On Tue, Jun 25, 2013 at 02:32:36PM -0600, Dan Urist wrote:
> I have several hosts, each running multiple vservers. Each
> vserver mounts several NFS volumes, including /home, from
> a NetApp via NFS v.3.
> The /home volume filled completely yesterday (bad users!).
When NFS volumes fill up, the strangest things can happen.
Why? Well, mainly because nobody uses NFS in sync mode
(it would be incredibly slow because of the turnaround time)
and thus all write requests are basically confirmed before
they actually happen ...
When the filesystem is full, the server will still try to
complete the ongoing transfer, which usually brings down
the entire system to a grinding halt as requests pile up.
> Now, some but not all of my vservers are not reporting any
> of the NFS volumes anymore in df or mount, but they are
> still mounted.
The vservers or the userspace tools?
Check what /proc/mounts has to say about your nfs mounts.
> This isn't a major problem (for me at least); just a curiosity
> I thought was worth bringing up in case it points to anything
> more serious going on under the covers.
Might be as simple as something in userspace overeagerly
cleaning up /etc/mtab or some tool modifying mtab not
being able to write a backup copy and accidentially removing
the original ...
> The host OS is Debian Squeeze, the vservers are a mix of
> Squeeze and Ubuntu Lucid (I'm seeing the same behavior in
> both vserver OS's).
> The kernel and util-vserver are the versions from the
> standard squeeze repos.
Not commenting on that,
best,
Herbert
> --
> Dan Urist
> durist@ucar.edu
> 303-497-2459
Received on Thu Jun 27 20:47:18 2013