Ceph OSDs require manual restarts after Octopuc -> Pacific upgrade after Victoria -> Wallaby openstack upgrade
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Ceph OSD Charm |
New
|
Undecided
|
Unassigned |
Bug Description
Hi,
Environment is:
* Focal-Victoria cloud during upgrade to Focal-Wallaby
* Juju 2.9.45
* All charms were refreshed to their victoria/latest on the 29th of May.
* All control plane and all Nova were Wallaby already.
* Somewhat outdated(because I took it before Nova upgrades) juju status --relations output: https:/
* Somewhat outdated(because I took it before Nova upgrades) sanitized bundle is here: https:/
How I reproduced it once:
* All OpenStack services were on Wallaby, all Ceph services except ceph-osd were on Pacific. Ceph osd were still Octopus
* Upgraded ceph-osd.
```
juju refresh ceph-osd --channel=
# wait until settles
juju config ceph-osd source=
# wait until settles
```
After that I saw that all my osds are still running octopus
```
$ sudo ceph tell osd.* version
... skipping ...
"version": "15.2.17",
"release": "octopus",
... skipping ...
```
I restarted all `systemctl restart ceph-osd.target` manually and all versions changed to
```
"version": "16.2.15",
"release": "pacific",
```
After that ceph complained about `require_
```
[WRN] OSD_UPGRADE_
all OSDs are running pacific or later but require_osd_release < pacific
```
I had manually switch it to pacific
```
$ sudo ceph osd require-osd-release pacific
```
After that ceph cluster became healthy. Seems like charm haven't done much except for package refresh.
The juju log for one of ceph-osd units during the upgrade: https:/
What I think important is that 6 out of 9 Nova hosts were affected by https:/