Stein->Train Upgrade : Old nova services not cleaned

Bug #1863107 reported by Keith Plant
20
This bug affects 4 people
Affects Status Importance Assigned to Milestone
OpenStack Compute (nova)
Invalid
Undecided
Unassigned
kolla-ansible
In Progress
Medium
Unassigned
Train
Triaged
Medium
Unassigned
Ussuri
Triaged
Medium
Unassigned

Bug Description

**This appears to be entirely superficial, I haven't been any to find any functional fallout from this behavior**

When upgrading from stable/stein to stable/train, the old nova-consoleauth and nova-scheduler services are not cleaned up:

Before upgrade: http://paste.openstack.org/show/82jvGkP67YZkzUl3dCJS/
After upgrade: http://paste.openstack.org/show/fNwE00IDdvzX4On8ZWyc/

**Environment**:
OS: CentOS 7.7.1908
Kernel: 3.10.0-1062.9.1.el7.x86_64
Docker version: 18.09.7, build 2d0083d
kolla-ansible version: stable/train (9.0.1-8-g4358395)
docker image type: source
docker image dist: centos7

Revision history for this message
Mark Goddard (mgoddard) wrote :

The nova-consoleauth container was removed in Train, so we should delete those services. I don't know why you have new scheduler services though. Did something change in the configuration? Perhaps something to do with the cells refactor?

Changed in kolla-ansible:
importance: Undecided → Medium
status: New → Triaged
Revision history for this message
Keith Plant (kplant) wrote :

Not to my knowledge, I may have missed some necessary configuration steps. I merged the stable/train globals.yml with the stable/stein globals.yml by doing a stare-and-compare with vimdiff.

Revision history for this message
Radosław Piliszek (yoctozepto) wrote :

I confirm this happens despite the same config. Wonder what happens with those schedulers.

Revision history for this message
xinliang (xin3liang) wrote :

I also met the same issue, when do a stein->train upgrade.

Changed in nova:
status: New → Invalid
Changed in kolla-ansible:
milestone: 10.0.0 → none
Revision history for this message
Oliver Horecny (horecoli) wrote :

I hit same issue during rolling upgrade Rocky->Stein->Train.

Since this change: https://github.com/openstack/kolla-ansible/commit/78a828ef42160d161f944308ec51af0303778ca7
nova-scheduler was moved to different DB than it was before. And after upgrade it is duplicated in openstack:

openstack compute service list --service nova-scheduler

+----+----------------+----------------------------+----------+---------+-------+----------------------------+
| ID | Binary | Host | Zone | Status | State | Updated At |
+----+----------------+----------------------------+----------+---------+-------+----------------------------+
| 1 | nova-scheduler | test-upgrade | internal | enabled | up | 2020-12-01T14:32:09.000000 |
| 4 | nova-scheduler | test-upgrade | internal | enabled | down | 2020-10-26T16:58:35.000000 |
+----+----------------+----------------------------+----------+---------+-------+----------------------------+

It is also visible in DB, where nova-scheduler is stored on two places:

MariaDB [(none)]> select created_at,updated_at,deleted_at,id,host,`binary` from nova.services where `binary`='nova-scheduler';
+---------------------+---------------------+------------+----+----------------------------+----------------+
| created_at | updated_at | deleted_at | id | host | binary |
+---------------------+---------------------+------------+----+----------------------------+----------------+
| 2020-09-29 13:19:35 | 2020-10-26 16:58:35 | NULL | 4 | test-upgrade | nova-scheduler |
+---------------------+---------------------+------------+----+----------------------------+----------------+

MariaDB [(none)]> select created_at,updated_at,deleted_at,id,host,`binary` from nova_cell0.services where `binary`='nova-scheduler';
+---------------------+---------------------+------------+----+----------------------------+----------------+
| created_at | updated_at | deleted_at | id | host | binary |
+---------------------+---------------------+------------+----+----------------------------+----------------+
| 2020-10-26 16:55:20 | 2020-12-01 13:35:39 | NULL | 1 | test-upgrade | nova-scheduler |
+---------------------+---------------------+------------+----+----------------------------+----------------+

After upgrade it seems, that openstack is fully functional, but that nova-scheduler which is in state "down" shouldn't be present in openstack. I think kolla-ansible should handle this situation by removing that old unused service from DB.
There is also additional unused service nova-consoleauth which was used in previous releases and need to be removed from DB after upgrade. So it can be done in same step.

Changed in kolla-ansible:
status: Triaged → In Progress
Revision history for this message
Radosław Piliszek (yoctozepto) wrote :
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.