Most consumer /hard drives/ lie about the result of fsync, as a 'performance optimization'.
It's generally possible to fsync, then cut the power before the data is physically on the disk.
It's generally possible to fsync, then cut the power before the data is physically on the disk.
There's also a Postgres contrib module called pg_test_fsync that tests various fsync modes on your hard drive, and if you get too high results, you can strongly suspect that the dist is lying.
On bigger hardware, like SAN storage arrays, the (redundant) batteries keep the whole thing running for a while after the loss of power.
My personal opinion is that raid controllers are of little value without battery backed cache.
I just use MD in front of 'enterprise sata' drives, as without doubling my total cost for storage, I can't get a raid card, as far as I can tell, which is better than MD, and really, if I was to double my storage cost, simply buying twice as many spindles would get me better bang for my buck, I think, than a hardware raid card that cost that much.
Of course that completely shafts your performance, but hopefully you were expecting that.