Delete Duplicate OpenStack Hypervisors and Services

Sunday, August 3, 2014

If you ever change the hostname of any of your OpenStack nodes and restart the OpenStack services on those nodes, the services are going to re-register to the OpenStack cluster under the new hostname. Because of this, when you run nova hypervisor-list, nova service-list, neutron agent-list, or cinder service-list you are going to have duplicate entries.

Unfortunately, there are not commands to clean up duplicate entries, so you have to modify the various OpenStack databases by hand.

It would be wise before making any changes to backup all of the OpenStack databases. This can be done with the following command (more information on the mysqldump command can be found here):

mysqldump --all-databases > /root/openstack-databases.sql

As mentioned by Dave Johnston in the comments below, OpenStack Juno has begun introducing commands to do what is described in this post so you do not have to manually modify the database.

Delete Duplicate Nova Hypervisors

Identity that you have duplicate hypervisor entries by running nova hypervisor-list:

root@controller1:~# nova hypervisor-list
+----+---------------------+
| ID | Hypervisor hostname |
+----+---------------------+
| 1  | compute1            |
| 3  | compute1.local      |
+----+---------------------+

Log into your OpenStack controller node as root and login to the MySQL command line as the root user:

mysql -u root

Select the nova database:

USE nova;

Run the following SQL command to identify what hypervisors have similar hostnames:

mysql> SELECT id, created_at, updated_at, hypervisor_hostname FROM compute_nodes;
+----+---------------------+---------------------+---------------------+
| id | created_at          | updated_at          | hypervisor_hostname |
+----+---------------------+---------------------+---------------------+
|  1 | 2014-08-03 19:37:12 | 2014-08-03 19:47:33 | compute1            |
|  3 | 2014-08-03 19:47:35 | 2014-08-03 20:02:35 | compute1.local      |
+----+---------------------+---------------------+---------------------+
2 rows in set (0.00 sec)

In this scenario, compute1 was the original hostname and compute1.local is the new hostname. compute1 is the entry you will want to delete.

First, make sure the new hostname’s nova-compute service is registering as up. If you run the nova service-list command, nova-compute on compute1 should be down and nova-compute on compute1.local should be up.

root@controller1:~# nova service-list
+------------------+----------------+----------+---------+-------+----------------------------+-----------------+
| Binary           | Host           | Zone     | Status  | State | Updated_at                 | Disabled Reason |
+------------------+----------------+----------+---------+-------+----------------------------+-----------------+
| nova-scheduler   | controller1    | internal | enabled | up    | 2014-08-03T19:49:16.000000 | None            |
| nova-conductor   | controller1    | internal | enabled | up    | 2014-08-03T19:49:16.000000 | None            |
| nova-cert        | controller1    | internal | enabled | up    | 2014-08-03T19:49:09.000000 | None            |
| nova-consoleauth | controller1    | internal | enabled | up    | 2014-08-03T19:49:11.000000 | None            |
| nova-compute     | compute1       | nova     | enabled | down  | 2014-08-03T19:47:34.000000 | None            |
| nova-compute     | compute1.local | nova     | enabled | up    | 2014-08-03T19:49:15.000000 | None            |
+------------------+----------------+----------+---------+-------+----------------------------+-----------------+

If this is the case, and the updated_at time for compute1 in the SQL output above is older than compute1.local, you can safely delete the compute1 entry from the various tables in the nova database.

First, get the id of compute1. In this scenario it is simply 1.

Delete the duplicate hypervisor from the nova database by running the following SQL commands:

DELETE FROM compute_node_stats WHERE compute_node_id='1';

DELETE FROM compute_nodes WHERE hypervisor_hostname='compute1';

In addition, and as seen above, there are going to be duplicate entries in the Nova services table. The duplicate entry can be deleted by running the following SQL command:

DELETE FROM services WHERE host='compute1';

nova hypervisor-list should now have no duplicate entries:

root@controller1:~# nova hypervisor-list
+----+---------------------+
| ID | Hypervisor hostname |
+----+---------------------+
| 3  | compute1.local      |
+----+---------------------+

nova service-list should also now have no duplicate entries:

root@controller1:~# nova service-list
+------------------+----------------+----------+---------+-------+----------------------------+-----------------+
| Binary           | Host           | Zone     | Status  | State | Updated_at                 | Disabled Reason |
+------------------+----------------+----------+---------+-------+----------------------------+-----------------+
| nova-scheduler   | controller1    | internal | enabled | up    | 2014-08-03T20:29:37.000000 | None            |
| nova-conductor   | controller1    | internal | enabled | up    | 2014-08-03T20:29:36.000000 | None            |
| nova-cert        | controller1    | internal | enabled | up    | 2014-08-03T20:29:31.000000 | None            |
| nova-consoleauth | controller1    | internal | enabled | up    | 2014-08-03T20:29:33.000000 | None            |
| nova-compute     | compute1.local | nova     | enabled | up    | 2014-08-03T20:29:36.000000 | None            |
+------------------+----------------+----------+---------+-------+----------------------------+-----------------+

Delete Duplicate Neutron Services

Even though you have deleted the duplicate Nova hypervisors and services, OpenStack compute nodes typically are running some Neutron services as well (in most cases it is just the Open vSwitch agent).

If you run neutron agent-list, you will most likely have duplicate entries for the Open vSwitch agent:

root@controller1:~# neutron agent-list
+--------------------------------------+--------------------+----------------+-------+----------------+
| id                                   | agent_type         | host           | alive | admin_state_up |
+--------------------------------------+--------------------+----------------+-------+----------------+
| 5a0296d1-18db-4b49-948c-15577b1116c1 | L3 agent           | controller1    | :-)   | True           |
| 70b694e3-e27a-418e-aeb7-ce2c0a521717 | Open vSwitch agent | compute1.local | :-)   | True           |
| a15b145d-ef9a-4fae-8368-072a938e329d | DHCP agent         | controller1    | :-)   | True           |
| da8a56c9-a117-45c0-b4c7-e46f66eeec01 | Open vSwitch agent | compute1       | xxx   | True           |
| edc42804-2f52-4cc7-a29d-1f09eb911643 | Open vSwitch agent | controller1    | :-)   | True           |
+--------------------------------------+--------------------+----------------+-------+----------------+

The Open vSwitch agent service for compute1.local is up, so you can safely delete the duplicate entry for compute1 from the Neutron database.

Again, log into your OpenStack controller node as root and login to the MySQL command line as the root user:

mysql -u root

Select the neutron database:

USE neutron;

Delete the duplicate Neutron service from the agents table by running the following SQL command:

DELETE FROM agents WHERE host='compute1';

neutron agent-list should now have no duplicate entries:

root@controller1:~# neutron agent-list
+--------------------------------------+--------------------+----------------+-------+----------------+
| id                                   | agent_type         | host           | alive | admin_state_up |
+--------------------------------------+--------------------+----------------+-------+----------------+
| 5a0296d1-18db-4b49-948c-15577b1116c1 | L3 agent           | controller1    | :-)   | True           |
| 70b694e3-e27a-418e-aeb7-ce2c0a521717 | Open vSwitch agent | compute1.local | :-)   | True           |
| a15b145d-ef9a-4fae-8368-072a938e329d | DHCP agent         | controller1    | :-)   | True           |
| edc42804-2f52-4cc7-a29d-1f09eb911643 | Open vSwitch agent | controller1    | :-)   | True           |
+--------------------------------------+--------------------+----------------+-------+----------------+

Delete Duplicate Cinder Services

If you happen to have a Cinder node and also changed its hostname, there will be duplicate entries in the Cinder services table:

root@controller1:~# cinder service-list
+------------------+---------------+------+---------+-------+----------------------------+
|      Binary      |      Host     | Zone |  Status | State |         Updated_at         |
+------------------+---------------+------+---------+-------+----------------------------+
| cinder-scheduler |  controller1  | nova | enabled |   up  | 2014-08-03T19:49:04.000000 |
|  cinder-volume   |    cinder1    | nova | enabled |  down | 2014-08-03T19:48:53.000000 |
|  cinder-volume   | cinder1.local | nova | enabled |   up  | 2014-08-03T19:49:08.000000 |
+------------------+---------------+------+---------+-------+----------------------------+

The cinder-volume service for cinder1.local is up, so you can safely delete the duplicate entry for cinder1 from the Cinder database.

Once again, log into your OpenStack controller node as root and login to the MySQL command line as the root user:

mysql -u root

Select the cinder database:

USE cinder;

Delete the duplicate Cinder service from the services table by running the following SQL command:

DELETE FROM services WHERE host='cinder1';

cinder service-list should now have no duplicate entries:

root@controller1:~# cinder service-list
+------------------+---------------+------+---------+-------+----------------------------+
|      Binary      |      Host     | Zone |  Status | State |         Updated_at         |
+------------------+---------------+------+---------+-------+----------------------------+
| cinder-scheduler |  controller1  | nova | enabled |   up  | 2014-08-03T20:31:47.000000 |
|  cinder-volume   | cinder1.local | nova | enabled |   up  | 2014-08-03T20:31:51.000000 |
+------------------+---------------+------+---------+-------+----------------------------+

References

How to remove compute node from havana

Manage services



comments powered by Disqus