Why can restarting a whole cache cluster take down the database behind it, and how do you avoid that?

Answer

An empty cache misses on every read, so the database suddenly gets the full read load instead of a few percent of it. Restart nodes one at a time, warm the cache with the hottest keys first, or admit traffic gradually, and have the database shed load rather than fall over.

Review this in your daily deck → All cards in Caching