Question

Every hour, almost exactly on the hour, your homepage feed service shows a 10–15 second cliff: cache hit rate drops from 98% to near 0, origin DB CPU spikes to saturation, p99 jumps 30x, and a burst of 500s appears — then it recovers until the next hour. The feed is cached in Redis with a 3600s TTL, and a nightly batch job repopulates many keys at the same wall-clock minute, so a large set of keys share the same expiry boundary. When they expire together, every request for those keys misses and hits the DB simultaneously. Recent context: a refactor last week consolidated cache writes into a single hourly warm job. Diagnose and mitigate.