Redis Outage: Fail-Open vs Fail-Closed

Your rate limiter enforces two policies against the same Redis cluster:

• General API limit: 1,000 req/min per API key, protects your own infrastructure from being overwhelmed.
• Login attempt limit: 5 attempts per 15 minutes per account, protects against credential stuffing / brute force.

Redis becomes unavailable for 6 minutes due to a failover.

1. What should happen to each policy during the outage, and why are the right answers different for the two?
2. Design a concrete fallback for the general API limit that avoids both "reject every request" and "allow unlimited requests" during the outage.
3. Six minutes is short. What would you do differently if Redis outages in this system typically last 30+ minutes?