yahoo-eng-team team mailing list archive

[Thomas Goirand <2125567@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

We did a (human) mistake doing the setup of a compute node, not
activating Ceph on it, while all the other compute do have a Ceph
backend.

Then we cold migrated a VM from a compute with Ceph backend, to that new
compute that had the VM system disk under
/var/lib/nova/instances/<UUID>/disk. Instead of refusing the cold
migration, Nova happily accepted it. As a consequence, our customer got
a fresh new disk from Glance (the one that was configured for the VM to
begin with), and booted that.

If this is a feature, that's not a funny one. The customer had down time
as his VM lost its system disk setup, and we thought the data of the VM
was lost. In fact, the data were in Ceph, but not used. We did a dump
from Ceph like this:

rbd -p nova export UUID_disk /srv/data/UUID_disk.raw

and imported that raw file in Glance, so the customer could destroy the
VM and re-create it with this image from Glance.

So, to avoid surprise, Nova should either:
- convert the VM from Ceph and create the disk file on the destination compute (ideal scenario)
- refuse the cold migration

If we want to keep things as they are right now, then at least we should
put an option in the scheduler (or as a policy in the API?) so this type
of cold migration that may loose the system disk can be forbidden.

** Affects: nova
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/2125567

Title:
  Nova allows migration between nodes with different backend (ceph or
  not)

Status in OpenStack Compute (nova):
  New

Bug description:
  We did a (human) mistake doing the setup of a compute node, not
  activating Ceph on it, while all the other compute do have a Ceph
  backend.

  Then we cold migrated a VM from a compute with Ceph backend, to that
  new compute that had the VM system disk under
  /var/lib/nova/instances/<UUID>/disk. Instead of refusing the cold
  migration, Nova happily accepted it. As a consequence, our customer
  got a fresh new disk from Glance (the one that was configured for the
  VM to begin with), and booted that.

  If this is a feature, that's not a funny one. The customer had down
  time as his VM lost its system disk setup, and we thought the data of
  the VM was lost. In fact, the data were in Ceph, but not used. We did
  a dump from Ceph like this:

  rbd -p nova export UUID_disk /srv/data/UUID_disk.raw

  and imported that raw file in Glance, so the customer could destroy
  the VM and re-create it with this image from Glance.

  So, to avoid surprise, Nova should either:
  - convert the VM from Ceph and create the disk file on the destination compute (ideal scenario)
  - refuse the cold migration

  If we want to keep things as they are right now, then at least we
  should put an option in the scheduler (or as a policy in the API?) so
  this type of cold migration that may loose the system disk can be
  forbidden.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/2125567/+subscriptions