Add ARCHITECTURE.md system design documentation

This commit is contained in:
courtmanr@gmail.com 2025-12-02 20:40:31 +00:00
parent fdf7de29f0
commit d96c5d3e11

88
ARCHITECTURE.md Normal file
View file

@ -0,0 +1,88 @@
# Pulse Architecture
Pulse is a real-time, agentless (mostly) monitoring system designed for Proxmox VE, Proxmox Backup Server, and Docker infrastructure. It is built with a **Go** backend and a **SolidJS** frontend, focusing on low latency, high concurrency, and a premium user experience.
## 🏗 High-Level Overview
The system operates as a single binary that serves both the API and the static frontend assets. It connects to infrastructure nodes via SSH (for Proxmox) or local/remote Docker sockets to gather metrics, which are then streamed to connected clients via WebSockets.
```mermaid
graph TD
User[User Browser] <-->|WebSocket / HTTP| Pulse[Pulse Server]
subgraph "Pulse Server (Go)"
API[REST API]
WS[WebSocket Hub]
Monitor[Monitoring Engine]
Config[Config Manager]
end
Pulse -->|SSH| PVE[Proxmox VE Node]
Pulse -->|SSH| PBS[Proxmox Backup Server]
Pulse -->|Docker Socket| Docker[Docker Host]
Monitor --> WS
Monitor --> API
```
## 🔌 Backend Architecture (Go)
The backend is a high-performance Go application designed for concurrent monitoring.
### Core Components
1. **Entry Point (`cmd/pulse/main.go`)**:
* Initializes the configuration, logger, and persistence layer.
* Starts the `ReloadableMonitor` which manages the lifecycle of monitoring routines.
* Launches the HTTP server and WebSocket hub.
2. **Monitoring Engine (`internal/monitoring`)**:
* **Polymorphic Monitors**: Uses interfaces to treat PVE, PBS, and Docker hosts uniformly where possible.
* **Goroutines**: Each host is monitored in its own lightweight goroutine to ensure non-blocking operations.
* **SSH Connection Pooling**: Maintains persistent SSH connections to Proxmox nodes to avoid handshake overhead during metric collection.
3. **WebSocket Hub (`internal/websocket`)**:
* Manages active client connections.
* Broadcasts metric updates in real-time.
* Handles "commands" from the frontend (e.g., requesting immediate updates).
4. **API Layer (`internal/api`)**:
* RESTful endpoints for configuration (adding nodes, setting thresholds).
* Handles authentication and secure token management.
### Data Flow
1. **Collection**: The `Monitoring Engine` ticks (default: 2s). It executes commands on remote hosts (e.g., `pvesh`, `docker stats`).
2. **Normalization**: Raw JSON/Text output is parsed into standardized Go structs (`HostMetrics`, `ContainerMetrics`).
3. **Broadcast**: Normalized data is sent to the `WebSocket Hub`.
4. **Delivery**: The Hub serializes the data to JSON and pushes it to all subscribed frontend clients.
## 🎨 Frontend Architecture (SolidJS)
The frontend is a modern Single Page Application (SPA) built with **SolidJS** and **TypeScript**. It prioritizes performance by using fine-grained reactivity instead of a Virtual DOM.
### Key Technologies
* **SolidJS**: For reactive UI components.
* **TailwindCSS**: For styling and theming (Dark/Light mode).
* **Vite**: For fast development and optimized builds.
### State Management
* **Stores (`frontend-modern/src/stores`)**:
* `websocket.ts`: The central nervous system. It maintains the WS connection, handles reconnection logic, and updates reactive signals when new data arrives.
* `metricsHistory.ts`: Buffers incoming metrics to drive historical charts (Sparklines) without needing a time-series database backend.
### Component Design
* **Atomic Design**: Small, reusable components (`MetricBar`, `StatusBadge`) compose into larger views (`NodeSummaryTable`).
* **Visualizations**: Custom SVG-based charts (Sparklines) are used instead of heavy charting libraries to keep the bundle size small and rendering fast.
## 🔒 Security
* **Encryption at Rest**: Sensitive configuration (passwords, API keys) is encrypted on disk using `AES-GCM` with a user-provided passphrase.
* **Transport Security**: All communications can be secured via TLS.
* **Authentication**: Session-based auth for API access.
## 🚀 Deployment
Pulse is distributed as:
1. **Docker Container**: Multi-stage build resulting in a scratch-based or alpine-based image containing just the binary and frontend assets.
2. **Single Binary**: The frontend is embedded into the Go binary using `embed`, allowing for a single-file deployment.