ytsm/app/YtManagerApp/management/downloader.py

112 lines
4.2 KiB
Python

from YtManagerApp.scheduler.jobs.download_video_job import DownloadVideoJob
from YtManagerApp.models import Video, Subscription, VIDEO_ORDER_MAPPING
from YtManagerApp.utils import first_non_null
from django.conf import settings as srv_settings
import logging
import requests
import mimetypes
import os
import PIL.Image
import PIL.ImageOps
from urllib.parse import urljoin
log = logging.getLogger('downloader')
def __get_subscription_config(sub: Subscription):
user = sub.user
enabled = first_non_null(sub.auto_download, user.preferences['auto_download'])
global_limit = user.preferences['download_global_limit']
limit = first_non_null(sub.download_limit, user.preferences['download_subscription_limit'])
order = first_non_null(sub.download_order, user.preferences['download_order'])
order = VIDEO_ORDER_MAPPING[order]
return enabled, global_limit, limit, order
def downloader_process_subscription(sub: Subscription):
log.info('Processing subscription %d [%s %s]', sub.id, sub.playlist_id, sub.id)
enabled, global_limit, limit, order = __get_subscription_config(sub)
log.info('Determined settings enabled=%s global_limit=%d limit=%d order="%s"', enabled, global_limit, limit, order)
if enabled:
videos_to_download = Video.objects\
.filter(subscription=sub, downloaded_path__isnull=True, watched=False)\
.order_by(order)
log.info('%d download candidates.', len(videos_to_download))
if global_limit > 0:
global_downloaded = Video.objects.filter(subscription__user=sub.user, downloaded_path__isnull=False).count()
allowed_count = max(global_limit - global_downloaded, 0)
videos_to_download = videos_to_download[0:allowed_count]
log.info('Global limit is set, can only download up to %d videos.', allowed_count)
if limit > 0:
sub_downloaded = Video.objects.filter(subscription=sub, downloaded_path__isnull=False).count()
allowed_count = max(limit - sub_downloaded, 0)
videos_to_download = videos_to_download[0:allowed_count]
log.info('Limit is set, can only download up to %d videos.', allowed_count)
# enqueue download
for video in videos_to_download:
log.info('Enqueuing video %d [%s %s] index=%d', video.id, video.video_id, video.name, video.playlist_index)
DownloadVideoJob.schedule(video)
log.info('Finished processing subscription %d [%s %s]', sub.id, sub.playlist_id, sub.id)
def downloader_process_all():
for subscription in Subscription.objects.all():
downloader_process_subscription(subscription)
def fetch_thumbnail(url, object_type, identifier, thumb_size):
log.info('Fetching thumbnail url=%s object_type=%s identifier=%s', url, object_type, identifier)
# Make request to obtain mime type
try:
response = requests.get(url, stream=True)
except requests.exceptions.RequestException as e:
log.error('Failed to fetch thumbnail %s. Error: %s', url, e)
return url
ext = mimetypes.guess_extension(response.headers['Content-Type'])
# Build file path
file_name = f"{identifier}{ext}"
abs_path_dir = os.path.join(srv_settings.MEDIA_ROOT, "thumbs", object_type)
abs_path = os.path.join(abs_path_dir, file_name)
abs_path_tmp = file_name + '.tmp'
# Store image
try:
os.makedirs(abs_path_dir, exist_ok=True)
with open(abs_path_tmp, "wb") as f:
for chunk in response.iter_content(chunk_size=1024):
if chunk:
f.write(chunk)
# Resize and crop to thumbnail size
image = PIL.Image.open(abs_path_tmp)
image = PIL.ImageOps.fit(image, thumb_size)
image.save(abs_path)
image.close()
# Delete temp file
os.unlink(abs_path_tmp)
except requests.exceptions.RequestException as e:
log.error('Error while downloading stream for thumbnail %s. Error: %s', url, e)
return url
except OSError as e:
log.error('Error while writing to file %s for thumbnail %s. Error: %s', abs_path, url, e)
return url
# Return
media_url = urljoin(srv_settings.MEDIA_URL, f"thumbs/{object_type}/{file_name}")
return media_url