UIDs of data-owning users might change between deployed images
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
tripleo |
Expired
|
High
|
Unassigned |
Bug Description
We had a rather confusing situation recently while doing image based updates in Helion:
* Create cloud using image without ceilometer
- time passes
* Create new image _with_ ceilometer
On the controllers, data was stored in /mnt as the UID for _cinder_, let's say 1001.
In the new image, because ceilometer ended up getting created before cinder, it took UID 1001 in the new /etc/passwd. On rebuild/reboot, /mnt/state/
So, we must have a static registry of user names to UID's, and a guard against this sort of problem turning up undetected. We will need to backup /etc/passwd to /mnt/state and verify that the new image has the same mappings for all existing users. If not, we should go into a failure state so that an admin can rollback or deploy a new compatible image.
Changed in tripleo: | |
assignee: | nobody → Clint Byrum (clint-fewbar) |
Image based updates seem less important to us now. It still will be a problem that needs solving, but for now we can side-step it by updating servers in-place.
Also unassigning myself as I won't have time to work on it.