Create qcow2 image failed on instance with ephemeral disk when use_cow_images=0

[Ekaterina Shutova]

Detailed bug description:
 Filed to create qcow2 image on instance with ephemeral disk when use_cow_images set to 0.
Steps to reproduce:
1. Check use_cow_images=0 value in nova config on all computes
2. Start instance with ephemeral disk
3. umount /mnt in instance
4. On instance create qcow2 image with baking_file
   linked to target host device in ephemeral block device
   something like: qemu-img create -f qcow2
   -o backing_file=/dev/sda3,backing_fmt=raw /dev/vdb 20G
5. Change flavor or migrate instance
6. Check that /vdb is not linked to host device
Link to test case:
https://mirantis.testrail.com/index.php?/cases/view/1617642

Expected results:
 Test case passed

Actual result:
Links to failed runs:
https://mirantis.testrail.com/index.php?/tests/view/21296762
https://mirantis.testrail.com/index.php?/tests/view/21282306
 Test case failed with:
command = 'sudo qemu-img create -f qcow2 -o backing_file=/dev/dm-1,backing_fmt=raw /dev/vdb 20G'
verbose = True

    def check_call(self, command, verbose=True):
        ret = self.execute(command, verbose)
        if ret['exit_code'] != 0:
            raise CalledProcessError(command, ret['exit_code'],
> ret['stdout'] + ret['stderr'])

E CalledProcessError: Command 'sudo qemu-img create -f qcow2 -o backing_file=/dev/dm-1,backing_fmt=raw /dev/vdb 20G' returned non-zero exit status 1

E ['sudo: unable to resolve host server-test-ubuntu\n', 'sudo: qemu-img: command not found\n']

Reproducibility:
 100%
Workaround:
 NA
Impact:
 Test run failed.
Description of the environment:
On both Ceph and non-Ceph configuration

[Timofey Durakov (tdurakov)]

@Ekaterina Shutova, could you please describe steps to reproduce here instead of providing a link to testrail, marking an Incomplete until this bug description is updated

[Timofey Durakov (tdurakov)]

It looks like that the issue is related to installing qemu-utils on the instance:
https://github.com/Mirantis/mos-integration-tests/blob/master/mos_tests/nova/nova_test.py#L967-L971
It's a potential race condition when ssh become available before qemu-utils is installed. Trying to reproduce this, to localize root cause.
Setting priority to Medium

[Roman Podoliaka (rpodolyaka)]

I don't think it's a race condition. Based on how the test case looks like:

https://github.com/Mirantis/mos-integration-tests/blob/master/mos_tests/nova/nova_test.py#L967-L971

it's more likely that qemu-utils failed to install (e.g. due to the fact that `apt-get update` wasn't run and apt cache in the cloud image was outdated and pointed to packages that had already been removed from mirrors).

Anyway, it's not a Nova problem. I suggest you check the instance console log when running this test case and at the very least add `apt-get update` call before trying to install qemu-utils package. Also the echoed marker should ideally be only printed when `apt-get install` finished successfully.

[Ekaterina Shutova (eshutova)]

Move to invalid since problem didn't appear since bug creation.