Instagram

Overview

Instagram is one of the world’s largest social media platforms, serving over 2 billion monthly active users. Its architecture handles the challenge of generating personalized feeds from billions of posts, processing massive volumes of image and video uploads, and delivering real-time interactions. Originally built as a Django monolith by a team of 3, it has evolved into a sophisticated microservices architecture within the Meta infrastructure.

Scale

Requirements

High-Level Architecture

Loading diagram...

Key Components

Data Model

Loading diagram...

Feed Generation Architecture

Instagram uses a hybrid fanout model for feed generation, combining push and pull strategies:

Fanout-on-Write (Push Model): When a user publishes a post, the Feed Service pushes the post ID into the feed caches of all followers. This pre-computes feeds, enabling instant feed loading.

• Used for users with < 10K followers
• Feed stored in Redis sorted sets (post_id, timestamp)
• Each user’s feed cache holds ~500 most recent post IDs

Fanout-on-Read (Pull Model): For celebrity accounts with millions of followers, fanout-on-write is too expensive. Instead, their posts are fetched on-demand when a follower loads their feed.

• Used for users with > 10K followers
• Merged with pre-computed feed at read time

Ranking: Once candidate posts are assembled, an ML ranking model scores each post based on ~1000 features (user affinity, content type, timeliness, engagement signals) and returns the final ordered feed.

Loading diagram...

Media Processing Pipeline

Every photo uploaded to Instagram goes through a sophisticated processing pipeline:

1. Upload: Client uploads the original image to a staging area
2. Validation: File format, size, and content policy checks
3. Processing: Apply user-selected filters using GPU-accelerated pipelines
4. Resizing: Generate multiple resolutions (thumbnail 150px, low 320px, medium 640px, high 1080px)
5. CDN Distribution: Push processed images to edge locations worldwide
6. Metadata Extraction: EXIF data, location, and computer vision (object detection, alt-text)

For videos, the pipeline also includes transcoding to multiple bitrates for adaptive streaming, generating preview thumbnails, and audio normalization.

Instagram stores images in a custom storage system built on top of Haystack (Meta’s photo storage), which reduces metadata overhead by storing multiple images in a single physical file.

Social Graph with TAO

Instagram uses Meta’s TAO (The Associations and Objects) system for the social graph — a distributed, highly-cached graph data store.

TAO Architecture:

• Objects: Nodes in the graph (users, posts, comments)
• Associations: Edges between nodes (follows, likes, comments-on)
• Caching Layer: Multi-tier cache (L1 per-region, L2 per-datacenter) with >99% hit rate
• Storage: MySQL sharded by object ID as the persistent backend

Query Pattern: TAO optimizes for the most common social queries:

• “Who does User X follow?” — Forward association query
• “Who follows User X?” — Inverse association query
• “Does User X follow User Y?” — Point association query

The cache handles billions of queries per second with median latency under 1ms.

Architectural Tradeoffs

# Learn From a Instagram-like System

You are an expert software architect and teacher. I want to learn from a Instagram-inspired system design (Photo and video sharing at massive scale with real-time feeds) without pretending this is the exact private production architecture.

Use the following reference as a learning model. Keep assumptions visible, avoid unsupported scale claims, and explain the engineering forces behind each design choice.

---

## System Design Reference
import { Mermaid, Scale, ComponentList, ComponentCard, Requirements, Tradeoff } from '@/components/mdx';

## Overview

Instagram is one of the world's largest social media platforms, serving over 2 billion monthly active users. Its architecture handles the challenge of generating personalized feeds from billions of posts, processing massive volumes of image and video uploads, and delivering real-time interactions. Originally built as a Django monolith by a team of 3, it has evolved into a sophisticated microservices architecture within the Meta infrastructure.

## Scale

<Scale items={{
  "Monthly active users": "2B+",
  "Photos uploaded daily": "~100M",
  "Likes per second": "~1M",
  "Media storage": "Exabytes",
}} />

## Requirements

<Requirements
  functional={[
    'Photo and video upload with filters',
    'Personalized news feed (algorithmic + chronological)',
    'Stories (ephemeral 24-hour content)',
    'Direct messaging',
    'Explore/discovery page',
    'Likes, comments, and sharing',
    'User profiles and following system',
  ]}
  nonFunctional={[
    'Sub-second feed generation',
    'Low-latency media delivery globally via CDN',
    'High availability (99.99%)',
    'Eventual consistency for social graph',
    'Support for viral content spikes',
    'Efficient storage of billions of images',
  ]}
/>

## High-Level Architecture

<Mermaid chart={`graph TB
    subgraph Clients
        A[iOS App]
        B[Android App]
        C[Web App]
    end
    subgraph Edge["Edge / CDN"]
    ...

[Full prompt copied to clipboard]