That's only because you've drawn such tight boundaries around your definition of "distributed system" that "User Experience" is actually "API design".
The same principle applies: The system can't do everything magically, so it exposes some of that through the API. I mean, I doubt the GFS API is simply a read() and write() method that are always guaranteed to work instantly.
Instead you start posing the same kind of "Hey, I need you to make a decision here" tradeoffs that the author is talking about:
"Sometimes X will happen, if so you need to check Y and try to recover with Z, unless you didn't really care about W. If you don't want to recover and would rather discard, make sure you call Q, otherwise we'll eventually destroy it after T minutes."
I would have to disagree. One of the main benefits of distributed systems is that you don't need to know that in reality your file system is sharded on to 5 different disks at 2 different colocs.
I understand that sometimes you need to manually recover, but OP sweepingly states that distributed systems is a UX problem, which I think is a ridiculous reduction and equivalent to saying "well this shit is hard so let's just spill the guts of the system internals up to the user"
