Container Registries and Versioning
What you'll learn: How to push, version, and manage Docker images in cloud registries so you can reliably deploy AI models and roll back when needed.
Why Registries Matter
Think of a container registry as GitHub for Docker images—a centralized place to store, version, and distribute your containerized AI applications. Instead of manually copying Docker images between machines, you push them once to a registry and pull them anywhere: dev laptops, staging servers, or production clusters.
Core Concepts
A container registry stores your built Docker images. Popular options include:
- Docker Hub (public and private)
- AWS Elastic Container Registry (ECR)
- Google Container Registry (GCR) / Artifact Registry
- Azure Container Registry (ACR)
Versioning uses image tags to track different builds. Just like semantic versioning for code, you tag images with meaningful identifiers:
# Bad: always overwrites
myapp:latest
# Good: semantic versioning
myapp:1.2.3
myapp:1.2.3-gpu
myapp:sha-a3f8b21
Workflow Example
After building your Dockerfile (lesson 1093), you tag and push:
# Tag with version
docker tag my-ai-app:latest myregistry/my-ai-app:v1.0.0
# Push to registry
docker push myregistry/my-ai-app:v1.0.0
# Pull on production server
docker pull myregistry/my-ai-app:v1.0.0
Rollback Strategy
When a deployment breaks, versioned images let you instantly return to the previous working version—just redeploy v1.0.0 instead of v1.1.0. This is critical when model serving goes wrong.
Key Takeaway: Container registries with semantic versioning create a reliable deployment pipeline—push once, pull anywhere, and roll back instantly when issues arise.