7. Deployment View

7.1 Infrastructure Overview

The system is deployed on a self-hosted server using Docker Compose. Services are organized into different profiles:

Tech Stack: Core infrastructure services (Nginx, RabbitMQ, PostgreSQL, MinIO, MLflow, Prometheus, Loki, Tempo, Mimir, Grafana).
Application Services: Streamlit UI, FastAPI Backend (Pipeline API), Data Processor.

The server runs all containers in an isolated Docker network, exposing only necessary ports through an Nginx reverse proxy with automated Let’s Encrypt SSL certificates.

7.2 Deployment Diagram

        graph TB
    subgraph Self-Hosted Server
        subgraph Docker Network
            subgraph Reverse Proxy
                Nginx[Nginx Proxy] -->|SSL Certificates| Acme[Acme Companion]
            end

            subgraph Core Services
                DB[PostgreSQL]
                MQ[RabbitMQ]
                DP[Data Processor]
                S3[MinIO]
                MLflow[MLflow Server] --> |Save artifact and model| DB
                Logs[Prometheus + Loki + Tempo + Mimir - Logging and Metrics] --> Grafana[Grafana]
            end

            subgraph Application Services
                UI[Streamlit UI] -->|API Calls| API[FastAPI Backend]
                API -->|Model Requests| MQ
                MQ --> |Model Requsts| DP
                DP -->|Fetch Model| MLflow
                DP -->|Store Data| S3
                API -->|Queue Tasks| MQ
            end
        end
    end

7.3 Network and Hosting Details

All services are deployed in Docker containers.
The Nginx reverse proxy exposes public services using HTTPS (port 443).
Internal communication happens within the Docker network.
Persistent data (DB, MinIO, MLflow) is stored in Docker volumes.
There is a staging and a production view of the services.

Deployment Services
Service	Host/URL - Production	Host/URL - Staging	Port	Exposed?
Nginx Proxy	/	/	443	Privat
RabbitMQ	https://rabbitmq.mlops.dns64.de/	https://rabbitmq.staging.mlops.iot64.de/	15672	Public
PostgreSQL	postgres	postgres	5432	Private
MinIO	https://s3.mlops.dns64.de/	https://s3.staging.mlops.iot64.de/	9001	Public
MLflow	https://mlflow.mlops.dns64.de/	https://mlflow.staging.mlops.iot64.de/	5000	Public
Prometheus	/	/	9090	Private
Grafana	https://monitoring.mlops.dns64.de/	https://monitoring.staging.mlops.iot64.de/	3000	Public
FastAPI Backend	https://api.mlops.dns64.de/docs	https://api.staging.mlops.iot64.de/docs	8000	Public
Streamlit UI	https://app.mlops.dns64.de/	https://app.staging.mlops.iot64.de/	8501	Public

7.4 Deployment Process

CI/CD Pipeline

All services are built and published as Docker images via GitHub Actions.
Images are pushed to GitHub Container Registry (GHCR).
The latest version is deployed using Docker Compose.
Startup Sequence

Nginx Proxy starts first.

Database (PostgreSQL) and RabbitMQ must be ready before application services start.
MinIO (S3 storage) must be running before MLflow can store artifacts.
Monitoring & Logging

Prometheus collects system metrics.

Loki stores logs from all services.
Tempo handles distributed tracing.
Grafana visualizes logs, metrics, and traces.

7.5 Scaling and Future Improvements

The system can be extended to Kubernetes for better scalability.
A load balancer can distribute API requests for high availability.
Auto-scaling workers can be introduced for handling data processing.