11. Cache Invalidation Nightmares: Cache Avalanche, Cache Stampede, Thundering Herd & Cache Penetration • High-Level System Design (HLD): Distributed Systems at Scale • Rishabh Ki Kaksha

High-Level System Design (HLD): Distributed Systems at Scale › 11. Cache Invalidation Nightmares: Cache Avalanche, Cache Stampede, Thundering Herd & Cache Penetration

50 mins

← Prev Next →

Course Syllabus

Module 1: Distributed Systems Core & Scalability Fundamentals

01. Scalability Dimensions: Vertical Scaling vs Horizontal Scaling & Latency vs Throughput 02. CAP Theorem vs PACELC Theorem: Consistency, Availability & Partition Tolerance Tradeoffs 03. Consistency Models: Strong Consistency, Eventual Consistency, Read-Your-Writes & Monotonic Reads

Module 2: Networking, Load Balancing & Reverse Proxies

04. Layer 4 vs Layer 7 Load Balancing: Round Robin, Weighted, Least Connections & IP Hash 05. High-Availability Load Balancers: Keepalived, VRRP, Anycast DNS & Failover Clustering 06. API Gateway Pattern: Authentication, Rate Limiting, Request Routing & Circuit Breakers

Module 3: Database Scaling: Sharding, Replication & Distributed Transactions

07. Database Replication: Master-Slave vs Multi-Master & Synchronous vs Asynchronous Replication 08. Database Sharding: Horizontal Partitioning, Directory-Based Sharding & Rebalancing Hotspots 09. Distributed Transactions: Two-Phase Commit (2PC) vs Saga Pattern (Choreography vs Orchestration)

Module 4: Caching Architectures, CDN & Consistent Hashing

10. Caching Strategies: Cache-Aside, Read-Through, Write-Through, Write-Around & Write-Back 11. Cache Invalidation Nightmares: Cache Avalanche, Cache Stampede, Thundering Herd & Cache Penetration 12. Consistent Hashing with Virtual Nodes (Solving Cache Clustering & Dynamic Scaling) 13. Content Delivery Networks (CDN): Edge Caching, Dynamic Site Acceleration & Origin Shielding

Module 5: Message Brokers, Event-Driven Architecture & Stream Processing

14. Message Queues vs Event Streams: RabbitMQ vs Apache Kafka (Partitioning, Consumer Groups) 15. Exactly-Once Processing Semantics (EOS), Idempotency Keys & Deduplication Architecture 16. Distributed Unique ID Generators: Twitter Snowflake, UUIDv7 & Database Ticket Servers

Module 6: FAANG Classic HLD Architectural Case Studies

17. System Design Case Study 1: URL Shortener (TinyURL / Bitly) at 100M Daily Requests 18. System Design Case Study 2: Distributed Rate Limiter & Web Application Firewall 19. System Design Case Study 3: Real-Time Chat & Notification System (WhatsApp / Slack) 20. System Design Case Study 4: Social Media News Feed (Twitter/Instagram) with Fan-out on Write/Read 21. System Design Case Study 5: Video Streaming Platform (YouTube / Netflix) Transcoding Architecture 22. System Design Case Study 6: Ride Hailing Geospatial Dispatch Architecture (Uber / Ola)

Interactive Lesson Studio

# 11. Cache Invalidation Nightmares: Cache Avalanche, Cache Stampede, Thundering Herd & Cache Penetration

> High-Level Design (HLD) Notes by Rishabh - Topic 11

---

### 1. The 4 Cache Disasters & Production Mitigations 💥

```
┌─────────────────────────────────────────────────────────────┐
│                 THE 4 CACHE DISASTERS                       │
├───────────────────┬─────────────────────────────────────────┤
│ 1. AVALANCHE      │ All keys expire at same second ➔ DB Down│
│ 2. STAMPEDE       │ Hot key expires ➔ 100K requests hit DB  │
│ 3. PENETRATION    │ Hacker queries non-existent ID ➔ Hits DB│
│ 4. BREAKDOWN      │ Extreme single-item spike (Black Friday)│
└───────────────────┴─────────────────────────────────────────┘
```

---

### 2. Deep-Dive Fixes

1. **Cache Avalanche Fix (TTL Jitter):**
   * Har key ke TTL me random variation add karo: `TTL = 3600 + Math.random() * 300` seconds.
2. **Cache Stampede / Thundering Herd Fix (Distributed Mutex Locks):**
   * Sirf 1 worker thread Redis lock acquire karke DB query karega; baki 99,999 requests lock release hone ka wait karengi.
3. **Cache Penetration Fix (Bloom Filters & Null Caching):**
   * **Bloom Filter:** In-memory probabilistic data structure jo 100% certainty ke sath batata hai ki key DB me exist karti hi nahi hai!
   * **Null Caching:** Non-existent key ko `null` value ke sath 60 seconds ke liye cache karo (`SET key "null" EX 60`).

---

### 3. Master Comparison Matrix

---

### 4. Interview Ke 3 Sawal Is Topic Se

1. **Bloom Filter me False Positive aur False Negative ki kya probability hoti hai?** -> Bloom Filter me False Negatives **ZERO (0%)** hote hain (agar filter ne bola nahi hai, toh guaranteed nahi hai); False Positives low probability me ho sakte hain jo safe DB query me fall through karte hain.
2. **Probabilistic Early Expiration (XFetch algorithm) kaise kaam karta hai?** -> Key expire hone se theek pehle statistical probability calculate karke background thread se pre-fetch kar leta hai bina single user request ko block kiye.
3. **Two-Tier Caching (L1 Local Memory + L2 Redis) me cache sync kaise handle hoti hai?** -> Redis Pub/Sub ke zariye jab L2 update hota hai, toh saare application nodes apne local L1 memory cache ko invalidate karte hain.

---
**Next Topic:** 12. Consistent Hashing with Virtual Nodes (Solving Cache Clustering & Dynamic Scaling)

Loading lesson content...

Python 3.12

Full IDE

Live Execution Terminal

Click "Execute Live" to run code.
                

Lesson Mastery AI Quiz

AI Doubt Solver

CERTIFICATE OF COMPLETION

Rishabh Gehlot

PHP 8+ & Laravel 11 Master Track