XenAPI doesn't clean up failed build's VHDs

[Josh Kearney]

Failed build's VHDs continue to remain on the file system:

[root@xs ~]# ls -l /var/run/sr-mount/e0f61540-2d8e-e98b-a043-f7a6f0038ac1/ | wc -l
11
[root@xs ~]# xe vm-list
uuid ( RO) : 8aa565eb-6c5d-b328-7f4c-19aa9cbc46ec
     name-label ( RW): Nova
    power-state ( RO): running
uuid ( RO) : 7a570ee8-1d5b-467b-9c12-4a58f92fe2bc
     name-label ( RW): Control domain on host: xs
    power-state ( RO): running
uuid ( RO) : 068d3d7d-1249-2332-a9da-8fe1731e56b3
     name-label ( RW): XenServer DDK
    power-state ( RO): halted

[Salvatore Orlando (salvatore-orlando)]

This could be a regression of bug #694935

However the original bug fix took care only of VDIs created from raw images in Glance/Objectstore.
The code for VHD download was added concurrently with this fix, and therefore currently lacks 'protection' against instance build failure. Therefore, if the bug has been observed while building instance from VHDs that would explain it. Otherwise it's a regression.

[Johannes Erdfelt (johannes.erdfelt)]

Rick Harris has made a lot of changes recently to fix problems where VDIs were leaking on failed builds:

https://review.openstack.org/#change,4713
https://review.openstack.org/#change,4731
https://review.openstack.org/#change,4836

I'm going to mark this as Fix Committed for now because of all of the work he has done. If VDIs are still seen leaking, please reopen (or open new bug) with details on the new failure.