# Service Architecture — Gateway Checker API

> Living document. Updated as we discuss and build. Read this before starting any work.

## Overview

REST API that wraps gateway checker scripts behind authenticated endpoints. Clients get an API key and hit endpoints — they never see source code. This is a **black-box service** — clients only know the API contract, not how it works internally.

### How the service works (the big picture)

```
┌─────────────────────────────────────────────────────────────────┐
│                        YOUR SERVER                               │
│                                                                  │
│  Client A ──►┐                                                   │
│  Client B ──►┤  FastAPI (api.py)                                 │
│  Client C ──►┤    │                                              │
│  Tg Bot   ──►┘    ├── Auth middleware (validate API key)         │
│                    ├── Rate limiter (per-key, per-minute)        │
│                    ├── Task queue (asyncio.Queue)                │
│                    │     │                                       │
│                    │     ▼                                       │
│                    ├── Gateway Registry                          │
│                    │     ├── comwave_forbot.py  [max_concurrent=1]│
│                    │     ├── comwave_forbot.py  [max_concurrent=5]│
│                    │     └── future_forbot.py   [max_concurrent=N]│
│                    │                                             │
│                    ├── Dedup cache (same card+gateway = cached)  │
│                    ├── Request logger (SQLite)                   │
│                    └── DB (api.db) — keys, usage, logs           │
│                                                                  │
│  Admin (you) ──► /admin/* endpoints (master key)                 │
└─────────────────────────────────────────────────────────────────┘
```

### The flow (step by step)

1. **Client sends request:** `POST /api/check/comwave` with API key + card
2. **Auth middleware:** Validates key exists, is active, not expired, not over limit
3. **Gateway access check:** Is this key allowed to use `comwave`?
4. **Rate limit check:** Has this key exceeded N requests/minute?
5. **Dedup check:** Same card + same gateway in last 60s? Return cached result
6. **Task created:** Card is queued, client gets back `task_id` immediately
7. **Worker picks it up:** Respects gateway's `max_concurrent` semaphore
8. **Gateway runs:** `comwave_forbot.py` does the actual check (5-15s)
9. **Result stored:** In-memory results dict + logged to SQLite
10. **Client polls:** `GET /api/result/{task_id}` → gets result when ready

---

## Stack

| Component | Choice | Why |
|---|---|---|
| Framework | **FastAPI** | Async native, auto-docs, fast |
| DB | **SQLite** (via `aiosqlite`) | Zero setup, file-based, good for this scale |
| Auth | API key in `X-API-Key` header | Simple, standard |
| Server | **uvicorn** | ASGI server for FastAPI |
| Config | **python-dotenv** | Already installed, clean `.env` loading |
| Keys | `secrets.token_urlsafe(32)` | 43-char random, unguessable |
| Task queue | **asyncio.Queue** | No Redis needed, in-process |
| Concurrency | **asyncio.Semaphore** per gateway | Controls parallel load |

**Packages to install:** `fastapi`, `uvicorn`, `aiosqlite`

---

## Files

| File | Purpose | Status |
|---|---|---|
| `api.py` | FastAPI server, gateway registry, auth, task queue, workers | ⬜ Not started |
| `db.py` | SQLite key management + request logging | ⬜ Not started |
| `comwave_forbot.py` | Comwave checker engine (existing, unchanged) | ✅ Done |
| `comwave_bot.py` | Telegram bot (existing, will update to call API later) | ✅ Done |
| `api.db` | SQLite database (auto-created at runtime) | Auto |
| `.env` | `MASTER_KEY`, `BOT_TOKEN`, `API_HOST`, `API_PORT`, credentials | ✅ Exists (needs `MASTER_KEY`) |
| `SERVICE_ARCHITECTURE.md` | This file | ✅ |

---

## API Endpoints

### Public (no auth)

| Method | Endpoint | Description |
|---|---|---|
| `GET` | `/health` | Health check — `{"status": "ok", "uptime": 3600}` |

### Client Endpoints (require API key)

| Method | Endpoint | Body | Description |
|---|---|---|---|
| `POST` | `/api/check/{gateway}` | `{"card": "cc\|mm\|yyyy\|cvv"}` | Submit card for checking, returns `task_id` |
| `GET` | `/api/result/{task_id}` | — | Poll for result (queued/processing/completed) |
| `GET` | `/api/usage` | — | Check own usage/limits/expiry |
| `GET` | `/api/gateways` | — | List available gateways + status |
| `GET` | `/api/cooldown` | — | Check own cooldown timers per gateway |

### Admin Endpoints (require master key)

| Method | Endpoint | Body | Description |
|---|---|---|---|
| `POST` | `/admin/keys` | `{"owner", "gateways", "request_limit", "expires_days", "rate_per_minute"}` | Create new API key |
| `GET` | `/admin/keys` | — | List all keys with usage |
| `GET` | `/admin/keys/{key}` | — | Get single key details |
| `PATCH` | `/admin/keys/{key}` | fields to update | Update key limits/expiry/gateways |
| `DELETE` | `/admin/keys/{key}` | — | Revoke a key |
| `GET` | `/admin/stats` | — | Live server stats (queues, 24h counts, gateway health) |

---

## Gateway Registry

Each gateway declares its function, concurrency limit, timeout, and optional lifecycle hooks:

```python
GATEWAYS = {
    "comwave": {
        "fn": checker.check_card,           # the async check function
        "type": "$0 auth",                  # display label
        "status": "online",                 # online/maintenance/offline
        "max_concurrent": 1,                # uses shared session, serialize
        "timeout": 30,                      # seconds before giving up
        "cooldown": 20,                     # per-user cooldown in seconds
        "max_queue": 50,                     # max tasks waiting in queue
        "init": checker.ensure_logged_in,   # called on API startup
        "shutdown": checker.shutdown,        # called on API shutdown
    },
    "comwave3": {
        "fn": check_card_3,
        "type": "$3.33 charge",
        "status": "online",
        "max_concurrent": 5,                # stateless, can run in parallel
        "timeout": 60,
        "cooldown": 20,                     # per-user cooldown in seconds
        "max_queue": 50,                     # max tasks waiting in queue
        "init": None,
        "shutdown": None,
    },
    # Adding a new gateway:
    # "duke": {
    #     "fn": duke_check,
    #     "type": "$1.00 charge",
    #     "status": "online",
    #     "max_concurrent": 3,
    #     "timeout": 45,
    #     "cooldown": 20,
    #     "max_queue": 50,
    #     "init": None,
    #     "shutdown": None,
    # },
}
```

**Adding a new gateway:**
1. Code `new_site_forbot.py` with `async def check_card(card: str) -> dict` function
2. Import and register in `GATEWAYS` dict in `api.py`
3. Create/update client keys to include the new gateway name
4. Done — client uses `POST /api/check/newsite`

**Gateway function contract:**
```python
async def check_card(card: str) -> dict:
    # card = "cc|mm|yyyy|cvv"
    # Must return: {"status": "approved"|"declined"|"error", "message": "...", "time": float}
```

**Cooldown:**
- Each gateway has a `cooldown` value (seconds) — **per-user**, not global
- After a user submits a card to a gateway, they must wait `cooldown` seconds before submitting another to the **same** gateway
- Different users are independent — User A's cooldown doesn't block User B
- Different gateways are independent — cooldown on `comwave` doesn't affect `comwave3`
- Tracked via in-memory dict: `last_request[api_key][gateway] → timestamp`
- If a user submits too soon: `429 {"detail": "Cooldown active", "retry_after": 12.5}` (remaining seconds)
- Trade-off accepted: site load scales with number of active users (10 users = up to 10 concurrent requests), but users never wait for each other

**Concurrency control:**
- Each gateway gets its own `asyncio.Semaphore(max_concurrent)`
- `comwave` max_concurrent=1 → only 1 check at a time (shared session with lock)
- `comwave3` max_concurrent=5 → up to 5 parallel checks (stateless)
- Excess requests wait in queue, not rejected

**Lifecycle hooks:**
- `init()` — called once when API server starts (e.g. comwave login)
- `shutdown()` — called once when API server stops (e.g. comwave logout)
- Both are optional (`None` for stateless gateways)

---

## Task Queue System

Instead of making clients wait 5-15s for a response, we use async tasks:

### Submit flow
```
POST /api/check/comwave {"card": "4111...|12|2025|123"}
     ↓
Returns immediately:
{"task_id": "abc123", "status": "queued", "gateway": "comwave", "position": 2, "estimated_wait": 15}
```

### Poll flow
```
GET /api/result/abc123
     ↓
While queued:      {"task_id": "abc123", "status": "queued", "position": 2, "expires_in": 290}
While processing:  {"task_id": "abc123", "status": "processing", "expires_in": 270}
When done:         {"task_id": "abc123", "status": "completed", "result": {...}, "remaining": 999}
```

### Implementation
- `asyncio.Queue` per gateway — tasks are queued and processed by workers
- Workers respect `max_concurrent` semaphore
- Results stored in `dict[task_id] → result` (in-memory, TTL 5 minutes)
- `task_id` = `uuid4()` short string

### Why this is better than synchronous
- No HTTP timeout issues (client gets response in <100ms)
- Client can submit multiple cards and poll all results
- Server controls processing speed via semaphores
- Batch support becomes trivial (submit N tasks, poll N results)

### Task ownership
- Each task is tied to the API key that created it
- Only the creating key can poll `GET /api/result/{task_id}`
- Other keys get `404 "Task not found"` — prevents cross-client result snooping
- Stored as `tasks[task_id] = {"api_key": key, "result": ..., ...}`

### Max queue depth
- Each gateway has a max queue size (default: 50 tasks)
- If queue is full: `503 {"detail": "Gateway overloaded", "queue_depth": 50}`
- Prevents a single client from flooding the queue during cooldown edge cases
- Configurable per gateway via `max_queue` field in registry

---

## Throttling Strategy

**Model: Per-user cooldown (20s default)**

Each user (API key) has an independent cooldown timer per gateway. After submitting a card, that user must wait `cooldown` seconds before submitting another card to the same gateway.

### How it works
```
User A submits card → comwave processes → User A must wait 20s
User B submits card → comwave processes → User B must wait 20s (independent of A)
User A submits to comwave3 → allowed immediately (different gateway)
```

### Implementation
```python
# In-memory tracker
last_request: dict[str, dict[str, float]] = {}  # {api_key: {gateway: timestamp}}

# Before queuing a task:
now = time.time()
last = last_request.get(api_key, {}).get(gateway, 0)
remaining = cooldown - (now - last)
if remaining > 0:
    raise HTTPException(429, {"detail": "Cooldown active", "retry_after": round(remaining, 1)})
last_request.setdefault(api_key, {})[gateway] = now
```

### Behavior with multiple users
| Time | User A | User B | User C | Site load |
|------|--------|--------|--------|-----------|
| 0s | submits card | submits card | — | 2 concurrent |
| 5s | waiting (15s left) | waiting (15s left) | submits card | 1 concurrent |
| 20s | submits next | submits next | waiting (5s left) | 2 concurrent |
| 25s | waiting | waiting | submits next | 1 concurrent |

**Max site load = number of active users** (each can send 1 request per 20s)

---

## Request Deduplication

If a client sends the **same card** to the **same gateway** within **60 seconds**, return the cached result instead of hitting the gateway again.

- **Key:** `hash(card + gateway)` (SHA256, not stored in plain text)
- **Cache:** In-memory dict with TTL
- **Why:** Prevents accidental double-charges on `comwave3`, saves gateway resources

---

## Database Schema (SQLite)

### `api_keys` table

| Column | Type | Description |
|---|---|---|
| `api_key` | TEXT PK | The key string (`sk_live_...`) |
| `owner` | TEXT | Client name/identifier |
| `created_at` | DATETIME | When key was created |
| `expires_at` | DATETIME NULL | NULL = never expires |
| `request_limit` | INTEGER NULL | NULL = unlimited |
| `requests_used` | INTEGER | Counter, starts at 0 |
| `rate_per_minute` | INTEGER | Max requests per minute (default 10) |
| `allowed_gateways` | TEXT | JSON list `["comwave","comwave3"]` or `"*"` for all |
| `is_active` | BOOLEAN | Can be deactivated without deleting |
| `is_paused` | BOOLEAN | Temporarily disabled (client sees "API key paused") |

### `request_log` table

| Column | Type | Description |
|---|---|---|
| `id` | INTEGER PK | Auto-increment |
| `api_key` | TEXT | Which key was used |
| `gateway` | TEXT | Which gateway |
| `card` | TEXT | Full card (`cc|mm|yyyy|cvv`) |
| `status` | TEXT | approved/declined/error |
| `response_time` | REAL | Seconds |
| `created_at` | DATETIME | Timestamp |
| `ip_address` | TEXT | Client IP |

### `admin_log` table

| Column | Type | Description |
|---|---|---|
| `id` | INTEGER PK | Auto-increment |
| `action` | TEXT | create_key / revoke_key / update_key / pause_key / unpause_key |
| `target_key` | TEXT | Which API key was affected |
| `details` | TEXT | JSON of what changed (e.g. `{"request_limit": [1000, 2000]}`) |
| `ip_address` | TEXT | Admin's IP |
| `created_at` | DATETIME | Timestamp |

---

## Security

| Layer | Implementation |
|---|---|
| **Auth** | `X-API-Key` header required on all client/admin endpoints |
| **Key format** | `sk_live_` + `secrets.token_urlsafe(32)` |
| **No docs in prod** | `FastAPI(docs_url=None, redoc_url=None)` |
| **HTTPS** | Behind nginx with SSL or Cloudflare tunnel |
| **Rate limit** | Per-minute limit per API key (in-memory sliding window) |
| **IP block** | Block IP after 10 failed auth attempts (in-memory, resets after 15min) |
| **Full card logging** | Full card stored in request_log (keep `api.db` protected) |
| **Dedup hashing** | Card+gateway hashed with SHA256, never stored raw |
| **Master key** | From `.env`, used for `/admin/*` only |
| **Health endpoint** | `/health` is the only unauthenticated endpoint, reveals nothing sensitive |
| **Graceful shutdown** | Finish in-progress checks before exit, call gateway shutdown hooks |

---

## Response Format

### Submit check
```json
{
    "task_id": "a1b2c3d4",
    "status": "queued",
    "gateway": "comwave",
    "position": 2,
    "estimated_wait": 15
}
```

### Poll result (queued)
```json
{
    "task_id": "a1b2c3d4",
    "status": "queued",
    "position": 2,
    "expires_in": 290
}
```

### Poll result (processing)
```json
{
    "task_id": "a1b2c3d4",
    "status": "processing",
    "expires_in": 270
}
```

### Poll result (completed)
```json
{
    "task_id": "a1b2c3d4",
    "status": "completed",
    "result": {
        "status": "declined",
        "gateway": "comwave",
        "message": "DECLINED [VISA] | Last4: 1111 | Auth: 000000 | Ref: 662321650018600120",
        "time": 5.28
    },
    "remaining": 999
}
```

### Usage
```json
{
    "owner": "client_john",
    "requests_used": 347,
    "requests_limit": 1000,
    "expires_at": "2026-04-24T00:00:00",
    "gateways": ["comwave", "comwave3"],
    "rate_per_minute": 10
}
```

### Gateways
```json
{
    "gateways": {
        "comwave":  {"status": "online", "type": "$0 auth"},
        "comwave3": {"status": "online", "type": "$3.33 charge"}
    }
}
```

### Errors
```json
{"detail": "Missing API key"}                     // 401
{"detail": "Invalid API key"}                      // 401
{"detail": "API key expired"}                      // 401
{"detail": "API key deactivated"}                  // 401
{"detail": "Request limit exceeded", "used": 1000, "limit": 1000}  // 429
{"detail": "Rate limit exceeded", "retry_after": 6}               // 429
{"detail": "Cooldown active", "retry_after": 12.5}                // 429
{"detail": "Access denied for gateway: duke"}      // 403
{"detail": "Gateway not found: xyz"}               // 404
{"detail": "Gateway offline: duke"}                // 503
{"detail": "Gateway overloaded", "queue_depth": 50}  // 503
{"detail": "API key paused"}                          // 401
{"detail": "Invalid card format"}                  // 400
{"detail": "Task not found"}                       // 404
{"detail": "IP blocked"}                           // 403
```

---

## Key Tiers (pricing model)

| Tier | Limit | Expiry | Gateways | Rate |
|---|---|---|---|---|
| Trial | 50 requests | 7 days | 1 gateway | 5/min |
| Standard | 1000/month | 30 days | specific | 10/min |
| Unlimited | no limit | no expiry | all | 50/min |

---

## Telegram Bot Integration (future)

**Option A (recommended):** Bot calls own API with its own key — cleanest separation, bot is just another API consumer. The bot gets an "unlimited" key with all gateways.

**Option B:** Bot imports checkers directly (current setup) — faster, no network hop but tightly coupled.

Decision: TBD — build API first, connect bot later.

---

## .env Structure

```env
# API Server
MASTER_KEY=your_admin_secret_here
API_HOST=0.0.0.0
API_PORT=8000

# Telegram Bot
BOT_TOKEN=your_bot_token
ALLOWED_USERS=123456,789012

# Gateway Credentials
COMWAVE_USERNAME=tu76502@gmail.com
COMWAVE_PASSWORD=root13579
```

---

## Build Order

### Phase 1: Core API (v1)
1. ⬜ Install `fastapi`, `uvicorn`, `aiosqlite`
2. ⬜ Build `db.py` — SQLite init, key CRUD, usage tracking, request logging
3. ⬜ Build `api.py` — FastAPI app, auth middleware, gateway registry, task queue + workers, all endpoints
4. ⬜ Test with comwave gateway — submit, poll, verify result
5. ⬜ Test admin endpoints — create key, list, revoke
6. ⬜ Test security — expired key, over-limit, wrong gateway, IP blocking
7. ⬜ Test concurrency — multiple clients hitting same gateway

### Phase 2: Bot + Polish
8. ⬜ Connect Telegram bot via API (Option A)
9. ⬜ Add more gateways as clients request them

### Phase 3: Future (v2)
10. ⬜ Webhook/callback support for async results
11. ⬜ Batch endpoint (`POST /api/check/batch/{gateway}`)
12. ⬜ IP whitelisting per API key
13. ⬜ Key rotation
14. ⬜ Read-only key scoping (usage-only vs full-access)
15. ⬜ Client dashboard (simple HTML)
16. ⬜ Gateway health monitoring (auto-disable on failure)
17. ⬜ Cloudflare Tunnel for easy HTTPS

---

## Existing Gateways Reference

### Comwave $0 auth (`comwave`)
- **File:** `comwave_forbot.py` → `ComwaveChecker.check_card()`
- **Type:** Account portal card update, Moneris Checkout v1
- **Amount:** $0 (auth only)
- **Session:** Single persistent login, `asyncio.Lock` for ticket serialization
- **Concurrency:** max_concurrent=1 (shared session)
- **Lifecycle:** init = `ensure_logged_in()`, shutdown = `shutdown()`
- **Note:** Server enforces one login at a time

### Comwave $3.33 charge (`comwave3`)
- **File:** `comwave_forbot.py` → `check_card_3()`
- **Type:** WooCommerce guest checkout, Moneris Checkout v1
- **Amount:** $3.33 CAD ($2.95 + 12% tax)
- **Session:** Stateless, each call gets own session
- **Concurrency:** max_concurrent=5 (fully parallel-safe)
- **Lifecycle:** no init/shutdown needed

---

*Last updated: 2026-03-24*