Skip to main content

🗓️ 09092025 1153

VIDEO STORAGE STRATEGY

Core Principle​

NOTE

Different storage layers serve different purposes - don't mix them

Storage Hierarchy​

Transit Storage (Temporary)​

Hold chunks during upload

  • Location: Local filesystem or Redis
  • Lifespan: Minutes to hours
  • Performance: Fast write, sequential read

Persistent Storage (Source of Truth)​

Store assembled original videos

  • Location: OSS/S3
  • Lifespan: Long-term
  • Performance: Reliable, backed up

Delivery Storage (Performance Layer)​

For serving content to users

  • Location: CDN edge locations
  • Lifespan: Based on cache rules
  • Performance: Optimized for read speed

Storage Decisions​

  • Never store large files in database (use file paths instead)
WARNING

🎯 Core Problem​

Databases are optimized for small, structured queries. Video chunks are large, unstructured binary data.

💥 The Big Issues​

  • Performance - 10MB chunk insert: 2-5 seconds vs 0.1 seconds on filesystem
  • Memory usage: Each chunk loads entirely into MySQL buffer pool
  • Query impact: Video operations slow down all other database queries

Cost​

  • Database storage: 10-15x more expensive than S3/OSS
  • Backup explosion: 500MB video = 500MB+ in every database backup
  • Replication cost: Video data replicates to every database replica

Scalability​

  • Size limits: MySQL practical limit much lower than theoretical 4GB
  • Concurrent uploads: Multiple video uploads can crash the database
  • Growth ceiling: Works for 10 users, fails at 1000 users
  • Separate upload from delivery (different performance needs)
  • Plan for cleanup (temporary files accumulate)
  • File Organization

References​