Design scalable media storage and delivery

Q: Design scalable media storage and delivery

This is a System Design interview question from Meta for Machine Learning Engineer roles. View the full question and solution on PracHub.

Q: How do I approach System Design interview questions?

System Design questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master system design interviews.

Question

System Design: Global UGC Photos and Short Videos Store/Delivery

Context

Design a globally distributed system to store and deliver user-generated photos and short videos. The system must support billions of users, high write/read throughput, multi-region presence, and strong reliability/security guarantees. Assume mobile and web clients; traffic is worldwide with regional peaks and diurnal patterns.

Functional Requirements

Uploads
- Single-part and multipart uploads (for large videos, unstable networks).
Reads/Downloads
- Range requests; CDN acceleration; thumbnails and transcoded variants.
Delete
- Soft delete with retention; hard delete; privacy-compliant erase.
List
- List by user/album with pagination and sort (time, size).
Derived media
- Thumbnail generation for images; transcode videos to multiple bitrates/resolutions.
Search
- Simple search by owner and metadata (filename, tags, mime, created_at).
Signed access URLs
- Time-bounded, scoped URLs for upload and download.
Quotas
- Per-user size/object-count quotas; enforcement on upload.

Non-Functional Requirements (address explicitly)

Durability targets (e.g., 11 9s), availability/SLOs.
Multi-region data layout and replication strategy.
Consistency model (metadata vs. object data) and read-after-write behavior.
CDN integration and cache invalidation.
Security: encryption in transit/at rest, key management.
Access control and authorization.
Abuse prevention and rate limiting.
Cost model and lifecycle policies (tiering, retention, cold storage, deletion).
Background processing semantics (idempotency/exactly-once behavior).
Metadata schema and indexing strategy.
Observability (metrics, tracing, audit logs).
Backfill/migrations approach.
Disaster recovery and regional failover.
Performance targets (p95/p99) and bottlenecks.
Testing and rollout plan.

Deliverables

API sketch (major endpoints for uploads, reads, listing, search, delete, signing).
Component diagram (textual/ASCII description acceptable).
Storage choices (object store vs. block/file; metadata store).
Capacity planning with rough numbers and assumptions.