zzhang2816/CS_TA

SecureMed — EC2 deployment

This package runs a FastAPI website/API and the supplied securemed-diagnosis:1 Docker image on one EC2 instance. SQLite records live in the securemed_data Docker volume. The API is public on HTTP port 80; the diagnosis container has no host port. Cognito is the only external identity service.

1. AWS and Cognito prerequisites

Use an EC2 Linux instance with Docker Engine, Docker Compose v2, Python 3, and sufficient disk space for the supplied ~1 GB diagnosis image. Allow inbound TCP 80 from the grading network. Restrict TCP 22 to your own IP. Do not open 8080 or a database port. Attach an IAM role that permits Cognito user administration (cognito-idp:DescribeUserPoolClient, CreateGroup, AdminCreateUser, AdminSetUserPassword, AdminAddUserToGroup, AdminRemoveUserFromGroup, AdminGetUser) or configure AWS CLI credentials privately for the provisioning step.

Create a Cognito user pool in the same region. Configure email sign-in and public self-registration. In its app client:

• Select no client secret.
• Enable ALLOW_USER_PASSWORD_AUTH and refresh token auth for the three grading accounts.
• Enable OAuth authorization code grant with openid email profile scopes and Cognito plus Google identity providers.
• Set the callback URL to the public web URL including trailing slash; set the sign-out URL too. The teaching team can change this URL to its own domain. Google/Cognito hosted login normally requires an HTTPS callback for a non-localhost domain. The API itself may remain available over HTTP as required by the assignment; configure the grader's HTTPS domain or reverse proxy for interactive Google login.
• Disable required MFA and force-change-password behavior for grading accounts. The provisioning script sets permanent passwords.

Create a Google OAuth web client in Google Cloud. Put Cognito's Google redirect URI (https://YOUR_COGNITO_DOMAIN/oauth2/idpresponse) into Google Authorized redirect URIs. Put the Google client ID and secret into Cognito's Google identity provider settings, then enable Google for the app client. Keep the Google secret out of this package.

Cognito must not let public users join the doctor group. Only the administrator provisioning script grants that group. First-time Google and self-registered users are automatically assigned patient records by the API when they call a protected patient endpoint. The doctor group and local doctor mapping are required to use doctor endpoints.

2. Start on EC2

Copy the full directory, including securemed-diagnosis.tar, to the EC2 instance. From this directory:

cp .env.example .env
# Edit all placeholder values in .env. PUBLIC_CALLBACK_URL must exactly match
# the callback registered in the Cognito app client, including trailing slash.
docker load -i securemed-diagnosis.tar
docker compose up -d --build
curl -i http://EC2_PUBLIC_IP/api/health

Expected health response: HTTP 200 and {"status":"ok"}. docker compose ps should show api and diagnosis. For logs use docker compose logs --tail=100 api diagnosis. After reboot, Docker's unless-stopped policy restarts both services; the named volume preserves patient and diagnosis records.

Environment keys are documented in .env.example. COGNITO_DOMAIN is the full https://... Cognito hosted UI domain. PUBLIC_CALLBACK_URL is the website callback URL. The grading URL in submission.json is the site origin plus /api.

3. Create test accounts and records

Run this after the containers start. Install boto3 for the host script, or run it in a Python virtual environment. Use synthetic email addresses accessible to the teaching team if required.

python3 -m venv .venv
.venv/bin/pip install boto3
.venv/bin/python scripts/provision_grading_accounts.py \
  --region ap-southeast-1 \
  --user-pool-id YOUR_POOL_ID \
  --app-client-id YOUR_PUBLIC_CLIENT_ID \
  --base-url http://EC2_PUBLIC_IP \
  --student-id YOUR_STUDENT_ID \
  --doctor-email doctor@example.com \
  --alice-email alice@example.com \
  --bob-email bob@example.com

The script creates Cognito doctor/patient groups, permanent-password Doctor/Alice/Bob users, local patient records P1001 and P1002, and local subject-to-role mappings. It writes non-secret submission.json and mode-0600 private_accounts.json. Every rerun resets all three passwords. Send private_accounts.json only via the private Canvas submission; never put it in a public repository or the public source ZIP.

For a fresh grading deployment, repeat the steps above with the grading team's own Cognito values, callback domain, and instance address. No original EC2 service is needed.

4. Check the deployment

python3 checklist.py --base-url http://EC2_PUBLIC_IP
.venv/bin/python checklist.py --base-url http://EC2_PUBLIC_IP --accounts private_accounts.json
.venv/bin/python checklist.py --base-url http://EC2_PUBLIC_IP \
  --accounts private_accounts.json --scan ../assigment2/disease_images/01_breast_invasive_ductal.jpg

The final command persists test diagnoses for Alice and Bob. Each check prints PASS or FAIL and the script exits nonzero if any check fails. For a submission machine where the sample images are elsewhere, give any provided JPEG/PNG path.

5. API and submission

The web interface is /; all API routes are under /api. Upload image as multipart form data. POST /api/doctor/scan accepts patient_id in the form or query string. Use the Cognito access token as Authorization: Bearer TOKEN. Public self-registration creates patients only; local mappings are keyed by Cognito sub, never by a client-supplied role or patient ID.

Run python3 scripts/package.py to create SecureMed.zip containing the supplied diagnosis tar, source, .env.example, scripts, README, and non-secret submission.json. The script excludes .env, private_accounts.json, virtual environments, and runtime data. Add screenshots of Google login and both doctor and patient workflows to the private Canvas submission.