13. BLOB

---

⭐ Interview Intro

“User content contains large files, so I would separate metadata and actual file storage. Metadata like user ID and file URL goes into the database, while images/videos go into object storage such as S3. Uploads happen through pre-signed URLs to reduce backend load, and large uploads use multipart upload. Downloads are served through CDN for low latency.”

🎯 Interview Keywords

✅ Metadata in DB - Metadata -> SQL/NoSQL DB & Files -> Object Storage ✅ Pre-Signed URL - Avoid routing files through backend to upload & download. ✅ Multipart Upload - large files can be uploaded in chunk. ✅ CDN in front of storage - Usually object storage is fronted by CDN.

Client → CDN → Object Storage

---

📌 Object Storage (Blob Storage)

Storage system for large binary files (BLOBs). Examples:

• Images
• Videos
• PDFs
• Logs
• ML datasets
• Static assets

Core Idea

Metadata → DB
Files → Object Storage

Example:

Post:
- post_id      → DB
- caption      → DB
- image_url    → DB
- actual image → S3

Metadata Service → stores where the file exists

---

🚀 Pre-Signed URL

Avoid sending files through backend. ❌

Client → Backend → Get URL
Client → S3 (Direct Upload)

Backend returns:

• Temporary permission
• Expiry
• Allowed operation

Benefits:

• Lower backend load
• Better scalability
• Faster upload/download

---

📦 Multipart Upload

Used for large files

10 GB
→ Chunk
→ Upload chunks
→ S3 merges

Benefits:

• Parallel upload
• Resume on failure
• Reliable large transfers

---

☁️ Popular Services

Cloud	Service
AWS	Amazon S3
GCP	Google Cloud Storage
Azure	Azure Blob Storage

---

⚙️ Why Object Storage Scales Well

1. Flat Namespace

No real folders. → Stored internally as single unique key

user123/profile/image.jpg

Benefits:

• Fast lookup
• Easy scaling

---

2. Immutable Writes

Objects are not updated in-place.

Allowed:

• Create new version
• Overwrite entire object

Benefits:

• No locking
• No race conditions
• Simpler distributed systems

---

3. Replication / Erasure Coding

Data stored across:

• Multiple servers
• Multiple racks
• Multiple DCs Result:

11 nines durability
99.999999999%

---