Skip to content

EPFL IT Compliance Mapping

This document maps CO₂ Calculator implementation to EPFL IT requirements (RFC-EPFL-IT-001). Use this to verify compliance and identify areas needing approval or exceptions.

Reading time: ~8 minutes

TL;DR

We meet all mandatory (P0) requirements except two: we use PostgreSQL instead of MariaDB/SQL Server (an exception request is needed), and transfer encryption is not enforced end to end. Most recommended (P1) features are implemented.

Status: ⚠️ P0: 8/10 (80%) | ✅ P1: 4/5 (80%) | ⚠️ P2: 1/3 (33%)

⚠️ This page has further stale claims (Azure references, unverified retention figures) that need a separate pass.

Authentication and Authorization (SVC0018)

Requirement Priority Status Notes
Strong Auth P0 ✅ Microsoft Entra ID via OIDC
OIDC/OAuth 2.0+ P0 ✅ Full implementation with PKCE
SAML 2.0+ P1 ⚠️ Not implemented (OIDC sufficient)
SCIM P2 ⚠️ Group sync via Entra ID
No LDAPS P0 ✅ Not used

Implementation: We use Microsoft Entra ID as identity provider with OpenID Connect protocol. JWT tokens provide stateless auth. Authorization uses role-based access control from Entra ID claims.

Code: app/core/security.py (backend), src/plugins/auth.ts (frontend)

Reference: Auth Flow | ADR-012

Infrastructure and Containers

Requirement Priority Status Implementation
IaC Support P1 ✅ Helm + ArgoCD (not Ansible)
Hypervisor Agnostic P0 ✅ OCI containers
OS Patching P0 ✅ Automated base image updates
OCI Compliance P0 ✅ Docker/Podman images

Container Strategy: We use Helm charts and ArgoCD for GitOps deployment. This is more suitable for Kubernetes than Ansible. All images are OCI-compliant and stored in GitHub Container Registry (ghcr.io).

Patching Process:

  1. Dependabot creates PR for base image updates
  2. CI runs security scans (Trivy)
  3. Merge and deploy with rolling updates
  4. Zero-downtime patching

Base Images:

  • Backend: Python 3.14 on Alpine Linux
  • Frontend: Nginx on Alpine Linux
  • Database: PostgreSQL 16 official image

Reference: Deployment Topology

Client Support

Status: ✅ COMPLIANT (P0)

Web application works on any OS with a modern browser:

  • Chrome 120+, Firefox 120+, Safari 17+, Edge 120+
  • Windows 10+, macOS 12+, Linux (any distribution)

Database (SVC0020 / SVC0021)

Status: ⚠️ EXCEPTION NEEDED (P0)

We use PostgreSQL 16 instead of MariaDB or SQL Server.

Why PostgreSQL:

  • Superior JSON support for flexible data structures
  • Strong ACID compliance and performance
  • Already supported by EPFL Kubernetes infrastructure
  • Easier migration path (SQLAlchemy ORM abstracts database)

Risk: Low. PostgreSQL is widely used across EPFL and meets all security and backup requirements.

Action Required: Submit exception request to IT Security.

Reference: ADR-004

Security Scanning (SVC0062)

Status: ✅ COMPLIANT (P0)

Container Scanning: Trivy scans in CI/CD pipeline plus GitHub Security Scanning for dependencies.

File Uploads: Stored in EPFL-hosted S3-compatible object storage, encrypted by the application before upload. Processed in isolated worker containers. CSV only.

Encryption (transfer and keys)

Status: ⚠️ PARTIAL (P0)

Encryption is active on every external hop — TLS 1.3 throughout, verified against each database — and keys are held in ENAC-IT's Infisical vault. Two gaps remain: the database connection negotiates TLS without pinning it via sslmode, and intra-cluster traffic is segmented by NetworkPolicy rather than encrypted.

Reference: Encryption and Key Management

Backup (SVC0003)

Status: ✅ COMPLIANT (P0)

Type Method Retention Testing
Database PostgreSQL WAL + daily full 30d daily, 1y monthly Monthly restore
Object Storage EPFL-hosted S3, encryption enabled at provisioning Not versioned by decision — staging only n/a

Recovery: Database restore is requested from EPFL DSI; procedures are in the Disaster Recovery Plan in the private ops repository. Its restore template still carries SLA: ?, RPO: ?, RTO: ? — no recovery timeframe is agreed yet.

Logging and Monitoring

Status: ✅ COMPLIANT (P0)

Logging: All services produce structured JSON logs with timestamp, level, service, correlation_id, user_id, and event. Logs go to stdout/stderr, collected by Kubernetes, forwarded to EPFL SIEM. Retention: 90d in SIEM, 1y archive.

Monitoring: Prometheus metrics at /metrics endpoint. Health checks at /health (liveness) and /ready (readiness). Metrics include request count, duration, errors, and custom business metrics.

Reference: Global Conventions | CI/CD Pipeline

Exception Requests

Database Deviation: Using PostgreSQL instead of MariaDB/MSSQL

Approval Needed: IT Security / IT Service Manager

Risk: Low. PostgreSQL is supported by EPFL Kubernetes and meets all security requirements. Database is containerized and can be migrated if needed via SQLAlchemy ORM.

Next Steps

  1. Submit database exception request to IT Security
  2. Configure log forwarding to EPFL SIEM
  3. Register with End-to-End Service monitoring
  4. Quarterly compliance review

Quick Verification Commands

# Check authentication config
kubectl get secret entra-config -o yaml

# Verify structured logs
kubectl logs -l app=backend --tail=100 | jq

# Test metrics endpoint
curl https://co2-calculator.epfl.ch/api/metrics

# Check backup job
kubectl get cronjob backup-database

Contacts

  • Project Lead: [Technical Contact]
  • IT Security: security@epfl.ch
  • IT Service Manager: itsm@epfl.ch

References