Skip to content

Advanced Backend Architecture

This document covers the "under the hood" technical decisions that make Bedrud's high-performance, single-binary architecture possible.

1. The Fiber-to-Standard Bridge

Bedrud uses Fiber for its speed, but some dependencies like Goth (OAuth) and LiveKit SDK (Twirp) expect standard Go http.ResponseWriter and http.Request.

To bridge these two worlds, Bedrud implements an Adapter Pattern in internal/handlers/auth.go:

responseWriter Adapter

type responseWriter struct {
    ctx     *fiber.Ctx
    headers http.Header
    status  int
}
This struct implements the http.ResponseWriter interface. When a standard library function (like Goth) calls Header().Add(), the adapter stores it in a local headers map. When WriteHeader() is called, it manually copies all headers back into the Fiber Context.

Why this is important:

It allows Bedrud to remain extremely fast (using Fiber) while maintaining full compatibility with the vast ecosystem of standard Go web libraries.

2. LiveKit Reverse Proxying

Bedrud avoids port management issues by routing all media signaling through its main port.

How it works:

  1. Detection: If LIVEKIT_INTERNAL_HOST points to 127.0.0.1, Bedrud starts the internal media engine.
  2. Mounting: It mounts a httputil.NewSingleHostReverseProxy on the /livekit route.
  3. Path Stripping: A custom Director function is used to strip the /livekit prefix so that the media server receives the request as if it were at root.
  4. WebSocket Support: Since the proxy operates on the underlying TCP stream via adaptor.HTTPHandler, it natively supports the WebSockets used by LiveKit for real-time signaling.

3. Storage Efficiency: Embedded Structs

Bedrud makes heavy use of GORM's embedded tag to keep the database schema flat and optimized.

Example from models/room.go: 

type Room struct {
    ID       string       `gorm:"primaryKey"`
    Settings RoomSettings `gorm:"embedded;embeddedPrefix:settings_"`
}
This causes GORM to create columns like settings_allow_chat directly in the rooms table. This is faster than a JOIN and more searchable than a JSON blob.

4. Testing Infrastructure

Bedrud follows a "DB-First" testing strategy using the internal/testutil package.

SetupTestDB logic:

  • Creates a unique SQLite in-memory database for every test.
  • Automatically runs all migrations.
  • Provides a clean slate, ensuring tests are deterministic and can run in parallel.

To run backend tests: 

go test ./internal/...

5. Security: Token Rotation & Rotation

Unlike simple JWT implementations, Bedrud rotates both tokens on every refresh.

  • Refresh Token Reuse Detection: Once a refresh token is used to get a new pair, it is blocked.
  • Cleanup: The BlockedRefreshToken table ensures that even if a refresh token is stolen, it cannot be used after the user has cleanly logged out or refreshed their session.