What is a Cache Stampede?

What is a Cache Stampede?

• A. A failure mode where many concurrent requests miss an expired cache entry and hit the database simultaneously.
• B. A cache pattern where writes are bounced to DB.
• C. A mismatch between cache and DB due to stale data.
• D. A round-trip optimization technique.

Answer: (A)

A cache stampede happens when an item in the cache has expired and a burst of requests all miss at the same time, each trying to recompute or fetch the data from the underlying store. Because they all observe a miss, they all hit the database (or the slow path) concurrently, which can overwhelm the database, spike latency for many users, and cause a ripple of performance problems. The scenario described matches this idea precisely: many concurrent requests miss an expired cache entry and hit the database simultaneously. Understanding this helps you see why simply relying on a cache without coordination is risky after expirations. To prevent stampedes, you can coordinate refreshes by locking per key so only one request refreshes and others reuse the refreshed value, implement a single-flight or request-coalescing mechanism, use refresh-ahead strategies (pre-warming or background refreshing), or apply smarter TTLs and jitter to spread out refresh work.

A cache stampede happens when an item in the cache has expired and a burst of requests all miss at the same time, each trying to recompute or fetch the data from the underlying store. Because they all observe a miss, they all hit the database (or the slow path) concurrently, which can overwhelm the database, spike latency for many users, and cause a ripple of performance problems. The scenario described matches this idea precisely: many concurrent requests miss an expired cache entry and hit the database simultaneously.

Understanding this helps you see why simply relying on a cache without coordination is risky after expirations. To prevent stampedes, you can coordinate refreshes by locking per key so only one request refreshes and others reuse the refreshed value, implement a single-flight or request-coalescing mechanism, use refresh-ahead strategies (pre-warming or background refreshing), or apply smarter TTLs and jitter to spread out refresh work.