Description of problem: NFS soft mounts can lead to silent data corruptions on write errors. It will be wise to change it in rhev-2.2 although late in the game. Even this change might expose other bugs, those shouldn't be drastic as the data corruption one.
Using hard mount however, (as I understand the difference between hard and soft mounts), reverts us back to the previous behaviour that the qemu_kvm process hangs on NFS disconnect. In the past the process was destroyed when this happens for more then 30secs. What is the current behaviour? We also had another check that run lstat on the NFS mount path - is this one still there? if so we can use the fact that this check does not return to understand that this is NFS freeze and refrain from killing the VM. There may be other implications that I may not thought of due to this change in behaviour.
Although there is no known issue with this in our use case it is still recommended to add support for hard mounts as it is much more widely tested at the nfs level. this requires however supporting intr properly and actually interrupting processes in the control path.
Closing this bug as duplicate on the upstream bug (this one was opened for rhev and contains irrelevant info). *** This bug has been marked as a duplicate of bug 810082 ***