I don't think end customer cares about why is his experience broken. Following a process that accepts things being broken, but shifts the responsibility doesn't

I don't think end customer cares about why is his experience broken. Following a process that accepts things being broken, but shifts the responsibility doesn't fix anything. Op's question is a great sign of ownership and customer obsession on his side. The actuall solution is tough, but I think something can be improved. First of all I would try to measure and root cause the problem. It's hard for me to believe that the DB has 90% availability, I'm not aware of any commercial DB that's this bad. I would avoid creating shards of db, it will lead to hard to debug data inconsistencies. In this regard, cache with reasonable TTL is a better choice wince it guarantees eventual consistency of the data. I would also take a deep look into retry approaches you can take, something like exponential backoff with jitter and properly adjusted number of retries can greatly increase your availability.