Enhanced FFmpeg Integration:

Added robust error handling and logging
Improved binary verification and initialization
Added proper GPU detection and hardware acceleration
Optimized encoding parameters for different content types
Improved File Operations:

Added retry mechanisms for file operations
Enhanced temporary directory management
Improved cleanup of failed downloads
Added proper permission handling
Enhanced Queue Management:

Fixed queue manager initialization
Added better error recovery
Improved status tracking and logging
Enhanced cleanup of failed items
Better Error Handling:

Added comprehensive exception hierarchy
Improved error logging and reporting
Added fallback mechanisms for failures
Enhanced error recovery strategies
This commit is contained in:
pacnpal
2024-11-15 03:21:25 +00:00
parent a04c576e0a
commit 8503fc6fdd
13 changed files with 1336 additions and 376 deletions

View File

@@ -33,16 +33,26 @@ class VideoDownloader:
enabled_sites: Optional[List[str]] = None,
concurrent_downloads: int = 3,
):
self.download_path = download_path
# Ensure download path exists with proper permissions
self.download_path = Path(download_path)
self.download_path.mkdir(parents=True, exist_ok=True)
# Ensure directory has rwx permissions for user and rx for group/others
os.chmod(str(self.download_path), 0o755)
logger.info(f"Initialized download directory: {self.download_path}")
self.video_format = video_format
self.max_quality = max_quality
self.max_file_size = max_file_size
self.enabled_sites = enabled_sites
self.url_patterns = self._get_url_patterns()
# Initialize FFmpeg manager
self.ffmpeg_mgr = FFmpegManager()
ffmpeg_path = self.ffmpeg_mgr.get_ffmpeg_path()
if not os.path.exists(ffmpeg_path):
raise FileNotFoundError(f"FFmpeg not found at {ffmpeg_path}")
logger.info(f"Using FFmpeg from: {ffmpeg_path}")
# Create thread pool for this instance
self.download_pool = ThreadPoolExecutor(
max_workers=max(1, min(5, concurrent_downloads)),
@@ -53,13 +63,14 @@ class VideoDownloader:
self.active_downloads: Dict[str, str] = {}
self._downloads_lock = asyncio.Lock()
# Configure yt-dlp options
# Configure yt-dlp options with absolute FFmpeg path
self.ydl_opts = {
"format": f"bestvideo[height<={max_quality}]+bestaudio/best[height<={max_quality}]",
"format": f"bv*[height<={max_quality}][ext=mp4]+ba[ext=m4a]/b[height<={max_quality}]/best", # More flexible format
"outtmpl": "%(title)s.%(ext)s", # Base filename only, path added later
"merge_output_format": video_format,
"quiet": True,
"no_warnings": True,
"quiet": False, # Enable output for debugging
"no_warnings": False, # Show warnings
"verbose": True, # Enable verbose output
"extract_flat": False,
"concurrent_fragment_downloads": concurrent_downloads,
"retries": self.MAX_RETRIES,
@@ -68,8 +79,22 @@ class VideoDownloader:
"extractor_retries": self.MAX_RETRIES,
"postprocessor_hooks": [self._check_file_size],
"progress_hooks": [self._progress_hook],
"ffmpeg_location": self.ffmpeg_mgr.get_ffmpeg_path(),
"ffmpeg_location": str(ffmpeg_path), # Convert Path to string
"prefer_ffmpeg": True, # Force use of FFmpeg
"hls_prefer_ffmpeg": True, # Use FFmpeg for HLS
"logger": logger, # Use our logger
"ignoreerrors": True, # Don't stop on download errors
"no_color": True, # Disable ANSI colors in output
"geo_bypass": True, # Try to bypass geo-restrictions
"socket_timeout": 30, # Increase timeout
"external_downloader": {
"m3u8": "ffmpeg", # Use FFmpeg for m3u8 downloads
},
"external_downloader_args": {
"ffmpeg": ["-v", "warning"], # Reduce FFmpeg verbosity
}
}
logger.info("VideoDownloader initialized successfully")
def __del__(self):
"""Ensure thread pool is shutdown and files are cleaned up"""
@@ -83,7 +108,7 @@ class VideoDownloader:
self.active_downloads.clear()
# Shutdown thread pool
if hasattr(self, 'download_pool'):
if hasattr(self, "download_pool"):
self.download_pool.shutdown(wait=True)
except Exception as e:
logger.error(f"Error during VideoDownloader cleanup: {str(e)}")
@@ -94,7 +119,7 @@ class VideoDownloader:
try:
with yt_dlp.YoutubeDL() as ydl:
for ie in ydl._ies:
if hasattr(ie, '_VALID_URL') and ie._VALID_URL:
if hasattr(ie, "_VALID_URL") and ie._VALID_URL:
if not self.enabled_sites or any(
site.lower() in ie.IE_NAME.lower()
for site in self.enabled_sites
@@ -120,21 +145,29 @@ class VideoDownloader:
"""Handle download progress"""
if d["status"] == "finished":
logger.info(f"Download completed: {d['filename']}")
elif d["status"] == "downloading":
try:
percent = d.get("_percent_str", "N/A")
speed = d.get("_speed_str", "N/A")
eta = d.get("_eta_str", "N/A")
logger.debug(f"Download progress: {percent} at {speed}, ETA: {eta}")
except Exception as e:
logger.debug(f"Error logging progress: {str(e)}")
def _verify_video_file(self, file_path: str) -> bool:
"""Verify video file integrity"""
try:
probe = ffmpeg.probe(file_path)
# Check if file has video stream
video_streams = [s for s in probe['streams'] if s['codec_type'] == 'video']
video_streams = [s for s in probe["streams"] if s["codec_type"] == "video"]
if not video_streams:
raise VideoVerificationError("No video streams found")
# Check if duration is valid
duration = float(probe['format'].get('duration', 0))
duration = float(probe["format"].get("duration", 0))
if duration <= 0:
raise VideoVerificationError("Invalid video duration")
# Check if file is readable
with open(file_path, 'rb') as f:
with open(file_path, "rb") as f:
f.seek(0, 2) # Seek to end
if f.tell() == 0:
raise VideoVerificationError("Empty file")
@@ -148,12 +181,11 @@ class VideoDownloader:
for attempt in range(self.MAX_RETRIES):
try:
ydl_opts = self.ydl_opts.copy()
ydl_opts['outtmpl'] = os.path.join(temp_dir, ydl_opts['outtmpl'])
ydl_opts["outtmpl"] = os.path.join(temp_dir, ydl_opts["outtmpl"])
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
info = await asyncio.get_event_loop().run_in_executor(
self.download_pool,
lambda: ydl.extract_info(url, download=True)
self.download_pool, lambda: ydl.extract_info(url, download=True)
)
if info is None:
@@ -171,7 +203,9 @@ class VideoDownloader:
except Exception as e:
logger.error(f"Download attempt {attempt + 1} failed: {str(e)}")
if attempt < self.MAX_RETRIES - 1:
await asyncio.sleep(self.RETRY_DELAY * (attempt + 1)) # Exponential backoff
await asyncio.sleep(
self.RETRY_DELAY * (attempt + 1)
) # Exponential backoff
else:
return False, "", f"All download attempts failed: {str(e)}"
@@ -206,7 +240,7 @@ class VideoDownloader:
)
compressed_file = os.path.join(
self.download_path,
f"compressed_{os.path.basename(original_file)}"
f"compressed_{os.path.basename(original_file)}",
)
# Configure ffmpeg with optimal parameters
@@ -225,11 +259,15 @@ class VideoDownloader:
)
if not os.path.exists(compressed_file):
raise FileNotFoundError("Compression completed but file not found")
raise FileNotFoundError(
"Compression completed but file not found"
)
# Verify compressed file
if not self._verify_video_file(compressed_file):
raise VideoVerificationError("Compressed file is not a valid video")
raise VideoVerificationError(
"Compressed file is not a valid video"
)
compressed_size = os.path.getsize(compressed_file)
if compressed_size <= (self.max_file_size * 1024 * 1024):
@@ -245,7 +283,9 @@ class VideoDownloader:
return False, "", f"Compression error: {str(e)}"
else:
# Move file to final location
final_path = os.path.join(self.download_path, os.path.basename(original_file))
final_path = os.path.join(
self.download_path, os.path.basename(original_file)
)
# Use safe move with retries
success = await self._safe_move_file(original_file, final_path)
if not success:
@@ -264,7 +304,11 @@ class VideoDownloader:
try:
if original_file and os.path.exists(original_file):
await self._safe_delete_file(original_file)
if compressed_file and os.path.exists(compressed_file) and not compressed_file.startswith(self.download_path):
if (
compressed_file
and os.path.exists(compressed_file)
and not compressed_file.startswith(self.download_path)
):
await self._safe_delete_file(compressed_file)
except Exception as e:
logger.error(f"Error during file cleanup: {str(e)}")