I think it is scary about Linux. They are still wrestling with reliable error reporting for buffered IO after decades, and while the changes in 4.13 are an improvement in some ways they make some highly questionable assumptions that don't seem to be based on POSIX. I'm not following it in detail but fortunately it sounds like something might be done to improve the situation on the Linux side?

I can't read it because it's paywalled, but how is the operating system's fsync behavior "fatally scary" with regards to PostgreSQL? How does MySQL handle durability in these cases? What's your point?