This document explains the security model used by FastAPI Production Template — including authentication flow, CSRF protection, session handling, and best practices for secure deployments.

It builds on the README's Authentication section and the Configuration Guide.

The template implements a Backend-for-Frontend (BFF) architecture using OpenID Connect (OIDC) with the Authorization Code + PKCE flow.
This ensures your frontend never directly handles tokens, and your backend maintains complete control over authentication and sessions.

1. Login Request
   The client (usually a browser) calls /auth/web/login, optionally specifying a provider.
   The server:

   • Generates state, nonce, and a code_verifier (PKCE).
   • Stores them in Redis or in-memory storage (ephemeral auth session with 10-minute TTL).
   • Creates an auth_session_id cookie for the client.
   • Redirects the user to the provider's authorization endpoint with:

     response_type=code
     client_id=...
     redirect_uri=http://localhost:8000/auth/web/callback
     state=<random>
     nonce=<random>
     code_challenge=<PKCE-hash>
     code_challenge_method=S256
     

2. User Authenticates
   The user logs into the OIDC provider (e.g., Keycloak, Google).

3. Callback Exchange
   The provider redirects the user back to /auth/web/callback with a short-lived authorization code.

   The backend:

   • Validates the returned state against the stored auth session.
   • Validates the client fingerprint to prevent session hijacking.
   • Marks the auth session as "used" (single-use enforcement).
   • Exchanges the code for tokens using the stored code_verifier (PKCE).
   • Verifies the ID token signature, issuer, audience, and nonce.
   • Fetches or parses user claims from the ID token or userinfo endpoint.
   • Provisions or updates the user in the database (JIT provisioning).
   • Issues a secure session cookie (user_session_id) with HttpOnly and Secure flags.
   • Deletes the ephemeral auth session.

4. Authenticated Session
   The browser now carries a secure, HttpOnly session cookie on subsequent requests. Sessions are stored server-side with client fingerprinting for additional security.

5. Logout / Refresh

   • /auth/web/logout (POST): Invalidates the session and clears cookies. Requires CSRF token and Origin validation.
   • /auth/web/refresh (POST): Rotates the session ID and CSRF token. Requires CSRF token and Origin validation.

Sessions are server-managed using secure cookies and Redis (or in-memory fallback) for storage.

1. Auth Session (Temporary, during OIDC flow):

• Duration: 10 minutes (600 seconds)
• Cookie: auth_session_id
• Purpose: Store PKCE verifier, state, nonce during authorization
• Single-use: Marked as "used" after token exchange

2. User Session (Persistent, after authentication):

• Duration: Configurable via app.session_max_age (default: 3600 seconds / 1 hour)
• Cookie: user_session_id
• Purpose: Maintain authenticated user state
• Features: Client fingerprinting, token storage, session rotation

Note: SameSite=Lax is optimal for BFF pattern OAuth flows. Use SameSite=None only if your frontend is on a different domain (requires Secure=true).

Sessions are stored in Redis (with automatic fallback to in-memory storage if Redis is unavailable):

• Auth sessions: Prefix auth:, 10-minute TTL
• User sessions: Prefix user:, configurable TTL (default 1 hour)
• Automatic cleanup: Redis handles TTL expiration automatically
• Session data: Pydantic models serialized as JSON

Storage Features:

• Automatic Redis connection health checking
• Graceful fallback to in-memory storage
• TTL-based expiration (no manual cleanup needed)
• Session listing and pattern matching capabilities

Cross-Site Request Forgery (CSRF) protection is implemented using HMAC-based tokens bound to sessions and Origin header validation.

CSRF tokens are generated using HMAC-SHA256:

token = timestamp:hmac(session_id + ":" + timestamp, csrf_signing_secret)

• Tokens are hour-based (timestamp truncated to hours)
• Maximum age: 12 hours (configurable)
• Constant-time comparison prevents timing attacks

1. Get CSRF token: Call /auth/web/me to receive auth state including csrf_token
2. Include in requests: Add X-CSRF-Token header to state-changing requests (POST, PUT, PATCH, DELETE)
3. Origin validation: Ensure correct Origin header is sent by browser

# Get auth state and CSRF token
CSRF=$(curl -s -b cookies.txt http://localhost:8000/auth/web/me | jq -r '.csrf_token')# Use token in state-changing request
curl -X POST -b cookies.txt \
-H "Origin: http://localhost:8000" \
-H "X-CSRF-Token: $CSRF" \
-H "Content-Type: application/json" \
-d '{"data":"value"}' \
http://localhost:8000/api/v1/resource

The following endpoints require CSRF tokens:

• POST /auth/web/logout - Logout user
• POST /auth/web/refresh - Refresh session

Custom endpoints can add CSRF protection using:

fromsrc.app.api.http.depsimportenforce_origin, require_csrf@router.post("/protected", dependencies=[Depends(enforce_origin), Depends(require_csrf)])asyncdefprotected_endpoint():
return {"message": "CSRF protected"}

• Missing CSRF token → 403 Forbidden: Missing CSRF token header
• Invalid/expired CSRF token → 403 Forbidden: Invalid CSRF token
• Missing or invalid Origin header → 403 Forbidden: Origin not allowed
• Missing session cookie → 401 Unauthorized: No session found

CSRF validation is disabled in development and test environments to simplify testing. Always test with APP_ENVIRONMENT=production before deployment.

Each session is bound to a lightweight client fingerprint to prevent session hijacking and cookie theft attacks.

The fingerprint is derived from:

• User-Agent: Browser and OS identification string
• Client IP: First IP from X-Forwarded-For, X-Real-IP, CF-Connecting-IP, or direct client IP
• Hashing: SHA256 hash of combined components

1. Session Creation: Fingerprint captured during login callback
2. Storage: Stored in UserSession model (client_fingerprint field)
3. Validation: Checked on every authenticated request
4. Comparison: Constant-time comparison to prevent timing attacks

• Stolen cookie protection: Cookies used from different devices/locations are rejected
• Session binding: Ties sessions to specific client contexts
• Defense in depth: Additional layer beyond HttpOnly and Secure cookies

# Fingerprint extraction (from src/app/core/security.py)defextract_client_fingerprint(request: Request) ->str:
"""Extract and hash client fingerprint from request."""user_agent=request.headers.get("user-agent")
# Try forwarded headers (proxy-aware)forwarded_headers= ["x-forwarded-for", "x-real-ip", "cf-connecting-ip"]
client_ip=Noneforheaderinforwarded_headers:
value=request.headers.get(header)
ifvalue:
client_ip=value.split(",")[0].strip()
break# Fallback to direct client IPifnotclient_ipandrequest.client:
client_ip=request.client.hostreturnhash_client_fingerprint(user_agent, client_ip)

• NAT/Proxy networks: Users behind shared IPs may have issues if IP changes
• Mobile networks: Cellular IPs can change frequently
• User-Agent changes: Browser updates invalidate fingerprint

Mitigation: The implementation uses strict matching by default but can be configured for fuzzy matching if needed.

The template supports both session-based (BFF) and JWT-based authentication for different client types.

When a JWT bearer token is provided in the Authorization header:

During the /auth/web/callback flow, ID tokens undergo enhanced validation:

# From auth_bff_enhanced.py callback handlerawaitjwt_verify_service.verify_jwt(
token=tokens.id_token,
expected_nonce=auth_session.nonce, # Single-use nonceexpected_issuer=provider_cfg.issuer,
expected_audience=provider_cfg.client_id,
)

Security guarantees:

• Nonce binding: Prevents ID token replay attacks
• Single-use enforcement: Nonce is deleted after first use
• Signature verification: Ensures token issued by trusted provider
• Audience validation: Confirms token intended for this application

Important: The backend never stores access tokens or ID tokens long-term.

• Auth flow: Tokens are used once during callback then discarded
• User sessions: Only session metadata (and optional refresh tokens) are stored
• Refresh tokens: Disabled by default. When enabled, they are persisted only if oidc.refresh_tokens.persist_in_session_store is true and are subject to a configurable lifetime cap.
• JWKS caching: Public keys cached with configurable TTL to reduce provider requests

# Automatic JWKS fetching and cachingjwks_cache=JWKSCacheInMemory(
max_entries=config.jwt.jwks_cache_max_entries,
ttl_seconds=config.jwt.jwks_cache_ttl_seconds,
)

Features:

• Automatic key rotation detection with forced refresh on kid misses
• Configurable TTL + cache size per environment
• Provider-specific key caching
• Thread-safe operations

Standard OIDC claims are mapped to internal user model:

jwt:
claims:
user_id: "sub"# Subject (unique user ID)email: "email"# Email addressroles: "roles"# User roles/permissionsgroups: "groups"# User groupsscope: "scope"# OAuth scopesname: "name"# Full namepreferred_username: "preferred_username"

When a valid JWT is received from a new user:

1. Identity lookup: Check if user identity exists (by uid or issuer:subject)
2. Create user: If new, create user record with claims data
3. Create identity: Link identity to user with issuer/subject
4. Return user: Authenticate request with provisioned user

This enables zero-touch onboarding for federated authentication.

Rate limiting is implemented via Redis (with in-memory fallback) using sliding window algorithm.

Configure rate limits in config.yaml:

config:
rate_limiter:
requests: 10# Max requests per windowwindow_ms: 5000# Window size in milliseconds (5 seconds)enabled: true # Enable/disable rate limitingper_endpoint: true # Separate limits per endpointper_method: true # Separate limits per HTTP method

Rate limiting uses fastapi-limiter with Redis backend:

fromsrc.app.api.http.middleware.limiterimportget_rate_limiter@router.post("/api/resource")asyncdefcreate_resource(
rate_limit: None=Depends(get_rate_limiter()) # Uses config defaults
):
return {"message": "created"}
# Custom rate limit for specific endpoint@router.post("/api/expensive-operation")asyncdefexpensive_op(
rate_limit: None=Depends(get_rate_limiter(requests=2, window_ms=60000)) # 2 req/min
):
return {"message": "processed"}

When rate limit is exceeded:

HTTP/1.1 429 Too Many RequestsContent-Type: application/json
{
"detail": "Rate limit exceeded"
}

Headers (may be included depending on configuration):

• X-RateLimit-Limit: Maximum requests allowed
• X-RateLimit-Remaining: Remaining requests in current window
• X-RateLimit-Reset: Timestamp when limit resets

Rate limit keys are generated based on:

• Client IP: From X-Forwarded-For, X-Real-IP, or direct connection
• Endpoint: Route path (if per_endpoint=true)
• Method: HTTP verb (if per_method=true)

Example key: ratelimit:127.0.0.1:/api/resource:POST

• Primary: Redis (shared across multiple instances)
• Fallback: In-memory (single instance only, loses state on restart)
• Automatic detection: Falls back gracefully if Redis unavailable

Production Settings:

# Recommended production rate limitsrate_limiter:
requests: 100# 100 requestswindow_ms: 60000# per minuteper_endpoint: trueper_method: true

Authentication Endpoints (stricter limits):

# Login/callback endpointsrequests: 10window_ms: 60000# 10 attempts per minute

Public API Endpoints (more lenient):

# Read-only public datarequests: 1000window_ms: 60000# 1000 reads per minute

• Use local Keycloak only for testing - never use development credentials in production
• Test with HTTPS using self-signed certificates if testing cross-origin flows
• Set environment variables:

  APP_ENVIRONMENT=development
  SESSION_SIGNING_SECRET=dev-secret-change-me
  CSRF_SIGNING_SECRET=dev-csrf-secret-change-me

• Development mode relaxations:
  • CSRF validation disabled
  • Origin validation disabled
  • Secure cookie flag disabled (allows HTTP)
  • Detailed error messages exposed

• Set environment:

  APP_ENVIRONMENT=test

• Use production-like settings but with relaxed validation for automated tests
• Seed test data with known credentials
• Mock OIDC providers to avoid external dependencies

Critical Security Checklist:

1. Environment Configuration:

   APP_ENVIRONMENT=production

2. Generate Strong Secrets:

   # Use infra/secrets/generate_secrets.sh
   ./infra/secrets/generate_secrets.sh --generate-pki
   # Verify secrets
   ./infra/secrets/generate_secrets.sh --verify

3. Use Managed Identity Provider:

   • ✅ Azure AD / Entra ID
   • ✅ Google Workspace
   • ✅ Auth0 / Okta
   • ✅ AWS Cognito
   • ✗ Development Keycloak instance

4. HTTPS Enforcement:

   • Enable HTTPS on all endpoints
   • Set Secure cookie flag (automatic in production)
   • Use TLS 1.2+ with strong cipher suites
   • Implement HSTS headers

5. Secret Management:

   # Use environment variables or secret managementsession_signing_secret: "${SESSION_SIGNING_SECRET}"# From secret managercsrf_signing_secret: "${CSRF_SIGNING_SECRET}"# From secret manager

   • Rotate secrets every 90 days
   • Never commit secrets to version control
   • Use .env files (gitignored) or external secret managers
   • Implement secret rotation procedures

6. CORS Configuration:

   app:
   cors:
   origins:
   - "https://app.yourdomain.com"# Explicit origins onlyallow_credentials: true

   • Never use wildcards (*) with credentials
   • List explicit allowed origins
   • Use https:// URLs only in production

7. Database Security:

   • Use connection pooling with reasonable limits
   • Enable SSL/TLS for database connections
   • Use separate database users with minimal privileges
   • Regular backup and recovery testing

8. Redis Security:

   • Enable password authentication
   • Use TLS for Redis connections in production
   • Set appropriate maxmemory limits
   • Monitor Redis memory usage

9. Logging & Monitoring:

   logging:
   level: "INFO"# Not DEBUG in productionformat: "json"

   • Never log sensitive data (tokens, passwords, PII)
   • Use structured JSON logging
   • Set up alerting for authentication failures
   • Monitor rate limit violations

10. Session Configuration:

    app:
    session_max_age: 3600# 1 hour, adjust based on security requirements

    • Balance security vs user experience
    • Shorter sessions for sensitive operations
    • Implement session refresh for long-lived applications

Regular Tasks:

1. Weekly:

   • Review authentication logs for anomalies
   • Check rate limit violations
   • Monitor failed login attempts

2. Monthly:

   • Rotate CSRF and session signing secrets
   • Review and update CORS origins
   • Audit active user sessions

3. Quarterly:

   • Update dependencies (uv lock --upgrade)
   • Review security advisories
   • Conduct security testing
   • Rotate database passwords

4. Annually:

   • Renew TLS certificates
   • Comprehensive security audit
   • Penetration testing
   • Update security documentation

If secrets are compromised:

1. Immediate: Rotate all affected secrets
2. Invalidate: Clear all active sessions
3. Notify: Inform affected users
4. Investigate: Audit logs for unauthorized access
5. Document: Record incident and response

If session hijacking detected:

1. Terminate: Delete affected session
2. Force re-authentication: Require user to log in again
3. Review: Check fingerprint validation logs
4. Enhance: Consider adding additional security layers

Provided out-of-the-box:

• ✅ OIDC Authorization Code + PKCE flow
• ✅ Nonce validation and single-use enforcement
• ✅ Client fingerprinting for session binding
• ✅ HMAC-based CSRF token generation
• ✅ Origin/Referer validation
• ✅ Secure cookie configuration (HttpOnly, Secure, SameSite)
• ✅ JWT signature verification with JWKS
• ✅ Rate limiting with Redis backend
• ✅ Session rotation on refresh
• ✅ JIT user provisioning
• ✅ Automatic Redis/in-memory fallback

Requires configuration:

• ⚠️ OIDC provider credentials (client_id, client_secret)
• ⚠️ Session signing secret (SESSION_SIGNING_SECRET)
• ⚠️ CSRF signing secret (CSRF_SIGNING_SECRET)
• ⚠️ CORS allowed origins
• ⚠️ Rate limit thresholds
• ⚠️ Session expiration times

You must:

• Generate and rotate secrets regularly
• Configure production OIDC providers (not dev Keycloak)
• Set appropriate CORS origins (no wildcards in production)
• Implement authorization logic (roles, permissions)
• Handle sensitive data appropriately
• Write secure business logic
• Test security configurations
• Review and update dependencies

You should not:

• Store tokens in localStorage (use HttpOnly cookies)
• Expose sensitive data in logs
• Use development credentials in production
• Bypass CSRF protection
• Disable security features without understanding implications

Infrastructure team must:

• Deploy with HTTPS/TLS
• Manage certificate lifecycle
• Secure Redis with password and TLS
• Secure PostgreSQL with SSL and network isolation
• Implement firewall rules
• Set up monitoring and alerting
• Perform regular security audits
• Implement backup and disaster recovery
• Monitor for security incidents
• Rotate infrastructure secrets

Authentication Flow:

• Login redirects to OIDC provider correctly
• Callback validates state parameter
• Invalid state is rejected with 400
• Client fingerprint is validated on callback
• Session cookie is set with correct attributes
• Logout clears session and cookies
• Session refresh rotates session ID

CSRF Protection:

• GET requests work without CSRF token
• POST without CSRF token returns 403
• POST with invalid CSRF token returns 403
• POST with valid CSRF token succeeds
• CSRF token expires after 12 hours
• Origin validation rejects invalid origins

Session Security:

• HttpOnly flag prevents JavaScript access
• Secure flag enforced in production
• Sessions expire after configured time
• Changed fingerprint invalidates session
• Session rotation works on refresh

Rate Limiting:

• Exceeding limit returns 429
• Rate limits reset after window
• Per-endpoint limits work independently
• Per-method limits work independently

Run security-focused tests:

# Unit tests for security functions
uv run pytest tests/unit/app/core/test_security.py -v
# Integration tests for auth flows
uv run pytest tests/integration/test_oidc_keycloak.py -v
# CSRF and Origin validation tests
uv run pytest tests/unit/app/api/test_dependencies.py -v
# Rate limiting tests
uv run pytest tests/unit/app/api/test_ratelimit.py -v

Dependencies:

# Check for known vulnerabilities
uv run pip-audit
# Update dependencies
uv lock --upgrade

Static Analysis:

# Type checking
uv run mypy src/
# Security linting
uv run bandit -r src/

SAST (Static Application Security Testing):

• Consider integrating tools like Snyk, Semgrep, or SonarQube
• Run scans in CI/CD pipeline
• Review and remediate findings regularly

• Configuration Guide - Complete configuration reference
• OIDC Documentation - OIDC setup and provider configuration
• Secrets Management - Generate and manage secrets
• README — Authentication API - API endpoint reference
• JavaScript Client Guide - Frontend integration examples
• Python Client Guide - Backend-to-backend authentication

Key Security Features:

1. 🔐 BFF Pattern: Backend controls all authentication, frontend never handles tokens
2. 🔑 PKCE Flow: Prevents authorization code interception attacks
3. 🎲 Nonce Validation: Single-use tokens prevent replay attacks
4. 🍪 Secure Sessions: HttpOnly, Secure, SameSite cookies with server-side storage
5. 🕵️ Fingerprinting: Client binding prevents session hijacking
6. 🛡️ CSRF Protection: HMAC-based tokens with Origin validation
7. ✅ JWT Verification: Full signature and claims validation with JWKS
8. 🚦 Rate Limiting: Redis-based abuse prevention with sliding windows
9. 📝 JIT Provisioning: Automatic user creation from trusted identity providers
10. 🔄 Session Rotation: Security through session ID rotation on refresh

Security Layers:

┌─────────────────────────────────────────────────────┐
│ Layer 1: Network (HTTPS/TLS, Firewall) │
├─────────────────────────────────────────────────────┤
│ Layer 2: Authentication (OIDC, PKCE, Nonce) │
├─────────────────────────────────────────────────────┤
│ Layer 3: Session Security (Cookies, Fingerprint) │
├─────────────────────────────────────────────────────┤
│ Layer 4: Request Validation (CSRF, Origin) │
├─────────────────────────────────────────────────────┤
│ Layer 5: Authorization (JWT, Roles, Scopes) │
├─────────────────────────────────────────────────────┤
│ Layer 6: Rate Limiting (Redis, Sliding Window) │
├─────────────────────────────────────────────────────┤
│ Layer 7: Monitoring (Logs, Alerts, Audit) │
└─────────────────────────────────────────────────────┘

Configuration Checklist:

• Set APP_ENVIRONMENT=production
• Generate secrets: ./infra/secrets/generate_secrets.sh
• Configure production OIDC provider
• Set explicit CORS origins (no wildcards)
• Enable HTTPS with valid TLS certificates
• Configure Redis with password and TLS
• Set appropriate rate limits
• Configure session expiration
• Set up logging and monitoring
• Test authentication flows
• Review security settings

This template provides defense in depth with multiple security layers. Each layer is designed to fail securely if bypassed, ensuring robust protection for your application.