Design YouTube: video upload, transcoding, and CDN
Youtube
Guided ModeSeparate Binary Storage with Blob Storage
Move video bytes out of the database and into a dedicated S3 Storage. The database now stores only metadata (title, URL, duration) β never file bytes.
Request Flow
API uploads to S3
Uploader POSTs to the API, which streams bytes to S3 Storage. Better than the database, but the API still handles every byte.
DB stores metadata only
The Database stores a lightweight row: title, s3_key, duration, uploader_id, status. No binary blobs β rows are tiny and queries are fast.
Viewer fetches from S3
GET /watch returns the S3 URL. The viewer's browser or player fetches bytes directly from S3 Storage. Better β but the API still mediates uploads.
Why We Need This
S3 stores petabytes at $0.023/GB/month with 99.999999999% durability. A relational DB costs 100Γ more per GB and wasn't designed for streaming binary data.
Key Insight
The database should only store the S3 key (a string like 'videos/abc123.mp4'), never the bytes. Separating metadata from binary data is the foundation of every scalable media system.
Step 2 of 9
