remove-services action deletes useful services

Bug #2054398 reported by Shunde Zhang

This bug report will be marked for expiration in 50 days if no further activity occurs. (find out why)

8
This bug affects 1 person
Affects Status Importance Assigned to Milestone
OpenStack Cinder Charm
Incomplete
Undecided
Unassigned

Bug Description

The remove-services action deletes a cinder service if its host is not 'cinder' or 'cinder@cinder-ceph'.
Some users use other drivers such as purestorage, netapp etc.

After initial deployment, there is only one backend service in cinder service list.
If that is deleted by the remove-services action, several entries will appear in the list if there are multiple volume services running, which is normal for HA in a production environment. These entries will have the same binary and host, and some may appear to be down, which may trigger alerting system to send out alerts.

This behavior is not right. The remove-services action shouldn't delete useful services.

Tags: sts
tags: added: sts
Revision history for this message
Alex Kavanagh (ajkavanagh) wrote :

Hi @shunde-zhang

Could you provide a more concrete example please? e.g. what was configured/installed before the remove-services action, what command you ran, and then what the resultant configuration/situation was afterwards. Please add juju-status, and relevant debug logs of the cinder service(s).

Thanks.

Changed in charm-cinder:
status: New → Incomplete
Revision history for this message
Shunde Zhang (shunde-zhang) wrote :

Hi @ajkavanagh

According to the code [1], it will delete all cinder services if the host is not 'cinder' or 'cinder@cinder-ceph'.
But I've seen 2 exceptions from user's environment.

1. In some deployments, cinder-scheduler service's host is hostname, e.g. juju-xxx-lxd-xxx, and not 'cinder'. In this case the cinder-scheduler service will be removed. Although it can be created again, this is not necessary.
| cinder-scheduler | juju-fa8549-4-lxd-1 | nova | enabled | up | 2024-02-01T07:10:58.000000 |
| cinder-scheduler | juju-fa8549-8-lxd-1 | nova | enabled | up | 2024-02-01T07:10:59.000000 |
| cinder-scheduler | juju-fa8549-6-lxd-1 | nova | enabled | up | 2024-02-01T07:10:59.000000 |

2. Some users use other storage backends, e.g. purestorage, and not ceph. Those services will also be removed.
| cinder-volume | cinder-volume@cinder-purestorage | nova | enabled | up | 2024-02-01T07:11:02.000000 |
When this is removed, if the user has multiple cinder units, several cinder-volume services will be created for purestorage. This is not right, since only one is required.

I think this action shouldn't remove a cinder service if it is up. The check on 'host' is not quite right since it doesn't consider those who use a backend which is not ceph.

Thanks!

[1] https://opendev.org/openstack/charm-cinder/src/branch/master/actions/cinder_manage.py#L123

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.