nova-compute accepts non-existing path for ephemeral-device

[Przemyslaw Hausman]

bionic-stein, 20.02 charms release.

nova-compute accepts non-existing path for ephemeral-device, for example '/dev/disk/by-dname/i-dont-exist'. Charm successfully deploys. VMs can be successfully launched.

However, ephemeral disks for VMs end up in /var/lib/nova/instances, instead of on the device '/dev/disk/by-dname/i-dont-exist' (as expected). Nova compute charm is silently ignoring incorrect setting for ephemeral-device. This may sound like a sane fallback.

But in case of e.g. typo in configuration of ephemeral-device, this results is an issue difficult to spot, because VMs run just fine and the charm does not report any errors.

I would suggest to set an error status on the charm with the message indicating that the config value for ephemeral-device does not match current configuration on the host. Or at least a warning in the charm message in juju status output. So that it is easy to spot and correct early.

[Chris MacNaughton (chris.macnaughton)]

The charm should set the status with a warning if the ephemeral device is not available

[Vladimir Grevtsev (vlgrevtsev)]

> The charm should set the status with a warning if the ephemeral device is not available

Indeed. We just hit this during the BS handover - one node was having incorrectly configured storage (sda/sdb devices were swapped, while charm had ephemeral-device config option set to /dev/sdb).

Could we please revisit this issue again to avoid similar problems in the future?

[Andre Ruiz (andre-ruiz)]

Just my 0,02: if we are asking for a specific path and it cannot be mounted, it should go into error state and require manual intervention. Even a warning in the logs may be not enough (unless the warning stays on juju status to be easily spotted.

[Vladimir Grevtsev (vlgrevtsev)]

For the record, we have just hit this again in another deployment.

[Alex Kavanagh (ajkavanagh)]

Hi Pedro. Are you still working on this? If not, please could you un-assign yourself. Many thanks, Alex.

[OpenStack Infra (hudson-openstack)]

Change abandoned by "Alex Kavanagh <email address hidden>" on branch: master
Review: https://review.opendev.org/c/openstack/charm-nova-compute/+/755780
Reason: This review is stale as it has not had an updates since the review comments.

[Kamal Bhaskar (kamalb)]

Hello

We are facing the same issue in a customer environment (jammy/yoga):

nova-compute-control 25.2.0 active 6 nova-compute yoga/stable 686 no Unit is ready

$ juju config nova-compute-control ephemeral-device
/dev/disk/by-dname/vg0-lvephemeral
$

And the actual/correct path is:

# ls -lrt /dev/disk/by-dname/vg0-lvephemeral
ls: cannot access '/dev/disk/by-dname/vg0-lvephemeral': No such file or directory
# ls -lrt /dev/disk/by-dname/vg1-lvephemeral
lrwxrwxrwx 1 root root 10 Dec 11 06:36 /dev/disk/by-dname/vg1-lvephemeral -> ../../dm-4
#

But, the charm deployed successfully and is in active/idle and ready state; while in the backend it failed to mount the above ephemeral-device path and fallen back to use root(/) partition with default images stored in /var/lib/nova/instances.

Please let me know if any other details are needed in this regard.

Thanks & regards