Files
MiruroAPI/docs/architecture.md
T
Shinei Nouzen 4293683ae4 fix: update docs to reflect CF Worker blocking reality
Both AniList and Miruro block Cloudflare Worker IPs (HTTP 403).
Worker approach tested and confirmed non-functional:
- /api/genres → AniList 403: 'manually blocked'
- /api/search → AniList 403: 'manually blocked'
- /api/episodes → Miruro 403: blocked

Updated README.md and docs/architecture.md:
- Worker section marked as experimental
- Troubleshooting reordered: FlareSolverr > ScraperAPI > self-host
- Architecture doc updated with actual test results
2026-07-02 09:58:21 +05:30

254 lines
8.4 KiB
Markdown

# Architecture
## Project Structure
```
MiruroAPI/
├── server.js # Express entry point, port 3000
├── worker.js # Cloudflare Worker edition (edge runtime)
├── package.json # name: "miruro-api"
├── vercel.json # Routes /api/* and /* to server.js
├── Dockerfile # Docker support (node:20-alpine)
├── .dockerignore # Docker ignore file
│
├── public/ # Static files served from process.cwd()
│ ├── index.html # Premium landing page (SVG icons, live console)
│ ├── docs.html # Swagger UI interactive documentation
│ ├── openapi.json # OpenAPI 3.0 spec
│ ├── 404.html # Glitch animation error page
│ ├── manifest.json # PWA manifest (theme: #A855F7)
│ ├── robots.txt # Crawler directives
│ ├── sitemap.xml # 4 pages (/, /tos, /privacy, /api)
│ ├── icon-dark.svg # Miruro dark mode SVG icon
│ ├── icon-light.svg # Miruro light mode SVG icon
│ ├── icon-512x512.png # Miruro app icon
│ ├── favicon.ico # Classic favicon
│ ├── apple-touch-icon-180x180.png # iOS home screen icon
│ ├── og-image.png # OG/Twitter share image
│ ├── privacy.html # Privacy policy (served at /privacy)
│ └── tos.html # Terms of service (served at /tos)
│
├── assets/ # Scraped Miruro assets
│ ├── favicons/ # All favicon variants
│ ├── logos/ # Status page logo
│ ├── fonts/ # Inter + FontAwesome
│ └── media/ # Testimonial avatars
│
├── docs/ # API documentation
│ ├── index.md # Overview, quick start, features
│ ├── endpoints.md # Full API reference (18 endpoints)
│ ├── streaming.md # Streaming flow guide (3-step)
│ ├── examples.md # cURL, JavaScript, Python
│ └── architecture.md # This file
│
├── src/
│ ├── helpers/
│ │ ├── anilist.js # AniList GraphQL integration
│ │ ├── pipe.js # Miruro pipe integration
│ │ └── cache.js # In-memory Map cache with TTL
│ │
│ └── routes/
│ └── apiRoutes.js # All route definitions
│
├── test.js # Integration test suite
├── CHANGELOG.md # Version history
├── LICENSE # MIT License
└── README.md # Full API documentation
```
## Tech Stack
| Component | Technology |
|-----------|------------|
| Runtime | Node.js |
| Framework | Express.js |
| Metadata | AniList GraphQL (`graphql.anilist.co`) |
| Streaming | Miruro pipe (`miruro.{to,ru,bz,tv}/api/secure/pipe`) |
| Deployment | Vercel (Serverless) |
| Caching | In-memory Map with 1-5 min TTL |
| Static Files | Express static middleware |
## Request Flow
```
Client Request
↓
Vercel Routes (/api/* → server.js)
↓
Express Router (apiRoutes.js)
↓
Helper (e.g., anilist.js or pipe.js)
↓
HTTP Request to AniList GraphQL / Miruro pipe
↓
Response decoded (base64url + gzip for pipe)
↓
Returns structured JSON
↓
Client Response
```
## Streaming Flow
```
/api/episodes/:anilistId
↓
Returns: providers with sub/dub episode lists
↓
/api/watch/:provider/:anilistId/:category/:slug
↓
Returns: streams[] with M3U8 URLs, quality, codec, fansub
↓
Play M3U8 in HLS player (hls.js, video.js, native)
```
## Caching Strategy
- **Type:** In-memory Map
- **TTL:** 1-5 minutes depending on endpoint
- **Max Size:** 100 entries with FIFO eviction
- **Key:** Full request URL
- **Behavior:** First request fetches from source, subsequent requests served from cache
- **Eviction:** Automatic when TTL expires or max size reached
## Data Sources
### AniList GraphQL
- **Endpoint:** `https://graphql.anilist.co`
- **Data:** Search, suggestions, info, characters, relations, recommendations, filter, schedule
- **Format:** Standard GraphQL queries
### Miruro Pipe
- **Endpoint:** `https://miruro.{to,ru}/api/secure/pipe`
- **Data:** Episodes, streaming sources (M3U8 URLs)
- **Encoding:** Base64url + gzip compressed request/response
- **Providers:** kiwi, pewe, bee, bonk, bun, ally, nun, twin, cog, moo, hop, telli
## Self-Healing Fallback System (v2.3.0)
When Cloudflare blocks streaming requests, the API automatically tries fallback methods in order:
1. **Direct** → Mirror rotation across 4 domains (miruro.to, miruro.ru, miruro.bz, miruro.tv)
2. **ScraperAPI** → Free proxy bypass via ScraperAPI (1K req/month, requires `SCRAPER_API_KEY`)
3. **FlareSolverr** → Self-hosted browser proxy for full Cloudflare bypass
The system tracks success/failure rates per method and auto-rotates to the most reliable option. Check status via `/api/pipe-health`.
## Vercel Configuration
```json
{
"version": 2,
"builds": [
{
"src": "server.js",
"use": "@vercel/node"
}
],
"routes": [
{
"src": "/api/(.*)",
"dest": "server.js"
},
{
"src": "/(.*)",
"dest": "server.js"
}
]
}
```
- `/api/*` routes to Express for API handling
- `/*` routes to Express for static file serving
- Static files served from `process.cwd()` (Vercel serverless compatible)
## Environment Variables
| Variable | Default | Description |
|----------|---------|-------------|
| `PORT` | `3000` | Server port (Express mode only) |
| `ALLOWED_ORIGINS` | `*` | Comma-separated allowed origins |
## Response Format
All API responses follow this structure:
```json
{
"success": true,
"results": { ... }
}
```
Error responses:
```json
{
"success": false,
"message": "Error description"
}
```
## Dependencies
| Package | Version | Purpose |
|---------|---------|---------|
| express | ^4.21.0 | Web framework |
| axios | ^1.8.0 | HTTP requests |
| cors | ^2.8.5 | CORS headers |
| dotenv | ^16.4.0 | Environment variables |
## Cloudflare Worker Architecture (v2.3.1)
The Worker edition (`worker.js`) runs the same API on Cloudflare's edge network. **⚠️ Experimental** — testing shows that both AniList and Miruro block CF Worker IPs (HTTP 403).
### Why Workers Don't Work (Current)
```
Vercel (datacenter IP) → miruro.to → Cloudflare WAF → 403 Blocked
Cloudflare Worker (edge) → miruro.to → Cloudflare bot detection → 403 Blocked
Cloudflare Worker (edge) → graphql.anilist.co → AniList blocks worker IPs → 403 Blocked
```
Cloudflare Worker IPs are recognized as automated traffic and blocked by both services.
### Worker Runtime Differences
| Feature | Express (Node.js) | Cloudflare Worker |
|---------|-------------------|-------------------|
| HTTP client | axios | native fetch |
| Decompression | zlib (Node.js) | DecompressionStream (Web API) |
| Caching | in-memory Map | Workers KV (optional) |
| npm packages | Yes | No (bundle only) |
| Cold starts | ~500ms | Zero |
| CPU limit | None | 10ms (free) / 30s (paid) |
### Worker Endpoints (15)
| Endpoint | Description |
|----------|-------------|
| `GET /api/health` | System health check |
| `GET /api/search?query=` | Full-text anime search |
| `GET /api/suggestions?query=` | Autocomplete suggestions |
| `GET /api/filter?genre=` | Advanced filter |
| `GET /api/trending` | Trending anime |
| `GET /api/popular` | Most popular |
| `GET /api/top` | Top scored |
| `GET /api/random` | Random anime |
| `GET /api/info/:id` | Full anime info |
| `GET /api/anime/:id/characters` | Characters + voice actors |
| `GET /api/anime/:id/relations` | Related anime |
| `GET /api/genres` | All genres |
| `GET /api/episodes/:id` | Episode list (pipe) |
| `GET /api/watch/:provider/:anilistId/:category/:slug` | Streaming sources (pipe) |
### Deploy (Experimental)
```bash
npm install -g wrangler
wrangler login
wrangler deploy worker.js --name miruroapi
```