Red Hat Bugzilla – Bug 517938
Cannot install KVM guests over NFS (rtl8139 emulation)
Last modified: 2014-03-25 21:00:46 EDT
Description of problem: For some reason, KVM install attempts of RHEL5.4 on RHEL5.4 will not succeed if an NFS mount is used as the install source. Version-Release number of selected component (if applicable): RHEL5.4 x86_64 -162 kernel kvm-83-105.el5 How reproducible: Every time Steps to Reproduce: 1. Run virt-manager 2. Install a new KVM guest 3. Use NFS URL as install source 4. Wait for a very very long time with no installed system Actual results: on screen F4, these messages repeated over and over again server $SERVERNAME not responding, till trying server $SERVERNAME not responding, till trying server $SERVERNAME OK server $SERVERNAME OK server $SERVERNAME OK server $SERVERNAME not responding, till trying server $SERVERNAME not responding, till trying server $SERVERNAME not responding, till trying server $SERVERNAME OK server $SERVERNAME not responding, till trying Expected results: Normal install Additional info: I have a machine that can reproduce this easily.
Did you suffer from network issues? Does it still happen? It's a very common use case so I doubt it happens now.
Development Management has reviewed and declined this request. You may appeal this decision by reopening this request.
sorry for taking too long to reply (again). This bug seems to be related to virt-manager not using virtio by default when installing RHEL. The change to use virtio for RHEL-5.4+ was addressed at: https://bugzilla.redhat.com/show_bug.cgi?id=547380 Have you seen this issue recently, or there's some time since the last time you tested NFS installs? Do you see this issue if you make sure virtio is used for networking, when installing the guest?
I've just tried again with the latest 5.4 bits (yum update from RHN) and the problem still occurs. Lots of <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME OK <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME OK <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME not responding, still trying <5>nfs: server $SERVERNAME not responding, still trying with no data appearing on the ctrl-alt-F1 console. I'm installing 5.5 beta now and will let you know how that behaves.
Are you using virtio-net on the failing setup? Could you show the 'vish dumpxml' output for the failing guest?
The RHEL 5.5 beta does not have this problem. I chose the "RHEL5.4 or later" option when setting up the guest (also RHEL5.5 beta) through virt-manager. I don't see any NFS errors on any consoles. Do you want me to probe further into the failing 5.4 installs?
If it works on 5.5 beta, it's good news. However, the problem may be still present when non-virtio network device emulation is used. Having virtio working may make the bug have lower priority, but it would be interesting to investigate and fix it for the benefit of users that have guests without virtio support. Could you show the 'virsh dumpxml' output for the guest on the failing setup?
Actually, 'virsh dumpxml' for both cases (working and non-working setup) is interesting, so I can be sure it's failing because of missing virtio support. Thanks for the help.
Created attachment 390557 [details] virsh dumpxml 1 from system RHEL5.4 system exhibiting NFS mount problem Attaching virsh dumpxml 1 output from RHEL5.4 system experiencing the bug
Created attachment 390581 [details] visrh dumpxml 1 from RHEL5.5 beta system that works correctly
The system that works correctly us using virtio, the buggy system is using the default (rtl8139). Using rtl8139 emulation is not recommended unless the guest system doesn't have support for virtio or e1000. Lowering priority and severity.