Cloud Architecture

Now Playing runs on a hybrid cloud architecture. The application services live on a homelab server for low-latency real-time operations, while durable storage, global edge delivery, and failover are handled by cloud providers.

High-Level Topology

             Your DJ software
                    │
                    ▼
           Now Playing Desktop
                    │
         HTTPS / Socket.IO / SSE
                    │
                    ▼
┌───────────────────────────────────────────┐
│          Cloudflare Edge                  │
│  - DNS, SSL, DDoS protection              │
│  - Load balancer (primary / failover)     │
│  - CDN for overlay assets and artwork     │
└────────────┬────────────────────┬─────────┘
             │                    │
   Cloudflare Tunnel          Direct HTTPS
             │                    │
             ▼                    ▼
┌─────────────────────┐   ┌─────────────────────┐
│  Proxmox Homelab    │   │  GCP Failover VM    │
│  (Primary, "Blue")  │   │  (Standby, "Green") │
│                     │   │                     │
│  Next.js + Python   │   │  Same image, stopped│
│  Redis              │   │  unless activated   │
└─────────┬───────────┘   └─────────┬───────────┘
          │                         │
          └────────────┬────────────┘
                       ▼
             ┌─────────────────────┐
             │  GCP Managed Data   │
             │  - PostgreSQL       │
             │  - Redis (replica)  │
             └─────────────────────┘

Where Things Run

Homelab (Primary)

The homelab server — a Proxmox host running an LXC container with a unified Docker image — hosts the application services:

• Next.js web application (dashboard, APIs, overlay pages).
• Track Service, Router Service, Overlay Service, Twitch Service, File Output Service, Mix Processor Service — Python microservices that deliver the real-time pipeline.
• Redis — in-memory pub/sub, streams, and short-lived cache.

All of these services run in a single container supervised by s6-overlay, which gives each service its own independent supervision and restart logic.

Cloudflare

Cloudflare sits at the edge:

• DNS and SSL for app.nowplayingapp.com.
• Load balancer that routes traffic between primary and failover pools based on health checks.
• Tunnel that connects the homelab outbound — no port forwarding needed, and no public IP exposed.
• CDN + R2 for artwork and static assets.

Google Cloud Platform

GCP hosts the data tier and the standby failover VM:

• Cloud SQL (PostgreSQL) — durable storage for users, tracks, settings.
• Managed Redis — persistent key-value data that survives deploys and failovers.
• Failover VM — a standby Compute Engine instance running the same application image, stopped by default.

Backblaze B2

Cold storage for database backups, retained for long-term durability.

Why Hybrid?

Pure cloud would be simpler operationally, but a few constraints push the architecture toward hybrid:

• Latency. The real-time pipeline (track → enrichment → overlay) is measured in hundreds of milliseconds. Keeping services co-located with Redis on local hardware keeps round-trips tight.
• Cost. Running Redis and the Python services on owned hardware avoids per-service cloud billing.
• Durability. Putting PostgreSQL in GCP Cloud SQL gives managed durability and backups that a single-host homelab cannot match.

The blended design gets the best of both: low-latency hot paths on local hardware, durable data on managed services.

What Traffic Goes Where

Destination	Path	Latency-Sensitive?
Overlay SSE stream	Cloudflare → Tunnel → Next.js → Redis	Yes
Track ingest from desktop	Cloudflare → Tunnel → Next.js → Redis	Yes
Track history / dashboard	Cloudflare → Tunnel → Next.js → Cloud SQL	No
Artwork image	Cloudflare CDN → R2	No (cached)
Authentication	Cloudflare → Tunnel → Next.js → Cloud SQL	No

Cost Envelope

The full stack runs at a low monthly cost because the compute stays on the homelab. Cloud charges come mainly from Cloud SQL (small instance) and R2 (egress for artwork). A standby failover VM costs nothing while stopped; it only runs when the primary is down or during a deployment window.