On Thu, Dec 12, 2013 at 04:06:26AM +0000, Allen Belletti wrote:
Hi All,
Ever since moving from oVirt 3.1 (with Fedora image-based nodes) to oVirt 3.3 (and now
3.3.1) with CentOS 6.4 (now 6.5) nodes, I've been having a serious problem with our
Windows guests. To my knowledge, none of the Linux guests have ever been affected.
The problem appears when a Windows guest is rebooted. Occasionally this happens during
maintenance work but more often as a result of Microsoft updates which require a reboot.
The most recent batch of these that we allowed resulted in something like 35 of 40 Windows
guests losing networking.
What we see is that the VirtIO Ethernet Adapter is enabled but shows no evidence of
actually being connected to the network. The MAC never appears in our switches, etc. The
solution is to connect to the guest via console and manually disable, then enable, the
NIC. Problem solved.
I've found at least two other references to this issue:
http://forum.proxmox.com/threads/11160-Windows-2008-R2-Server-Standard-Ed...
(exactly our issue)
http://forum.proxmox.com/threads/16550-Windows-server-2012-blue-screen-vi...
(similar, though we never see a crash)
Needless to say, this is super annoying. Right now it seems my options seem to be:
1. Implement a script to reset the NIC as suggested in the first link.
2. Revert to e1000 but not sure how much that will hurt performance and/or whether it
will limit us to 1 Gbps throughput.
Has anyone experienced this? Would I likely have better luck on the VirtIO list?
We've been running VirtIO Ethernet 61.63.103.3000 from virtio-win-0.1-65.iso but I
just now found 61.65.104.7400 included in virtio-win-0.1-74.iso. Perhaps that will help
although I don't see anything relevant in the changelog.
Which version of the host kernel exposes this?
Which version (in ovirt-3.1) did not?
Would you copy your qemu command line?
I hope that Gal (CCed) and his friends can help if they get this info.