Prolonged degradation after Redis failover

At 11:20 your primary Redis (ElastiCache) cluster failed over to a replica after a node fault. The failover took 25 seconds. During and after it, your search service didn't just blip, it stayed degraded for 8 minutes: connection errors to Redis for 25s, then a sustained period where origin DB CPU was pinned and p99 was 10x normal, slowly recovering. Your Redis client has a 50ms connect timeout but retries with no backoff. How do you triage and what are the failure modes to fix?