aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl')
-rw-r--r--youtube_dl/extractor/douyutv.py35
1 files changed, 25 insertions, 10 deletions
diff --git a/youtube_dl/extractor/douyutv.py b/youtube_dl/extractor/douyutv.py
index 3915cb182..ce6962755 100644
--- a/youtube_dl/extractor/douyutv.py
+++ b/youtube_dl/extractor/douyutv.py
@@ -18,7 +18,7 @@ class DouyuTVIE(InfoExtractor):
'display_id': 'iseven',
'ext': 'flv',
'title': 're:^清晨醒脑!T-ara根本停不下来! [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
- 'description': 'md5:f34981259a03e980a3c6404190a3ed61',
+ 'description': 're:.*m7show@163\.com.*',
'thumbnail': 're:^https?://.*\.jpg$',
'uploader': '7师傅',
'uploader_id': '431925',
@@ -43,7 +43,7 @@ class DouyuTVIE(InfoExtractor):
'params': {
'skip_download': True,
},
- 'skip': 'Romm not found',
+ 'skip': 'Room not found',
}, {
'url': 'http://www.douyutv.com/17732',
'info_dict': {
@@ -51,7 +51,7 @@ class DouyuTVIE(InfoExtractor):
'display_id': '17732',
'ext': 'flv',
'title': 're:^清晨醒脑!T-ara根本停不下来! [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
- 'description': 'md5:f34981259a03e980a3c6404190a3ed61',
+ 'description': 're:.*m7show@163\.com.*',
'thumbnail': 're:^https?://.*\.jpg$',
'uploader': '7师傅',
'uploader_id': '431925',
@@ -75,13 +75,28 @@ class DouyuTVIE(InfoExtractor):
room_id = self._html_search_regex(
r'"room_id"\s*:\s*(\d+),', page, 'room id')
- prefix = 'room/%s?aid=android&client_sys=android&time=%d' % (
- room_id, int(time.time()))
-
- auth = hashlib.md5((prefix + '1231').encode('ascii')).hexdigest()
- config = self._download_json(
- 'http://www.douyutv.com/api/v1/%s&auth=%s' % (prefix, auth),
- video_id)
+ config = None
+ # Douyu API sometimes returns error "Unable to load the requested class: eticket_redis_cache"
+ # Retry with different parameters - same parameters cause same errors
+ for i in range(5):
+ prefix = 'room/%s?aid=android&client_sys=android&time=%d' % (
+ room_id, int(time.time()))
+ auth = hashlib.md5((prefix + '1231').encode('ascii')).hexdigest()
+
+ config_page = self._download_webpage(
+ 'http://www.douyutv.com/api/v1/%s&auth=%s' % (prefix, auth),
+ video_id)
+ try:
+ config = self._parse_json(config_page, video_id, fatal=False)
+ except ExtractorError:
+ # Wait some time before retrying to get a different time() value
+ self._sleep(1, video_id, msg_template='%(video_id)s: Error occurs. '
+ 'Waiting for %(timeout)s seconds before retrying')
+ continue
+ else:
+ break
+ if config is None:
+ raise ExtractorError('Unable to fetch API result')
data = config['data']