Please try to do a defragmentation and do not disable it for the data location.
See also
The problem with not reliable readability/writeability was there before, just checkers did not have a timeout, so for partial hanging it was not effective and some such nodes were disqualified in the past.
Now the timeout is in place and crashes node in such half-hanging state.
However, you may disable it, and make it behave as before - just increase a timeout to some big number like a week.
But I think it’s better to know that there is some problem with disk subsystem than close eyes on it.