Redis Sentinel failover takes about 2 seconds beyond down-after setting, tests show

A software engineer deliberately killed the primary node 15 times across three in-memory database engines — Redis 8, Valkey 8, and Dragonfly — to measure real-world failover duration under Redis Sentinel. Tests on a one-primary, two-replica, three-sentinel Docker setup showed that write outage consistently equals the configured down-after-milliseconds value plus roughly two additional seconds for quorum voting and replica promotion. With the common default of 5,000 ms, all three engines restored writes in approximately 7.3–7.4 seconds, indicating the overhead belongs to Sentinel's election process rather than any specific engine. The single most controllable factor in reducing outage duration is tuning the down-after-milliseconds setting, which the author notes is rarely adjusted in production deployments. The full test lab, kill scripts, and reproducible benchmarks have been published on GitHub for independent verification.