Bug 1902153

I turned out that filestore-to-bluestore.yaml skipped the steps for migration

/var/lib/mistral/4eacc9bf-622b-43cc-9301-c8c1f6e328b6/ceph-ansible/ceph_ansible_command.log
~~~
Running /var/lib/mistral/4eacc9bf-622b-43cc-9301-c8c1f6e328b6/ceph-ansible/ceph_ansible_command.sh
...
2020-11-27 11:42:08,293 p=143980 u=root n=ansible | ok: [ceph-0] => {"ansible_facts": {"current_objectstore": "bluestore"}, "changed": false}
2020-11-27 11:42:08,343 p=143980 u=root n=ansible | TASK [warn user about osd already using bluestore] *****************************
2020-11-27 11:42:08,343 p=143980 u=root n=ansible | Friday 27 November 2020  11:42:08 +0900 (0:00:00.075)       0:00:05.484 ******* 
2020-11-27 11:42:08,368 p=143980 u=root n=ansible | ok: [ceph-0] => {
    "msg": "WARNING: ceph-0 is already using bluestore. Skipping all tasks."
}
...
~~~

This is because the playbook has a logic to skip migration when osd_objectstore is "bluestore"

https://github.com/ceph/ceph-ansible/tree/stable-4.0/infrastructure-playbooks/filestore-to-bluestore.yml
~~~
- hosts: "{{ osd_group_name }}"
  become: true
  serial: 1
  vars:
    delegate_facts_host: true
  tasks:
    - name: gather and delegate facts
      setup:
      delegate_to: "{{ item }}"
      delegate_facts: True
      with_items: "{{ groups[mon_group_name] }}"
      run_once: true
      when: delegate_facts_host | bool

    - import_role:
        name: ceph-defaults

    - name: set_fact current_objectstore
      set_fact:
        current_objectstore: '{{ osd_objectstore }}'

    - name: warn user about osd already using bluestore
      debug:
        msg: 'WARNING: {{ inventory_hostname }} is already using bluestore. Skipping all tasks.'
      when: current_objectstore == 'bluestore'
~~~

Our current documentation says that we should set "osd_objectstore: bluestore" in
CephAnsibleDiskConfig and this is causing the issue.

Even if we remove that line bluestore seems to be the default value in ceph-ansible
so I'm afraid the issue is not solved.

https://access.redhat.com/documentation/en-us/red_hat_openstack_platform/16.1/html/framework_for_upgrades_13_to_16.1/osd-migration-from-filestore-to-bluestore#migrating-OSDs-from-FileStore-to-BlueStore

I don't think this is a documentation BZ. This sounds like an engineering BZ.

(In reply to Dan Macpherson from comment #2)
> I don't think this is a documentation BZ. This sounds like an engineering BZ.

One possible solution without code change would be to set
 osd_objectstore: filestore
before the migration command and reset the parameter to bluestore after migration completes.
We can use "openstack overcloud deploy --stack-only" to change the parameter without
triggering actual deployment steps.

However I tend to agree with you about this is an engineering BZ and the above parameter settings
should be handled in tripleo, ideally.

Do you want me to change the assigned component to tripleo-heat-templates(or any different package
if we have better one), or can we get some insights from Ceph squad before moving this bz ?

Thanks Takashi

Hi @Francesco. Could you take a look and advise how best to address this BZ please? I agree with Dan that it seems more like an engineering BZ.

Thanks :)

Thanks for the updated note and clear explanation Takashi.
Here is the update:

https://access.redhat.com/documentation/en-us/red_hat_openstack_platform/16.1/html-single/framework_for_upgrades_13_to_16.1/index?lb_target=production#OSD-migration-from-filestore-to-bluestore

Thanks Naomi,

The updated version looks good to me !