We should implemented a tiered fallback cache for entries (at least for entitlements, maybe not auth?):

Happy Path:

• cache miss
• call service
• populate short TTL cache and long (fallback) TTL cache (hours)
• cache hit
• serve from short TTL cache
• repeat

Sad Path:

• cache miss
• call service
• populate short TTL cache and long (fallback) TTL cache (hours)
• cache hit
• serve from short TTL cache
• cache miss
• service call fails
• serve from long TTL cache as "last known state"
• ensure we check the live service at set intervals to prevent from serving from the long cache longer than necessary