Daniel Feenberg
2013-05-04 19:08:43 UTC
When we change the exportfs file on our FreeBSD 9.1 fileserver and signal
mountd to reread the file:
kill -HUP `cat /var/run/mountd.pid`
it kills the jobs on clients that have files open on the fileserver. They
terminate with an I/O error. The same thing happens if NFS is restarted.
This is pretty inconvenient for users (and us). Is there a way around
this? We have noticed that a Linux fileserver can restart nfs without
distrubing clients (other than a short pause). The Linux restart doesn't
restart the locking mechanism - is that the difference? We could do
without locks, even without NFSv4, for that matter, if it would let us
change exports without disturbing users. Perhaps there is an NFS shutdown
procedure that we should be using?
Daniel Feenberg
NBER
mountd to reread the file:
kill -HUP `cat /var/run/mountd.pid`
it kills the jobs on clients that have files open on the fileserver. They
terminate with an I/O error. The same thing happens if NFS is restarted.
This is pretty inconvenient for users (and us). Is there a way around
this? We have noticed that a Linux fileserver can restart nfs without
distrubing clients (other than a short pause). The Linux restart doesn't
restart the locking mechanism - is that the difference? We could do
without locks, even without NFSv4, for that matter, if it would let us
change exports without disturbing users. Perhaps there is an NFS shutdown
procedure that we should be using?
Daniel Feenberg
NBER