Editing Openai/694ab010-3e7c-8000-b0e8-b98c22b7a063 (section)

==== There are two ECS “compute” modes: ====
* ECS on EC2: you manage worker EC2 instances inside your cluster
* ECS on Fargate: AWS runs the worker machines for you (simpler)

For a beginner: use Fargate.

===== AWS Console: =====
* ECS → Clusters → Create cluster
* “Networking only (Fargate)” or equivalent
* Name: ml-chatbot-cluster

===== ECS → Task Definitions → Create =====
* Launch type: Fargate
* CPU/Memory: start small (e.g., 0.25 vCPU / 0.5GB or 1GB)
* Container: - Image URI: 123456789012.dkr.ecr.us-east-1.amazonaws.com/ml-chatbot:latest - Container port: 8501 - Environment variables: - OPENAI_API_KEY = your key (for demo)

Important security note: For real deployments, you should NOT store secrets in plain env vars in the task definition. Use AWS Secrets Manager. For learning, env var is acceptable.

===== In the cluster: =====
* Create Service
* Launch type: Fargate
* Task definition: select your task def
* Desired tasks: 1
* Networking: - VPC: default is fine for demo - Subnets: 2+ subnets recommended - Security group inbound: - allow port 8501 from your IP (or from ALB if you attach one)

===== If you expose Fargate tasks directly, you end up with awkward networking and changing IPs. The standard is: =====

ALB (public) → Target Group → ECS Service tasks

In the service creation wizard, enable Load Balancer:
* Create ALB
* Listener: HTTP 80 (or HTTPS 443 later)
* Target group: port 8501
* Health check path: / (Streamlit returns 200 on /)

Then you access:

<syntaxhighlight>http://ALB_DNS_NAME

</syntaxhighlight>

===== ECS → Tasks → select running task → Logs =====
If you configured CloudWatch logs in the task definition, you will see Streamlit output there.