Cache Stampede

🔴 Q. What is a cache stampede, and how do you stop one hot key overwhelming the database?

Answer. A cache stampede happens when a popular entry expires and many requests miss together, so they all fetch or recompute the same value and send a burst to the backing store.

The protections solve different parts of the problem:

• Request coalescing / single-flight: one caller refreshes; others wait.
• Jittered TTLs: spread expirations so many keys do not fail at once.
• Stale-while-revalidate: serve bounded stale data during one refresh.
• Early refresh: probabilistically refresh hot data before expiry.

A distributed lock can coordinate hosts, but it needs a short lease and safe ownership; otherwise it becomes another outage source. Load shedding and database limits remain the final backstop because no cache strategy can make the backing service survive arbitrary demand.

Found a mistake? Edit this page on GitHub.