For much better scaling with large datasets please run your YoutubeDL-Material instance with MongoDB backend rather than the json file-based default. It will fix a lot of performance problems (especially with datasets in the tens of thousands videos/audios)!