Bug 1207808

+++ This bug was initially created as a clone of Bug #1206722 +++

+++ This bug was initially created as a clone of Bug #1206355 +++

Description of problem:

After completing an active layer merge, vdsm's representation of the new volume chain might not be properly synchronized.  When this happens, engine will report a failure to delete the snapshot even though libvirt successfully merged it.


Version-Release number of selected component (if applicable):
libvirt-1.2.8-16.el7_1.2.x86_64 from http://download.devel.redhat.com/brewroot/packages/libvirt/1.2.8/16.el7_1.2/x86_64/ which is a candidate build for RHEL-7.1 zStream

vdsm-4.16.12-28.gitf03bb74.el7.x86_64 : 3.5 branch tip

How reproducible: Intermittent but easy to reproduce


Steps to Reproduce:
1. Create a VM with 3 BLOCK disks (1 thin and 2 preallocated)
2. Start the VM
3. Create a snapshot
4. Delete the snapshot

Actual results:
Snapshot may fail to delete on at least one disk


Expected results:
Snapshot removal is successful


Additional info:
The vdsm log reveals errors such as the following:

Thread-50::ERROR::2015-03-26 10:53:09,314::sampling::488::vm.Vm::(collect) vmId=`b3fbc637-9cc9-4b15-ba94-5a2ee1607785`::Stats function failed: <AdvancedStatsFunction _highWrite at 0x3133870>
Traceback (most recent call last):
  File "/usr/share/vdsm/virt/sampling.py", line 484, in collect
    statsFunction()
  File "/usr/share/vdsm/virt/sampling.py", line 359, in __call__
    retValue = self._function(*args, **kwargs)
  File "/usr/share/vdsm/virt/vm.py", line 292, in _highWrite
    self._vm.extendDrivesIfNeeded()
  File "/usr/share/vdsm/virt/vm.py", line 2537, in extendDrivesIfNeeded
    extend = [x for x in self._getExtendCandidates()
  File "/usr/share/vdsm/virt/vm.py", line 2489, in _getExtendCandidates
    capacity, alloc, physical = self._dom.blockInfo(drive.path, 0)
  File "/usr/share/vdsm/virt/vm.py", line 689, in f
    ret = attr(*args, **kwargs)
  File "/usr/lib/python2.7/site-packages/vdsm/libvirtconnection.py", line 111, in wrapper
    ret = f(*args, **kwargs)
  File "/usr/lib64/python2.7/site-packages/libvirt.py", line 646, in blockInfo
    if ret is None: raise libvirtError ('virDomainGetBlockInfo() failed', dom=self)
libvirtError: invalid argument: invalid path /rhev/data-center/dee92f86-2fef-4edf-ac51-135d12646bde/01f563fa-1aef-41ab-92d3-0d6561d4a731/images/31cfd267-e627-4cfe-8f33-f66b0db627c3/1863e178-c40a-42eb-803a-3d1f1ac6658a not assigned to domain

After talking to Eric Blake on IRC, it looks like another problem with libvirt where the domain XML is not updated before the virDomainBlockJobAbort API (used for pivoting an active layer merge) returns.  Therefore, our volume chain sync code gets an old value for the disk path which causes _highWrite to fail in this manner.