Created attachment 1040971 [details] VDSM log of failed start Description of problem: When a random number generator (virtio-rng) is added to an existing VM it won't start any longer. Engine webui shows: 2015-Jun-19, 15:39 Failed to run VM example-vm on Host vhost. 2015-Jun-19, 15:39 VM example-vm is down with error. Exit message: 'device'. 2015-Jun-19, 15:39 VM example-vm was started by user@FreeIPA (Host: vhost). Version-Release number of selected component (if applicable): On engine: otopi.noarch 1.3.2-1.el7.centos otopi-java.noarch 1.3.2-1.el7.centos ovirt-engine.noarch 3.5.3.1-1.el7.centos ovirt-engine-backend.noarch 3.5.3.1-1.el7.centos ovirt-engine-cli.noarch 3.5.0.6-0.1.20141107.gitcf7f1a1.el7.centos ovirt-engine-dbscripts.noarch 3.5.3.1-1.el7.centos ovirt-engine-extension-aaa-ldap.noarch 1.0.2-1.el7 ovirt-engine-extension-aaa-misc.noarch 1.0.0-1.el7 ovirt-engine-extensions-api-impl.noarch 3.5.3.1-1.el7.centos ovirt-engine-jboss-as.x86_64 7.1.1-1.el7 ovirt-engine-lib.noarch 3.5.3.1-1.el7.centos ovirt-engine-restapi.noarch 3.5.3.1-1.el7.centos ovirt-engine-sdk-python.noarch 3.5.2.1-1.el7.centos ovirt-engine-setup.noarch 3.5.3.1-1.el7.centos ovirt-engine-setup-base.noarch 3.5.3.1-1.el7.centos ovirt-engine-setup-plugin-ovirt-engine.noarch 3.5.3.1-1.el7.centos ovirt-engine-setup-plugin-ovirt-engine-common.noarch 3.5.3.1-1.el7.centos ovirt-engine-setup-plugin-websocket-proxy.noarch 3.5.3.1-1.el7.centos ovirt-engine-tools.noarch 3.5.3.1-1.el7.centos ovirt-engine-userportal.noarch 3.5.3.1-1.el7.centos ovirt-engine-webadmin-portal.noarch 3.5.3.1-1.el7.centos ovirt-engine-websocket-proxy.noarch 3.5.3.1-1.el7.centos ovirt-guest-agent-common.noarch 1.0.10-2.el7 ovirt-host-deploy.noarch 1.3.1-1.el7 ovirt-host-deploy-java.noarch 1.3.1-1.el7 ovirt-image-uploader.noarch 3.5.1-1.el7.centos ovirt-iso-uploader.noarch 3.5.2-1.el7.centos On host: vdsm.x86_64 4.16.20-0.el7.centos vdsm-cli.noarch 4.16.20-0.el7.centos vdsm-jsonrpc.noarch 4.16.20-0.el7.centos vdsm-python.noarch 4.16.20-0.el7.centos vdsm-python-zombiereaper.noarch 4.16.20-0.el7.centos vdsm-xmlrpc.noarch 4.16.20-0.el7.centos vdsm-yajsonrpc.noarch 4.16.20-0.el7.centos How reproducible: Every time Steps to Reproduce: 1. Enable Random Number Generator on cluster level 2. Enable Random Number Generator on VM 3. Start VM Actual results: VM does not start. It fails with error "Exit message 'device'" Expected results: VM is started and has a virtio-rng device attached. Additional info: Logfile from VM startup ist attached (vdsm.log). According to discussion in #vdsm the problem is that the device specification misses the 'device' attribute, and VDSM depends on it. There are quite some places on which VDSM checks for this attribute, so it is hard and counterproductive to workaround it in VDSM
Hi Patrick, is it possible that you check if https://gerrit.ovirt.org/#/c/40095/ solves the issue? Thanks!
This is exactly the issue that I've found and fixed with change https://gerrit.ovirt.org/#/c/40095/. vdsm 4.16.20 doesn't include this change. I cannot reproduce the issue with the latest build of vdsm from master. Please, try it with your setup.
Thanks Shmuel, lets try to fix for the next release.
I already tried that before opening this bug. The mentioned gerrit change https://gerrit.ovirt.org/#/c/40095/ seems to at least need https://gerrit.ovirt.org/#/c/38982/, too. Both do not cleanly apply to vdsm 4.16.20, there are more changes needed. An attempt to backport both changesets to 4.16.20 is attached (vdsm-virt-rng-device.diff). But this does not change anything. The error is still the same. VDSM-Log with the patch applies is also attached to this bz.
Created attachment 1041709 [details] backport of gerrit changes 38982 and 40095
Created attachment 1041710 [details] VDSM log of failed start with patch applied
After discussion on irc this was partly solved. Starting a vm with attached virtio-rng works, when virtio-console is detached. Vm start still fails with the same error message, when virtio-console is attached. So its the combination of virtio-console and virtio-rng that renders the vm unusable. Both virtio devices are usable when added separately.
seems the issue with both rng and virtio-console happens in 3.6 as well, re-targeting.
Verified. rhevm-3.6.1.2-0.1.el6 libvirt-client-1.2.17-13.el7_2.2.x86_64 qemu-kvm-rhev-2.3.0-31.el7_2.4.x86_64 vdsm-4.17.13-1.el7ev.noarch sanlock-3.2.4-1.el7.x86_64 Verification scenario: 1. Enable cluster rng 2. Enable rng on VM and attach virtio-console. 3. Run VM, verify rng is working. 4. Power off VM, detach virtio-console. 5. Run VM. 6. Verify VM is running. Verify rng is working.
oVirt 3.6.0 has been released and the bz verified, moving to closed current release.