The central question to answer is whether your boss is held responsible for failure to meet the SLA or whether you alone are. Assuming your boss is held respon

The central question to answer is whether your boss is held responsible for failure to meet the SLA or whether you alone are. Assuming your boss is held responsible, I would design the system to treat the database like an unreliable API. Wrap up all operations in a DAL and interact with a local database. Do regular transactional sync operations in the background. In case of data conflict, determine who wins and why in each case. Don't spend too much time on the design. Just enough to get the point across. Bring the design to your boss and say this is what I'm thinking given that we have an SLA but this critical dependency does not and has an observed uptime of 90%. And then let your boss decide what to do. If you are the only one who will be held responsible and your boss doesn't support your design or go to bat for you, you work in a bad environment and should find a new job.