Files
flyer-crawler.projectium.com/docs/adr/0032-rate-limiting-strategy.md

148 lines
6.0 KiB
Markdown

# ADR-032: Rate Limiting Strategy
**Date**: 2026-01-09
**Status**: Accepted
**Implemented**: 2026-01-09
## Context
Public-facing APIs are vulnerable to abuse through excessive requests, whether from malicious actors attempting denial-of-service attacks, automated scrapers, or accidental loops in client code. Without proper rate limiting, the application could:
1. **Experience degraded performance**: Excessive requests can overwhelm database connections and server resources
2. **Incur unexpected costs**: AI service calls (Gemini API) and external APIs (Google Maps) are billed per request
3. **Allow credential stuffing**: Login endpoints without limits enable brute-force attacks
4. **Suffer from data scraping**: Public endpoints could be scraped at high volume
## Decision
We will implement a tiered rate limiting strategy using `express-rate-limit` middleware, with different limits based on endpoint sensitivity and resource cost.
### Tier System
| Tier | Window | Max Requests | Use Case |
| --------------------------- | ------ | ------------ | -------------------------------- |
| **Authentication (Strict)** | 15 min | 5 | Login, registration |
| **Sensitive Operations** | 1 hour | 5 | Password changes, email updates |
| **AI/Costly Operations** | 15 min | 10-20 | Gemini API calls, geocoding |
| **File Uploads** | 15 min | 10-20 | Flyer uploads, avatar uploads |
| **Batch Operations** | 15 min | 50 | Bulk updates |
| **User Read** | 15 min | 100 | Standard authenticated endpoints |
| **Public Read** | 15 min | 100 | Public data endpoints |
| **Tracking/High-Volume** | 15 min | 150-200 | Analytics, reactions |
### Rate Limiter Configuration
All rate limiters share a standard configuration:
```typescript
const standardConfig = {
standardHeaders: true, // Return rate limit info in headers
legacyHeaders: false, // Disable deprecated X-RateLimit headers
skip: shouldSkipRateLimit, // Allow bypassing in test environment
};
```
### Test Environment Bypass
Rate limiting is bypassed during integration and E2E tests to avoid test flakiness:
```typescript
export const shouldSkipRateLimit = (req: Request): boolean => {
return process.env.NODE_ENV === 'test';
};
```
## Implementation Details
### Available Rate Limiters
| Limiter | Window | Max | Endpoint Examples |
| ---------------------------- | ------ | --- | --------------------------------- |
| `loginLimiter` | 15 min | 5 | POST /api/auth/login |
| `registerLimiter` | 1 hour | 5 | POST /api/auth/register |
| `forgotPasswordLimiter` | 15 min | 5 | POST /api/auth/forgot-password |
| `resetPasswordLimiter` | 15 min | 10 | POST /api/auth/reset-password |
| `refreshTokenLimiter` | 15 min | 20 | POST /api/auth/refresh |
| `logoutLimiter` | 15 min | 10 | POST /api/auth/logout |
| `publicReadLimiter` | 15 min | 100 | GET /api/flyers, GET /api/recipes |
| `userReadLimiter` | 15 min | 100 | GET /api/users/profile |
| `userUpdateLimiter` | 15 min | 100 | PUT /api/users/profile |
| `userSensitiveUpdateLimiter` | 1 hour | 5 | PUT /api/auth/change-password |
| `adminTriggerLimiter` | 15 min | 30 | POST /api/admin/jobs/\* |
| `aiGenerationLimiter` | 15 min | 20 | POST /api/ai/analyze |
| `aiUploadLimiter` | 15 min | 10 | POST /api/ai/upload-and-process |
| `geocodeLimiter` | 1 hour | 100 | GET /api/users/geocode |
| `priceHistoryLimiter` | 15 min | 50 | GET /api/price-history/\* |
| `reactionToggleLimiter` | 15 min | 150 | POST /api/reactions/toggle |
| `trackingLimiter` | 15 min | 200 | POST /api/personalization/track |
| `batchLimiter` | 15 min | 50 | PATCH /api/budgets/batch |
### Usage Pattern
```typescript
import { loginLimiter, userReadLimiter } from '../config/rateLimiters';
// Apply to individual routes
router.post('/login', loginLimiter, validateRequest(loginSchema), async (req, res, next) => {
// handler
});
// Or apply to entire router for consistent limits
router.use(userReadLimiter);
router.get('/me', async (req, res, next) => {
/* handler */
});
```
### Response Headers
When rate limiting is active, responses include standard headers:
```
RateLimit-Limit: 100
RateLimit-Remaining: 95
RateLimit-Reset: 900
```
### Rate Limit Exceeded Response
When a client exceeds their limit:
```json
{
"message": "Too many login attempts from this IP, please try again after 15 minutes."
}
```
HTTP Status: `429 Too Many Requests`
## Key Files
- `src/config/rateLimiters.ts` - Rate limiter definitions
- `src/utils/rateLimit.ts` - Helper functions (test bypass)
## Consequences
### Positive
- **Security**: Protects against brute-force and credential stuffing attacks
- **Cost Control**: Prevents runaway costs from AI/external API abuse
- **Fair Usage**: Ensures all users get reasonable service access
- **DDoS Mitigation**: Provides basic protection against request flooding
### Negative
- **Legitimate User Impact**: Aggressive users may hit limits during normal use
- **IP-Based Limitations**: Shared IPs (offices, VPNs) may cause false positives
- **No Distributed State**: Rate limits are per-instance, not cluster-wide (would need Redis store for that)
## Future Enhancements
1. **Redis Store**: Implement distributed rate limiting with Redis for multi-instance deployments
2. **User-Based Limits**: Track limits per authenticated user rather than just IP
3. **Dynamic Limits**: Adjust limits based on user tier (free vs premium)
4. **Monitoring Dashboard**: Track rate limit hits in admin dashboard
5. **Allowlisting**: Allow specific IPs (monitoring services) to bypass limits