nova compute fails to update inventory in placement after changing [DEFAULT]/host
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| OpenStack Compute (nova) |
Triaged
|
Undecided
|
Harshavardhan Metla | ||
Bug Description
* build a single node devstack (hostname=aio)
* when everything up and running change the /etc/nova/
* restart the nova-compute service
The below exception is periodically visible in the nova-compute logs
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Nov 22 11:50:52 aio nova-compute[
Problems:
Nova fails to create RP and update the inventory in placement BUT nova compute status is reported UP. I would argue that in this case this nova compute service is not operational as it cannot keep its resource view consistent in placement.
What happens:
* during the first that of the nova-compute service (before the reconfiguration) nova created a service object with hostname=aio and generated a uuid for it. Then used this uuid and the hypervisor_hostname returned by the virt driver (libvirt returns the result of gethostname, which was aio as well in this case) and used this uuid and name to create the compute RP in placement.
* when nova-compute is restarted during after the [DEFAULT]/host is change to other than 'aio' nova creates new service object and generates a new uuid for it. The libvirt driver still returns 'aio' as hypervisor hostname. So nova tries to create a new compute RP with a new uuid but the old 'aio' name. As in placement the RP name needs to be unique this RP creation fails.
| Balazs Gibizer (balazs-gibizer) wrote | #1 |
| tags: | added: compute |
| tags: | added: placement resource-tracker |
| Changed in nova: | |
| status: | New → Triaged |
| Changed in nova: | |
| assignee: | nobody → Harshavardhan Metla (harsha24) |
| Balazs Gibizer (balazs-gibizer) wrote | #2 |
I guess at least we need to document in the [DEFAULT]/host config option that changing the value of it after initial deploy needs special care (like cleaning up the old service object and placement)