icon-footer-language yt-sprite"></span></span><span class="yt-uix-button-content"> <span class="yt-picker-button-label">
- Language:
- </span>
- English
- </span><span class="yt-uix-button-arrow yt-sprite"></span></button>""" in validationtest.text, "Please make sure your YouTube and Google account language is set to English (United States)"
-
- del validationtest
-
- open("cookies.txt", "w").write("""# HTTP Cookie File
- .youtube.com TRUE / FALSE 1663793455 SID [SID]
- .youtube.com TRUE / FALSE 1663793455 HSID [HSID]
- .youtube.com TRUE / TRUE 1663793455 SSID [SSID]""".replace("[SID]", cookies["SID"]).replace("[HSID]", cookies["HSID"]).replace("[SSID]", cookies["SSID"]))
-
- del cookies
-
- #Graceful Shutdown
- class GracefulKiller:
- kill_now = False
- def __init__(self):
- signal.signal(signal.SIGINT, self.exit_gracefully)
- signal.signal(signal.SIGTERM, self.exit_gracefully)
-
- def exit_gracefully(self,signum, frame):
- self.kill_now = True
-
- gkiller = GracefulKiller()
-
- def prrun():
- while not jobs.empty():
- global recvids
- global recchans
- global recmixes
- global recplayl
- global ccenabledl
-
- item = jobs.get()
-
- print("Video ID:", str(item).strip())
- while True:
- try:
- info = getmetadata(str(item).strip())
- break
- except BaseException as e:
- print(e)
- print("Error in retrieving information, waiting 30 seconds")
- #raise
- sleep(30)
-
- # Add any discovered videos
- recvids.update(info[2])
- recchans.update(info[3])
- recmixes.update(info[4])
- recplayl.update(info[5])
-
- if info[0] or info[1]: # ccenabled or creditdata
- if not isdir("out/"+str(item).strip()):
- mkdir("out/"+str(item).strip())
-
- if info[1]: # creditdata
- open("out/"+str(item).strip()+"/"+str(item).strip()+"_published_credits.json", "w").write(dumps(info[1]))
-
- if info[0]: #ccenabled
- ccenabledl.append(item)
- jobs.task_done()
-
- return True
-
- while not gkiller.kill_now:
- collect() #cleanup
-
- try:
- mkdir("out")
- except:
- pass
-
- try:
- mkdir("directory")
- except:
- pass
-
- batchcontent.clear()
-
- # Get a batch ID
- batchthreads = []
-
- for r in range(50):
- batchrunthread = Thread(target=batchfunc)
- batchrunthread.start()
- batchthreads.append(batchrunthread)
- del batchrunthread
-
- for xc in batchthreads:
- xc.join() #bug (occurred once: the script ended before the last thread finished)
- batchthreads.remove(xc)
- del xc
-
-
-
- #for ir in range(501):
- # batchcontent.append(tracker.request_item_from_tracker())
-
-
-
- threads = []
-
- for i in range(50):
- runthread = Thread(target=prrun)
- runthread.start()
- threads.append(runthread)
- del runthread
-
- for x in threads:
- x.join()
- threads.remove(x)
- del x
-
- print("Sending discoveries to tracker...")
-
- submitjobs = Queue()
-
- #IDK how to handle mixes so just send them for now
- print("Videos:", len(recvids))
- for itemvid in recvids:
- submitjobs.put((tracker.ItemType.Video, itemvid))
-
- print("Channels:", len(recchans))
- for itemchan in recchans:
- submitjobs.put((tracker.ItemType.Channel, itemchan))
-
- print("Mix Playlists:", len(recmixes))
- for itemmix in recmixes:
- submitjobs.put((tracker.ItemType.MixPlaylist, itemmix))
-
- print("Playlists:", len(recplayl))
- for itemplayl in recplayl:
- submitjobs.put((tracker.ItemType.Playlist, itemplayl))
-
- #open("out/discoveries.json", "w").write(dumps({"recvids": sorted(recvids), "recchans": sorted(recchans), "recmixes": sorted(recmixes), "recplayl": sorted(recplayl)}))
- #clear
- recvids.clear()
- recchans.clear()
- recmixes.clear()
- recplayl.clear()
-
- submitthreads = []
-
- for r in range(50):
- submitrunthread = Thread(target=submitfunc, args=(submitjobs,))
- submitrunthread.start()
- submitthreads.append(submitrunthread)
- del submitrunthread
-
- for xb in submitthreads:
- xb.join() #bug (occurred once: the script ended before the last thread finished)
- submitthreads.remove(xb)
- del xb
-
- sleep(1)
-
-
- subtjobs = Queue()
- while ccenabledl:
- langcontent = langs.copy()
- intvid = ccenabledl.pop(0)
-
- while langcontent:
- subtjobs.put((langcontent.pop(0), intvid, "default"))
- del intvid
- del langcontent
-
- subthreads = []
-
- for r in range(50):
- subrunthread = Thread(target=subprrun, args=(subtjobs,mysession))
- subrunthread.start()
- subthreads.append(subrunthread)
- del subrunthread
-
- for xa in subthreads:
- xa.join() #bug (occurred once: the script ended before the last thread finished)
- subthreads.remove(xa)
- del xa
-
- sleep(1) #wait a second to hopefully allow the other threads to finish
-
- for fol in listdir("out"): #remove extra folders
- try:
- if isdir("out/"+fol):
- rmdir("out/"+fol)
- except:
- pass
-
- #https://stackoverflow.com/a/11968881
-
- # TODO: put the data somewhere...
- # TODO: put the discoveries somewhere...
-
- for fol in listdir("out"):
- if isdir("out/"+fol):
- make_archive("directory/"+fol, "zip", "out/"+fol) #check this
-
- targetloc = None
- while not targetloc:
- targetloc = tracker.request_upload_target()
- if targetloc:
- break
- else:
- print("Waiting 5 minutes...")
- sleep(300)
-
- if targetloc.startswith("rsync"):
- system("rsync -rltv --timeout=300 --contimeout=300 --progress --bwlimit 0 --recursive --partial --partial-dir .rsync-tmp --min-size 1 --no-compress --compress-level 0 --files-from=- directory/ "+targetloc)
- elif targetloc.startswith("http"):
- for filzip in listdir("directory"):
- if filzip.endswith(".zip"):
- system("curl --data-binary @directory/"+filzip+" "+targetloc)
-
- # Report the batch as complete
- for itemb in batchcontent:
- if isfile("directory/"+itemb.split(":", 1)[1]+".zip"):
- size = getsize("directory/"+itemb.split(":", 1)[1]+".zip")
- else:
- size = 0
- tracker.mark_item_as_done(itemb, size)
-
- # clear the output directory
- rmtree("out")
- rmtree("directory")
|