Bug 1242558

Cross-filed case #01476059 on the Red Hat customer portal.

Hello,

The customer logs are attached in the case #01476059, but they are around 116MBs ( i can't attach them here)

Do you need any specific files?

Release    :  Red Hat Enterprise Linux Server release 7.1 (Maipo)
Kernel     :  3.10.0-229.7.2.el7.x86_64

vdsm	    : Not Installed          	 libvirt     : 1.2.8-16.el7_1.3      
qemu-img   : 1.5.3-86.el7_1.2       	 qemu-kvm    : 1.5.3-86.el7_1.2

The only code paths where libvirt can log that error are the virDomainUndefineFlags API and the migration APIs with the use of the VIR_MIGRATE_UNDEFINE_SOURCE flag.

So from libvirt's point of view everything is fine, somebody must have asked libvirt to undefine the machine.

I do not see any logs related to libvirt in the customer case, but this is probably done by a different application. Is there anything running that could ask libvirtd to undefine a machine?

Ján, could that be caused by /usr/lib/ocf/resource.d/heartbeat/VirtualDomain
somehow? We added the virtual machine to pacemaker like this:

mkdir -p /var/lib/libvirt/qemu/pacemaker
chown qemu:qemu /var/lib/libvirt/qemu/pacemaker
pcs cluster cib vm_cfg
pcs -f vm_cfg \
  resource create vm ocf:heartbeat:VirtualDomain \
    config=/etc/libvirt/qemu/vm.xml \
    snapshot=/var/lib/libvirt/qemu/pacemaker
pcs -f vm_cfg \
  resource op remove vm monitor interval=10 timeout=30   # See: RHBZ#1031141
pcs -f vm_cfg \
  resource op remove vm start interval=0s timeout=90000  # See: RHBZ#1031141
pcs -f vm_cfg \
  resource op remove vm stop interval=0s timeout=90000   # See: RHBZ#1031141
pcs -f vm_cfg \
  resource op add vm monitor interval=60s timeout=30s
pcs -f vm_cfg \
  resource op add vm start interval=0 timeout=120s
pcs -f vm_cfg \
  resource op add vm stop interval=0 timeout=120s
pcs -f vm_cfg \
  constraint colocation add vm libvirtd INFINITY
pcs -f vm_cfg \
  constraint order libvirtd then vm
pcs cluster cib-push vm_cfg

And VirtualDomain_Start() calls verify_undefined() which runs itself the
"virsh undefine <domain>" command.

Created attachment 1056605 [details]
Diff between VirtualDomain from RHEL 6 and 7

The "virsh undefined" stuff was newly added with RHEL 7 - and as it works
on RHEL 6, this could be the cause...maybe?

The undefine call was added for bug 1016140:
https://github.com/ClusterLabs/resource-agents/commit/f00dcaf19

The following upstream change copies the file back into the original location if it disappeared during the undefine:
https://github.com/ClusterLabs/resource-agents/commit/897c03a3
which effectively makes the undefine last only until the next libvirtd restart.

If the domain needs to be undefined for the VirtualDomain agent, I think the config should be stored somewhere else, not in libvirt's /etc/libvirt/qemu - all the machines there are marked as defined when the libvirt daemon starts up.

I am not sure if 897c03a3 is really a good fix for this issue, given that it
is messing around dynamically in /etc/libvirt and /tmp. Note that e.g. in our
environment, /etc/libvirt, /var/lib/libvirt and /var/log/libvirt reside on a
DRBD device and we currently also seem to loose VMs during pacemaker takeover.

https://github.com/ClusterLabs/resource-agents/commit/aaadbdd79d0748f3d316e474a1fd39e8aef87cc5#diff-c14cdbae624e591a873d76bd66e96498

That patch just seems to copy the config back if it disappears during undefine.
If the config file disappears during undefine, it should stay deleted - otherwise what's the point of undefining it?

This is to keep the configuration when doing undefine as virsh removes it if it is in the libvirt-directories, but not if you put it outside of those directories, so this is a patch to avoid users upgrading from earlier versions losing their configuration.

This is how it is done in upstream, and you can read the discussion/reasoning here: https://github.com/ClusterLabs/resource-agents/issues/487.

Step-by-step reproduction:
Keep a backup of vm.xml, so it doesnt get lost in the Before test.

Before:
# rpm -q resource-agents
resource-agents-3.9.5-54.el7_2.6.x86_64
# pcs resource disable vm
# virsh define /etc/libvirt/qemu/vm.xml
# pcs resource enable vm
# ls /etc/libvirt/qemu/vm.xml
ls: cannot access /etc/libvirt/qemu/vm.xml: No such file or directory

After:
# rpm -q resource-agents
resource-agents-3.9.5-61.el7.x86_64
# pcs resource disable vm
# virsh define /etc/libvirt/qemu/vm.xml
# pcs resource enable vm
# ls /etc/libvirt/qemu/vm.xml 
/etc/libvirt/qemu/vm.xml

I have verified that xml file defining virtual machine does not disappear from
/etc/libvirt/qemu/ after the machine is started with resource-agents
3.9.5-79.el7.x86_64.

----

common environment:
# pcs resource create vm ocf:heartbeat:VirtualDomain config=/etc/libvirt/qemu/rhel-7.xm
# pcs resource show vm
 Resource: vm (class=ocf provider=heartbeat type=VirtualDomain)
  Attributes: config=/etc/libvirt/qemu/rhel-7.xml
  Utilization: cpu=2 hv_memory=1024
  Operations: start interval=0s timeout=90 (vm-start-interval-0s)
              stop interval=0s timeout=90 (vm-stop-interval-0s)
              monitor interval=10 timeout=30 (vm-monitor-interval-10)
# pcs resource disable vm
# ls -l /etc/libvirt/qemu/rhel-7.xml
-rw-------. 1 root root 4004 26. čec 10.59 /etc/libvirt/qemu/rhel-7.xml
# virsh define /etc/libvirt/qemu/rhel-7.xml

before the fix (resource-agents-3.9.5-54.el7.x86_64)
----------------------------------------------------
# pcs resource enable vm
# ls -l /etc/libvirt/qemu/*.xml
ls: cannot access /etc/libvirt/qemu/rhel-7.xml: No such file or directory


after the fix (resource-agents-3.9.5-79.el7.x86_64)
---------------------------------------------------
# pcs resource enable vm
# ls -l /etc/libvirt/qemu/*.xml
-rw-------. 1 root root 4004 26. čec 10.59 /etc/libvirt/qemu/rhel-7.xml

# pcs resource  | grep vm
 vm     (ocf::heartbeat:VirtualDomain): Started kiff-01.cluster-qe.lab.eng.brq.redhat.com

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://rhn.redhat.com/errata/RHBA-2016-2174.html