When using OpenStack, the nova-scheduler service is the one picking the specific host where a new VM (or instance) will be deployed.

Yet, sometimes you do not want this to happen. Maybe a specific host in your availability zone and/or cluster is configured in such a way that you would still prefer to have the new instance running there – even at the cost of over-provisioning.

The basics

OpenStack’s nova-scheduler decides internally where a new server instance will be placed based on multiple factors (few of them being the classical amount of RAM, CPU and disk).

All of that also has into account a large amount of filters that can be properly enabled or defined; for instance, to “hint” that a specific host should not be used or on the other hand, to suggest that a set of instances should all be deployed in the same host or to filter by availability zone; among many others.

Thus, if you have three physical hosts and two of them are full, it seems highly likely that a new instance will be deployed in that third host. However, if this host is excluded from the list, the result will vary.

Define specific location during server creation

The easiest way I went for is the one where the command to create the instance also includes the specific host to place this new instance.

Whilst this is documented in the OpenStack’s nova guide, you can find here a quick example to adapt. Note that this is based on the Mitaka version. As you can see from the Nova guide, the basic syntax can follow any of the options below:

1
2
3

openstack server create --availability-zone $ZONE ... $SERVER_ID
openstack server create --availability-zone $ZONE:$HOST ... $SERVER_ID
openstack server create --availability-zone $ZONE::$NODE ... $SERVER_ID

Finding the zone, the host and the node

The following command can be run to easily get the “zone” and the “host”. It is more interesting than the openstack availability zone list command since that one just provides the “zone”.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17

$ openstack host list
+-------------------------------------------+-------------+----------+
| Host Name                                 | Service     | Zone     |
+-------------------------------------------+-------------+----------+
| server2-nova-conductor-container-76cf50fd | conductor   | internal |
| server0-nova-conductor-container-64a4948b | conductor   | internal |
| server1-nova-conductor-container-a302505e | conductor   | internal |
| server2-nova-scheduler-container-8145a5da | scheduler   | internal |
| server0-nova-scheduler-container-e68205ad | scheduler   | internal |
| server1-nova-scheduler-container-993a8796 | scheduler   | internal |
| server1-nova-console-container-d4a3c746   | consoleauth | internal |
| server0-nova-console-container-6647cc1a   | consoleauth | internal |
| server2-nova-console-container-f6a024bf   | consoleauth | internal |
| server1                                   | compute     | nova     |
| server2                                   | compute     | nova     |
| server3                                   | compute     | nova     |
+-------------------------------------------+-------------+----------+

In this example, we have three servers (1, 2, 3) with the “compute” service. Any of these can be chosen for the placement of a new server (or VM instance).

Creating the instance with the specific location

Assuming you want to deploy the instance in “server3” (under the “nova” availability zone), the command would be as follows:

1

openstack server create --flavor=ec1b56f0-4df3-43c2-9663-833a3fad909c --image=aacb59b3-a029-41d5-a79a-35e87db68f22 --security-group=55665ae6-5fe4-4118-a473-4c0457fcfc2c --key-name=test1 --availability-zone=nova:server3 --nic net-id=1fef4ecc-da6b-4db1-9271-89486adf2b91 --nic net-id=00d1e117-2579-4818-938e-b3cbc6dda086 --user-data=/home/test/vm-test1_cloud-init.txt vm-test1

Other approaches

As introduced above, there are different options to force this deployment on the desired host.

Disabling specific hosts

As explained in this thread, it should also be theoretically possible to disable the host(s) you would not wish to deploy to by using the following:

1

nova service-disable $HOST_NAME nova-compute

This approach did not work for me, but again; I have not access to the Nova configuration to investigate why that would be overridden by any other setting.

Migrating from another host

The option of migrating a server instance from a source compute node to another is as well another theoretical option. Again, I tried this but the process seemed stuck and I had no access to the logs for further investigation.

If you want to try it, check the documentation and try this command:

1

openstack server migrate $SERVER_ID --live $HOST_NAME