Coit2019 - El Chatarras del código

HLD Interview Notes: Day 6/30 Topic: Cache Stampede, Cache Penetration and Cache Avalanche Your cache is supposed to protect the database. But what happens when caching itself creates a traffic problem? Three common failure scenarios every system design interview candidate should understand. 1. Cache Stampede Problem: A popular cache key expires, and thousands of requests try to fetch the same data from the database simultaneously. Example: A product page's cache expires during a sale. Every incoming request hits the database. Solutions: Request coalescing: Let one request fetch the data while others wait. Distributed locking: Allow one worker to refresh the cache. Background refresh: Update popular entries before they expire. 2. Cache Penetration Problem: Requests repeatedly ask for data that doesn't exist in the database, causing cache misses every time. Example: Bots request thousands of nonexistent product IDs. Every request reaches the database. Solutions: Cache negative results for a short TTL. Use a Bloom Filter to identify keys that definitely don't exist. Validate requests before querying the database. 3. Cache Avalanche Problem: Many cache entries expire simultaneously, causing a sudden surge of requests to the database. Example: Thousands of product entries share the same TTL and expire together. Solutions: Add random jitter to TTLs. Use staggered expiration times. Apply rate limiting and load shedding to protect the database. 4. Quick Comparison Stampede: Many requests for the same expired key. Penetration: Requests for nonexistent data. Avalanche: Many keys expire around the same time. 5. Interview Tip When diagnosing a cache problem, ask: Is one hot key causing too many requests? Are requests targeting data that doesn't exist? Are many keys expiring together? The answer helps you choose the right mitigation. FAANG Takeaway Caching is not just about hit rates. It's also about protecting the database when the cache fails to serve a request. Design for cache misses, expiration spikes, and malicious traffic before they become production incidents. Next: Day 7/30 Cache-Aside vs Read-Through vs Write-Through vs Write-Back