This chapter describes how to create and access instances.
Using an image, create a new instance via the Dashboard options.
Log into the Dashboard, and select the Compute project from the drop down list.
Click the Images option.
Locate the image that will act as the instance base from the Images table.
Click Launch from the Actions column.
Check the Launch Instances dialog, and find the details tab. Enter the appropriate values for the instance.
In the Launch Instance dialog, click the Access & Security tab. Select the keypair. Set the security group as “default”.
Click the Networking tab. This tab will be unavailable if OpenStack networking (neutron) has not been enabled. If networking is enabled, select the networks on which the instance will reside.
Click the Volume Options tab. This tab will only be available if a Block Storage volume exists for the instance. Select Don’t boot from a volume for now.
For more information on attaching Block Storage volumes to instances for persistent storage, see the Managing volumes for persistent storage section below.
Add customisation scripts, if needed, by clicking the Post-Creation tab. These run after the instance has been created. Some instances support user data, such as root passwords, or admin users. Enter the information specific to the instance here if required.
Click Advanced Options. Specify whether the instance uses a configuration drive to store metadata by selecting a disk partition type.
Click Launch to create the instance. The instance will start on a compute node. The Instance page will open and start creating a new instance. The Instance page that opens will list the instance name, size, status, and task. Power state and public and private IP addresses are also listed here.
The process will take less than a minute to complete. Instance creation is complete when the status is listed as active. Refresh the page to see the new active instance.
Field Name | Required | Details |
---|---|---|
Availability Zone | Optional | The availability zone in which the image service creates the instance. If no availability zones is defined, no instances will be found. The cloud provider sets the availability zone to a specific value. |
Instance Name | Required | The name of the new instance, which becomes the initial host name of the server. If the server name is changed in the API or directly changed, the Dashboard names remain unchanged |
Image | Required | The type of container format, one of ami , ari , aki ,
bare , or ovf |
Flavor | Required | The vCPU, Memory, and Disk configuration. Note that larger flavors can
take a long time to create. If creating an instance for the first time
and want something small with which to test, select m1.small . |
Instance Count | Required | If creating multiple instances with this configuration, enter an integer
up to the number permitted by the quota, which is 10 by default. |
Instance Boot Source | Required | Specify whether the instance will be based on an image or a snapshot. If it is the first time creating an instance, there will not yet be any snapshots available. |
Image Name | Required | The instance will boot from the selected image. This option will be
pre-populated with the instance selected from the table. However, choose
Boot from Snapshot in Instance Boot Source, and it will default
to Snapshot instead. |
Security Groups | Optional | This option assigns security groups to an instance. The default security group activates when no customised group is specified here. Security Groups, similar to a cloud firewall, define which incoming network traffic is forwarded to instances. |
Keypair | Optional | Specify a key pair with this option. If the image uses a static key set (not recommended), a key pair is not needed. |
Selected Networks | Optional | To add a network to an instance, click the + in the Networks field. |
Customisation Script | Optional | Specify a customisation script. This script runs after the instance launches and becomes active. |
On the command line, instance creation is managed with the openstack server create command. Before launching an instance, determine what images and flavors are available to create a new instance using the openstack image list and openstack flavor list commands.
Log in to any utility container.
Issue the openstack server create command with a name for the instance, along with the name of the image and flavor to use:
$ openstack server create --image precise-image --flavor=2 --key-name example-key example-instance
+-------------------------------------+--------------------------------------+
| Property | Value |
+-------------------------------------+--------------------------------------+
| OS-DCF:diskConfig | MANUAL |
| OS-EXT-SRV-ATTR:host | None |
| OS-EXT-SRV-ATTR:hypervisor_hostname | None |
| OS-EXT-SRV-ATTR:instance_name | instance-0000000d |
| OS-EXT-STS:power_state | 0 |
| OS-EXT-STS:task_state | scheduling |
| OS-EXT-STS:vm_state | building |
| accessIPv4 | |
| accessIPv6 | |
| adminPass | ATSEfRY9fZPx |
| config_drive | |
| created | 2012-08-02T15:43:46Z |
| flavor | m1.small |
| hostId | |
| id | 5bf46a3b-084c-4ce1-b06f-e460e875075b |
| image | precise-image |
| key_name | example-key |
| metadata | {} |
| name | example-instance |
| progress | 0 |
| status | BUILD |
| tenant_id | b4769145977045e2a9279c842b09be6a |
| updated | 2012-08-02T15:43:46Z |
| user_id | 5f2f2c28bdc844f9845251290b524e80 |
+-------------------------------------+--------------------------------------+
To check that the instance was created successfully, issue the openstack server list command:
$ openstack server list
+------------------+------------------+--------+-------------------+---------------+
| ID | Name | Status | Networks | Image Name |
+------------------+------------------+--------+-------------------+---------------+
| [ID truncated] | example-instance | ACTIVE | public=192.0.2.0 | precise-image |
+------------------+------------------+--------+-------------------+---------------+
The Actions column includes the following options:
Note
Terminate the instance under the Actions column.
Volumes attach to instances, enabling persistent storage. Volume storage provides a source of memory for instances. Administrators can attach volumes to a running instance, or move a volume from one instance to another.
Nova is capable of live migration instances from one host to a different host to support various operational tasks including:
Depending on the OpenStack-Ansible version in use, Nova can
be configured to force configuration drive attachments to instances.
In this case, a ISO9660 CD-ROM image will be made available to the
instance via the /mnt
mount point. This can be used by tools,
such as cloud-init, to gain access to instance metadata. This is
an alternative way of accessing the Nova EC2-style Metadata.
To allow live migration of Nova instances, this forced provisioning of the config (CD-ROM) drive needs either be turned off, or the format of the configuration drive needs to be changed to a disk format like vfat, a format which both Linux and Windows instances can access.
This work around is required for all Libvirt versions prior 1.2.17.
To turn off the forced provisioning of the config drive, add the following
override to the /etc/openstack_deploy/user_variables.yml
file:
nova_force_config_drive: False
To change the format of the configuration drive, to a hard disk style format,
use the following configuration inside the same
/etc/openstack_deploy/user_variables.yml
file:
nova_nova_conf_overrides:
DEFAULT:
config_drive_format: vfat
force_config_drive: false
In the default configuration, Nova determines the correct transport
URL for how to transfer the data from one host to the other.
Depending on the nova_virt_type
override the following configurations
are used:
qemu+tcp://%s/system
qemu+tcp://%s/system
xenmigr://%s/system
Libvirt TCP port to transfer the data to migrate.
OpenStack-Ansible changes the default setting and used a encrypted SSH connection to transfer the instance data.
live_migration_uri = "qemu+ssh://nova@%s/system?no_verify=1&keyfile={{ nova_system_home_folder }}/.ssh/id_rsa"
Other configurations can be configured inside the
/etc/openstack_deploy/user_variables.yml
file:
nova_nova_conf_overrides:
libvirt:
live_migration_completion_timeout: 0
live_migration_progress_timeout: 0
live_migration_uri: "qemu+ssh://nova@%s/system?keyfile=/var/lib/nova/.ssh/id_rsa&no_verify=1"
The live migration is accessible via the nova client.
nova live-migration [--block-migrate] [--force] <uuid> [<host>]
Examplarery live migration on a local storage:
nova live-migration --block-migrate <uuid of the instance> <nova host>
Once the live migration request has been accepted, the status can be monitored with the nova client:
nova migration-list
+-----+------------+-----------+----------------+--------------+-----------+-----------+---------------+------------+------------+------------+------------+-----------------+
| Id | Source Node | Dest Node | Source Compute | Dest Compute | Dest Host | Status | Instance UUID | Old Flavor | New Flavor | Created At | Updated At | Type |
+----+-------------+-----------+----------------+--------------+-----------+-----------+---------------+------------+------------+------------+------------+-----------------+
| 6 | - | - | compute01 | compute02 | - | preparing | f95ee17a-d09c | 7 | 7 | date | date | live-migration |
+----+-------------+-----------+----------------+--------------+-----------+-----------+---------------+------------+------------+------------+------------+-----------------+
To filter the list, the options --host
or --status
can be used:
nova migration-list --status error
In cases where the live migration fails, both the source and destination compute nodes need to be checked for errors. Usually it is sufficient to search for the instance UUID only to find errors related to the live migration.
Besides the live migration, Nova offers the option to migrate entire hosts in a online (live) or offline (cold) migration.
The following nova client commands are provided:
host-evacuate-live
Live migrate all instances of the specified host to other hosts if resource utilzation allows. It is best to use shared storage like Ceph or NFS for host evacuation.
host-servers-migrate
This command is similar to host evacuation but migrates all instances off the specified host while they are shutdown.
resize
Changes the flavor of an Nova instance (increase) while rebooting and also migrates (cold) the instance to a new host to accommodate the new resource requirements. This operation can take considerate amount of time, depending disk image sizes.
Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.