On Thu, 2008-03-27 at 09:28 +0100, Laurent Vaills wrote:
> Hi.
>
> On my Fedora box I have defined some NFS mount points like this in
> my /etc/fstab :
> <hostname>:/path/to/exported/dir /nfs/mount-point nfs
> user,hard,intr 0 0
>
> But sometimes the machine <hostname> is not powered up.
> When I run "yum -y upgrade" it freezes on "Running transaction test" .
> So I decided to strace it and I saw it stops at
> stat64("/nfs/mount-point", ....
>
> My first question is :
> - Why yum/rpm has to go in that directory ?
Is /nfs/mount-point the actual name of the directory? Is some other part
of your filesystem symlinked into some part of it?
> - What are the correct options to put in the fstab in order not to
> have any process frozen when they try access an NFS directory that
> does not respond ?
It freezes because that's how NFS hard mounts are defined to work when
the remote filesystem is not available. The process will hang for as
long as it takes for the filesystem to come online again. This preserves
Unix filesystem semantics at the cost of possibly having to wait
"forever".
You can use 'soft' instead of 'hard' in the mount options, but be sure
you know what the consequences are -- any access to a file on the
unavailable NFS filesystem will behave as if the file doesn't exist. Is
that what you really want, especially as you don't seem to know what
file(s) are being accessed? What happens if the server changes state in
the middle of the update process? For a yum update it probably isn't
that critical (anything it misses will likely be caught at some future
update) but of course the 'soft' option applies to *all* processes
accessing *all* files on the mounted system.
poc
