valet/doc/valet_os.md

valet-openstack

Valet gives OpenStack the ability to optimize cloud resources while simultaneously meeting a cloud application's QoS requirements. Valet provides an api service, an optimizer (Ostro), and a set of OpenStack plugins.

This document covers installation of valet-openstack, a set of OpenStack plugins used to interact with Valet.

IMPORTANT: Overall Installation of Valet is covered in a separate document. These instructions are a component of the overall instructions.

Prerequisites

Prior to installation:

• Ubuntu 14.04 LTS
• Python 2.7.6 with pip
• An OpenStack Kilo cloud
• Access to a valet-api endpoint

Throughout this document, the following installation-specific items are required. Have values for these prepared and ready before continuing. Suggestions for values are provided in this document where applicable.

Name	Description	Example
USER	User id	user1234
$VENV	Python virtual environment path (if any)	/etc/valet/venv
$VALET_OS_PATH	Local git repository's valet_os directory	/home/user1234/git/valet/valet_os
$VALET_HOST	valet-api hostname	localhost
$VALET_USERNAME	OpenStack placement service username	valet
$VALET_PASSWORD	OpenStack placement service password
$VALET_TENANT_NAME	OpenStack placement service default tenant	service
$VALET_FAILURE_MODE	Desired failure mode for Nova ValetFilter	reject
$KEYSTONE_AUTH_API	Keystone Auth API publicurl endpoint	http://controller:5000/
$KEYSTONE_REGION	Keystone Region	RegionOne

Root or sufficient sudo privileges are required for some steps.

A Note About Python Virtual Environments

As valet-openstack works in concert with OpenStack services, if heat and nova have been installed in a python virtual environment (venv), valet-openstack must be installed and configured in the same environment. (A venv helps avoid instabilities and conflicts within the default python environment.)

Installation

Install valet-openstack on the OpenStack controller node containing heat-engine and nova-scheduler. If these services are distributed across multiple nodes, install and configure valet-openstack as appropriate on each node.

valet-openstack is located in valet_os.

$ cd $VALET_OS_PATH
$ sudo pip install .

OpenStack Configuration

valet-openstack requires edits to the heat and nova configuration files, and a stop/start of the heat-engine and nova-scheduler services.

Prerequisites

The following keystone commands must be performed by an OpenStack cloud administrator.

Add a user ($VALET_USERNAME), giving it an admin role in one tenant ($VALET_TENANT_NAME, usually service):

$ keystone user-create --name $VALET_USERNAME --pass $VALET_PASSWORD
$ keystone user-role-add --user $VALET_USERNAME --tenant $VALET_TENANT_NAME --role admin

Create the service entity and API endpoints. While this is not used by Valet 1.0, it is being reserved for future use.

$ keystone service-create --type placement --name valet --description "OpenStack Placement"
$ keystone endpoint-create --region $KEYSTONE_REGION --service valet --publicurl 'http://$VALET_HOST:8090/v1' --adminurl 'http://$VALET_HOST:8090/v1' --internalurl 'http://$VALET_HOST:8090/v1'

Note: In OpenStack parlance, Valet is canonically referred to as a placement service.

The administrator may choose to use differing hostnames/IPs for public vs. admin vs. internal URLs, depending on local architecture and requirements.

Heat

The following changes are made in /etc/heat/heat.conf.

Set plugin_dirs in the [DEFAULT] section such that Heat can locate and use the Valet Stack Lifecycle Plugin.

[DEFAULT]
plugin_dirs = /usr/local/etc/valet_os/heat

Note: If a virtual environment is in use, change the path to be relative to the virtual environment's location, e.g. $VENV/etc/valet_os/heat.

If plugin_dirs is already present, separate entries by commas. The order of entries does not matter. See the OpenStack heat.conf documentation for more information.

Enable stack lifecycle scheduler hints in the [DEFAULT] section:

[DEFAULT]
stack_scheduler_hints = True

Add a [valet] section. This will be used by the Valet Stack Lifecycle Plugin:

[valet]
url = http://$VALET_HOST:8090/v1

Restart heat-engine using separate stop and start directives:

$ sudo service heat-engine stop
$ sudo service heat-engine start

Examine the heat-engine log (usually in /var/log/heat/heat-engine.log). The ATT::Valet plugin should be found and registered:

INFO heat.engine.environment [-]  Registered: [Plugin](User:False) ATT::Valet::GroupAssignment -> <class 'heat.engine.plugins.GroupAssignment.GroupAssignment'>

The heat command line interface (python-heatclient) can also be used to verify plugin registration:

$ heat resource-type-list | grep ATT
| ATT::Valet::GroupAssignment              |

Nova

The following changes are made in /etc/nova/nova.conf.

The nova-scheduler service requires manual configuration so that Nova can locate and use Valet's Scheduler Filter.

Edit the [DEFAULT] section so that scheduler_available_filters and scheduler_default_filters reference Valet, for example:

[DEFAULT]
scheduler_available_filters = nova.scheduler.filters.all_filters
scheduler_available_filters = valet_os.nova.valet_filter.ValetFilter
scheduler_default_filters = RetryFilter, AvailabilityZoneFilter, RamFilter, ComputeFilter, ComputeCapabilitiesFilter, ImagePropertiesFilter, ServerGroupAntiAffinityFilter, ServerGroupAffinityFilter, ValetFilter

When referring to additional filter plugins, multiple scheduler_available_filters lines are required. The first line explicitly makes all of nova's default filters available. The second line makes Valet's filter available. Additional lines may be required for additional plugins.

When setting scheduler_default_filters, ensure that ValetFilter is placed last so that Valet has the final say in scheduling decisions.

Note: scheduler_available_filters denotes filters that are available for use. scheduler_default_filters denotes filters that are enabled by default.

Add a [valet] section. This will be used by the Valet Scheduler Filter:

[valet]
url = http://$VALET_HOST:8090/v1
failure_mode = $VALET_FAILURE_MODE
admin_tenant_name = $VALET_TENANT_NAME
admin_username = $VALET_USERNAME
admin_password = $VALET_PASSWORD
admin_auth_url = $KEYSTONE_AUTH_API

$VALET_FAILURE_MODE refers to the action ValetFilter takes in the effect of a scheduling failure. It can be set to yield (defer to the other filter choices) or reject (block all other filter choices). The default action is reject so as to protect the integrity of Valet exclusivity groups. If exclusivity groups will never be used, or maintaining exclusivity group integrity is not required/practical, it may be desirable to set this to yield.

Restart nova-scheduler using separate stop and start directives:

$ sudo service nova-scheduler stop
$ sudo service nova-scheduler start

Uninstallation

Activate a virtual environment (venv) first if necessary, then uninstall with:

$ sudo pip uninstall valet-openstack

Remove previously made configuration file changes, OpenStack user accounts, and other settings as needed.