Error recovery with data on NFS

[Mihai Moldovan]

Hi


I have a few VMs managed over libvirt, whose image files are being served via
NFS (4.2).

This connection is not particularly stable (driver or hardware issues,
probably). Whenever it breaks off, NFS connections seem to be severed and the
file descriptors used by Qemu likewise seem to be invalidated (possibly, I don't
really know).

I'm not quite sure why that happens in the first place, given that my mount
options are pretty standard:
rw,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,local_lock=none
(especially notice "hard", which I would assume to hold the connection open
until data flows again).

To counter this issue, I started giving Qemu the rerror=stop,werror=stop options
on the disk devices, which worked to a degree, but not exactly the way I would
have liked it. The VM is correctly stopped on read or write errors, but it's
impossible to resume operation. Resuming just leads directly back to the paused
(I/O error) state. The only option is to forcefully stop the VM and start it up
again, which means I'm back to square one.

Is my use case so weird that there's no useful recovery option? I surely can't
be the only person serving images via NFS over an unreliable connection?

I've found a rather old mailing list thread in which it was explained that if
FDs go stale, there's nothing Qemu can do to rectify the issue (at least not
without data loss). While that generally makes sense to me, isn't the whole
point of NFS that clients can just reconnect after disconnects transparently,
without losing open FDs? Whenever Qemu stops, I certainly can still see the FDs
being open, so they obviously weren't completely closed.



Mihai

OpenPGP_signature
Description: OpenPGP digital signature