Cloudflare Containers Beta: Docker Edge Workers Auto-scale Pay-per-running

Cloudflare Containers Public Beta 2025: รัน Docker Edge 320+ Cities Auto-scale

Cloudflare Containers รันคอนเทนเนอร์ทั่วโลกผ่าน Workers + Durable Objects Latency ต่ำสุด Sandbox per user Media processing AI inference Pay-per-running-time

Cloudflare Containers vs Competition

Core Architecture

Request → Workers → Durable Objects → Containers
                        ↓
              Auto-scale + Health checks

คุณสมบัติหลัก

🌍 Global deployment (320+ cities)
⚙️ Workers orchestration
💤 Sleep after idle (pay-per-use)
🔒 Per-container sandboxing
🌐 HTTP/WebSocket support
📈 Auto-scaling (CPU/memory)

Quick Deploy Example

// wrangler.toml
[[containers]]
name = "ffmpeg-converter"
image = "myregistry/ffmpeg:latest”
port = 8080
sleep_after = "1m”
min_instances = 1

// worker.js  
export default {
  async fetch(request, env) {
    const container = env.FFMPEG.get(env.FFMPEG.idFromName("user123"));
    return container.fetch(request);
  }
}

npx wrangler deploy

Real-world Use Cases

1. Code Sandbox (Per User)

User1 → Container1 (Node.js)
User2 → Container2 (Python)
User3 → Container3 (Go)

2. FFmpeg Media Processing

Upload MP4 → Edge container → GIF output
Latency: 150ms (vs 2s centralized)

3. AI Model Inference

User request → Boot Llama.cpp → Inference → Sleep
Cost: $0.0001 per request

Pricing Model (Pay-per-running)

Running: $0.000015/GB-second
Sleep: $0 (free)
Boot time: 200ms (billed)
Min 1 instance: Configurable

ตัวอย่าง: 1000 req/day × 30s = $0.45/เดือน

Production Config

# wrangler.toml
[[containers]]
name = "app"
image = "docker.io/myapp:latest"
port = 3000
min_instances = 2
max_instances = 50
sleep_after = "5m"
cpu_threshold = 0.75
memory_threshold = 0.8

Advanced Patterns

Service Mesh via Workers

// Load balance 10 containers
const containers = Array(10).fill().map((_, i) => 
  env.APP.get(env.APP.idFromName(`inst-${i}`))
);
const container = containers[Math.floor(Math.random() * 10)];
return container.fetch(request);

Health Checking

export class ContainerProxy {
  async fetch(request) {
    try {
      const res = await this.container.fetch('/health');
      if (!res.ok) throw new Error('Unhealthy');
      return this.container.fetch(request);
    } catch {
      // Spin up new container
      await this.restart();
    }
  }
}

Limitations

❌ No TCP/UDP (HTTP/WebSocket only)
❌ Cloudflare network only
⚠️ Cold start 200ms (acceptable)
✅ Perfect for stateless workloads

Migration from Other Platforms

Docker → Cloudflare: 5 mins setup
K8s → Containers: No YAML hell
Lambda → Containers: Full runtime
Vercel → Edge Containers: Global