Redis Caching

Most code should use lib/redis.ts — it handles the singleton, fallback, and error handling.

Basic Cache Operations (lib/redis.ts)

import { getCached, setCached, deleteCached, isRedisAvailable } from '@/lib/redis';

// Check availability first (optional — functions fail gracefully)
if (isRedisAvailable()) {
  // Set with TTL (seconds)
  await setCached('my-key', { data: 'value' }, 3600); // 1 hour

  // Get typed value
  const data = await getCached<{ data: string }>('my-key');

  // Delete
  await deleteCached('my-key');
}

Key behaviors from actual implementation:

• Initialized via REDIS_URL environment variable
• enableOfflineQueue: false — fail fast when Redis is down
• lazyConnect: true — connects on first command
• All operations return null/false/0 when Redis is unavailable (never throws)
• Auto-serializes to JSON via JSON.stringify/JSON.parse

Pattern-Based Invalidation

import { clearCachePattern } from '@/lib/redis';

// Delete all keys matching a glob pattern
const deletedCount = await clearCachePattern('cache:projectId:*');

Rate Limiting (lib/rate-limiter.ts)

Uses Redis counters with in-memory fallback:

import {
  checkRateLimit,
  RATE_LIMITS,
  getRateLimitIdentifier,
  getClientIp,
  createRateLimitHeaders,
} from '@/lib/rate-limiter';

export async function POST(req: NextRequest) {
  const session = await getServerSession(authOptions);
  const ip = getClientIp(req);
  const identifier = getRateLimitIdentifier(session?.user?.id || null, ip);

  const rateLimit = await checkRateLimit(
    `chat:${identifier}`,
    RATE_LIMITS.CHAT
  );

  if (!rateLimit.success) {
    return NextResponse.json(
      { error: 'Rate limit exceeded' },
      { status: 429, headers: createRateLimitHeaders(rateLimit) }
    );
  }

  // ... proceed with request
}

Predefined rate limits:

RATE_LIMITS.CHAT     // 20 messages/minute
RATE_LIMITS.UPLOAD   // 10 uploads/minute
RATE_LIMITS.API      // 60 requests/minute
RATE_LIMITS.AUTH     // 5 login attempts/5 minutes
RATE_LIMITS.RENDER   // 5 renders/10 minutes

Redis Counter Operations

import { incrementCounter, getCounter } from '@/lib/redis';

// Increment with TTL (for rate limiting)
const count = await incrementCounter('ratelimit:user:123', 60); // 60s TTL
// Returns number or null if Redis unavailable

const current = await getCounter('ratelimit:user:123');

Redis Cache Adapter (lib/redis-cache-adapter.ts)

For structured cache with prefix namespacing and stats:

import { RedisCacheAdapter } from '@/lib/redis-cache-adapter';

const cache = new RedisCacheAdapter('myfeature', 3600000); // prefix, TTL in ms

await cache.set('key', { data: 'value' });
const result = await cache.get<{ data: string }>('key');
await cache.delete('key');
await cache.invalidatePattern(/project-.*-budget/);

const stats = await cache.getStats();
// { size, entries, hits, misses, hitRate, evictions }

Cache adapter details:

• Wraps values in CacheEntry<T> with { value, expiresAt, size }
• Uses redis.setex() with seconds-based TTL
• Uses SCAN instead of KEYS for pattern operations (safe for production)
• Tracks hit/miss stats in memory

Query Cache (lib/query-cache.ts)

Specialized cache for LLM query responses with Redis + in-memory redundancy:

import { getCachedResponse, cacheResponse } from '@/lib/query-cache';

// Check cache before calling LLM
const cached = await getCachedResponse(query, projectId, documentIds);
if (cached) return cached;

// Cache response after LLM call
await cacheResponse(query, response, projectId, documentIds, complexity, model);

Features: semantic similarity matching (Jaccard > 0.7), construction term normalization, dynamic TTL (48h standard, 72h high-value), LRU eviction prioritizing low-value entries.

Anti-Patterns

• Never import ioredis directly in feature code — use lib/redis.ts or lib/redis-client.ts
• Never use redis.keys('*') in production — use SCAN with count limits
• Never assume Redis is available — always handle the null/false return
• Never store large blobs (>1MB) in Redis — use S3/R2 for file storage
• Never create new Redis connections per request — use the singleton

Quick Reference