XenAPI: Fix race-condition with cached images.

The core problem is that XenServer's `VDI.copy` call drops the
destination file directly into the SR. This means that half-completed
files are visible with no way to distinguish these from fully-copied
files.

We had some code that attempted to mitigate this issue by checking
physical_utilisation against an expected value. The problem with this
code is that it didn't account for VDI chaining where the
physical_utilisation would not necessarily match the parent.

The net effect of this was that 'cloned' VDIs would never be found
because their physical_utilisation was far below what was expected.

The work around is to create our own `_safe_copy_vdi` which is isolated
and atomic. Long term, `VDI.copy` should be fixed so that half-completed
files are never stored in the SR.

Change-Id: I6eb3cb5259f9ee1c7394e58f76105a8b39bfc720

1 files changed, 9 insertions, 0 deletions

diff --git a/nova/tests/test_xenapi.py b/nova/tests/test_xenapi.py
index a5b88bd0d..f858fed25 100644
--- a/nova/tests/test_xenapi.py
+++ b/nova/tests/test_xenapi.py
@@ -288,6 +288,15 @@ class XenAPIVMTestCase(stubs.XenAPITestBase):
         self.stubs.Set(vmops.VMOps, 'inject_instance_metadata',
                        fake_inject_instance_metadata)
 
+        def fake_safe_copy_vdi(session, sr_ref, instance, vdi_to_copy_ref):
+            name_label = "fakenamelabel"
+            disk_type = "fakedisktype"
+            virtual_size = 777
+            return vm_utils.create_vdi(
+                    session, sr_ref, instance, name_label, disk_type,
+                    virtual_size)
+        self.stubs.Set(vm_utils, '_safe_copy_vdi', fake_safe_copy_vdi)
+
     def tearDown(self):
         super(XenAPIVMTestCase, self).tearDown()
         fake_image.FakeImageService_reset()