If you can't get the team who owns the data to meet the needed uptime, then you probably shouldn't start the project in the first place. It's a classic case of

If you can't get the team who owns the data to meet the needed uptime, then you probably shouldn't start the project in the first place. It's a classic case of organizational differences in priority; solving them with tech pretty much never works. Any caching layer will be more expensive than fixing the root cause, and add consistency issues. As an IC, you'll feel pressure to make it work the best you can with what you have, but sometimes the answer is it's doomed from the start if the right folks aren't bought in. Some tips if you have to proceed: 1) Be clear you can promise a SLO for you app, but not it's dependancies. If the dependancies are down, some features won't work. 2) Have honest clear errors that explain the clear root cause. Cloudflare's "Web server is down" page is a great example. It pretty clearly say's "we're up, but they aren't".