Cache שכתוב לא נכון = data stale, bugs מוזרים. Cache שכתוב נכון = 10x performance. הנה איך.

שכבות ה-Cache

(1) Browser cache: HTTP headers, service workers.

(2) CDN: static assets, edge caching.

(3) Application cache: Redis, Memcached.

(4) Database cache: query result cache.

(5) In-memory: local server memory.

Browser Cache

Static assets (JS, CSS, images): Cache-Control: public, max-age=31536000, immutable.

Hash in filename: main.abc123.js. שינוי = filename חדש = cache invalidation.

HTML: Cache-Control: no-cache. Always revalidate.

CDN Cache

Cloudflare, CloudFront, Fastly, Vercel Edge.

Static: cache everything.

Dynamic: cache by URL + query params + cookies. Careful with user-specific content.

Purge: API call to invalidate. Or short TTL + revalidate.

Redis Cache

Use cases: session storage, rate limiting, computed results, aggregations.

Patterns: cache-aside (check cache first, fetch from DB if miss), write-through (write to cache + DB simultaneously), write-behind (write to cache, async write to DB).

Recommended: cache-aside. פשוט וbetter consistency.

Cache Invalidation – הבעיה הכי קשה

Approaches:

(1) TTL: set expiration. Simple but data can be stale.

(2) Event-based: כשdata משתנה, מזמן invalidation. Complex but fresh.

(3) Version keys: include version בkey. update version = invalidate.

Cache Keys

Good format: entity:id:field.

Examples: user:12345:profile, product:widget-123:pricing.

Include version: user:12345:profile:v2. Update version instead of invalidate.

Serialization

JSON: readable, universal. Slower.

MessagePack, Protocol Buffers: faster, smaller. Requires schema.

For most projects: JSON is fine.

Memory Management

Redis eviction policies:

allkeys-lru: evict least recently used. Default for most use cases.

volatile-ttl: evict keys with shortest TTL.

noeviction: return error when memory full. Only for critical data.

Monitoring

Hit rate: % של requests שהgot cache. Target: 80%+.

Memory usage: warn at 70%, alert at 85%.

Slow commands: any command >100ms.

Connection pool: watch for exhaustion.

Common Mistakes

Caching everything: overhead > benefit for rarely-accessed data.

Ignoring cache stampede: קreadthundering herd – זכ many requests לfetch same missing key. Use locks or singleflight.

No monitoring: cache breaks silently. Metrics essential.

מבוסס על פרויקטים אמיתיים

המדריך הזה מבוסס על עבודה עם:

• Platforma365
• Podd

קריאה נוספת

אם המדריך הזה עזר לכם, אולי תרצו לקרוא גם על פיתוח SaaS מותאם – המדריך המקיף שלנו לתחום.