fuel-library/deployment/puppet/openstack

The Openstack modules:

Introduction

The Openstack Puppet Modules are a flexible Puppet implementation capable of configuring the core Openstack services:

• nova (compute service)
• glance (image database)
• swift (object store)
• keystone (authentication/authorization)
• horizon (web front end)

A 'Puppet Module' is a collection of related content that can be used to model the configuration of a discrete service.

These modules are based on the adminstrative guides for openstack compute and object store

Dependencies:

Puppet:

• Puppet 2.7.12 or greater
• Facter 1.6.1 or greater (versions that support the osfamily fact)

Platforms:

These modules have been fully tested on Ubuntu Precise and Debian Wheezy.

For instructions of how to use these modules on Debian, check out this excellent link:

The instuctions in this document have only been verified on Ubuntu Precise.

Network:

Each of the machines running the Openstack services should have a minimum of 2 NICS.

• One for the public/internal network
  • This nic should be assigned an IP address
• One of the virtual machine network
  • This nic should not have an ipaddress assigned

If machines only have one NIC, it is necessary to manually create a bridge called br100 that bridges into the ip address specified on that NIC

All interfaces that are used to bridge traffic for the internal network need to have promiscuous mode set.

Below is an example of setting promiscuous mode on an interface on Ubuntu.

    #/etc/network/interfaces   
    auto eth1
    iface eth1 inet manual
      up ifconfig $IFACE 0.0.0.0 up
      up ifconfig $IFACE promisc

Volumes:

Every node that is configured to be a nova volume service must have a volume group called nova-volumes.

Compute nodes

Compute nodes should be deployed onto physical hardware.

If compute nodes are deployed on virtual machines for testing, the libvirt_type must be configured as 'qemu'.

class { 'openstack::compute':
  ...
  libvirt_type => 'qemu'
  ...
}

class { 'openstack::all':
  ...
  libvirt_type => 'qemu'
  ...
}

Installation

Install Puppet

• Puppet should be installed on all nodes:

  apt-get install puppet

• A Puppet master is not required for all-in-one installations. It is, however, recommended for multi-node installations.

  • To install the puppetmaster:

    apt-get install puppetmaster

  • Rake and Git should also be installed on the Puppet Master:

    apt-get install rake git

  • Some features of the modules require storeconfigs to be enabled on the Puppet Master.

  • Create a site manifest site.pp on the master:

      cat > /etc/puppet/manifests/site.pp << EOT
      node default {
        notify { 'I can connect!': }
      }
      EOT
    

  • Restart the puppetmaster service:

    service puppetmaster restart

  • Configure each client to connect to the master and enable pluginsync. This can be done by adding the following lines to /etc/puppet.conf:

      [agent]
        pluginsync = true
        server     = <CONTROLLER_HOSTNAME>
    

  • Register each client with the puppetmaster:

    puppet agent -t --waitforcert 60

  • On the puppetmaster, sign the client certificates:

    puppet cert sign <CERTNAME>

Install the Openstack modules

• The Openstack modules should be installed into the module path of your master or on each node (if you are running puppet apply).

  Modulepath:

  • open source puppet - /etc/puppet/modules
  • Puppet Enterprise - /etc/puppetlabs/puppet/modules

• To install the released versions from the forge:

  puppet module install puppetlabs-openstack

• To install the latest revision of the modules from git (for developers/ contributors):

  cd <module_path>
  git clone git://github.com/puppetlabs/puppetlabs-openstack openstack
  cd openstack
  rake modules:clone
  

puppetlabs-openstack

The 'puppetlabs-openstack' module was written for those who want to get up and running with a single or multi-node Openstack deployment as quickly as possible. It provides a simple way of deploying Openstack that is based on best practices shaped by companies that contributed to the design of these modules.

Classes

openstack::all

The openstack::all class provides a single configuration interface that can be used to deploy all Openstack services on a single host.

This is a great starting place for people who are just kicking the tires with Openstack or with Puppet deployed OpenStack environments.

Usage Example:

An openstack all in one class can be configured as follows:

class { 'openstack::all':
  public_address       => '192.168.1.12',
  public_interface     => 'eth0',
  private_interface    => 'eth1',
  admin_email          => 'some_admin@some_company',
  admin_password       => 'admin_password',
  keystone_admin_token => 'keystone_admin_token',
  nova_user_password   => 'nova_user_password',
  glance_user_password => 'glance_user_password',
  rabbit_password      => 'rabbit_password',
  rabbit_user          => 'rabbit_user',
  libvirt_type         => 'kvm',
  fixed_range          => '10.0.0.0/24',
}

For more information on the parameters, check out the inline documentation in the manifest:

<module_path>/openstack/manifests/all.pp

openstack::controller

The openstack::controller class is intended to provide basic support for multi-node Openstack deployments.

There are two roles in this basic multi-node Openstack deployment:

• controller - deploys all of the central management services
• compute - deploys the actual hypervisor on which VMs are deployed.

The openstack::controller class deploys the following Openstack services:

• keystone
• horizon
• glance
• nova (ommitting the nova compute service and, when multi_host is enabled, the nova network service)
• mysql
• rabbitmq

Usage Example:

An openstack controller class can be configured as follows:

class { 'openstack::controller':
  public_address          => '192.168.101.10',
  public_interface        => 'eth0',
  private_interface       => 'eth1',
  internal_address        => '192.168.101.10',
  floating_range          => '192.168.101.64/28',
  fixed_range             => '10.0.0.0/24',
  multi_host              => false,
  network_manager         => 'nova.network.manager.FlatDHCPManager',
  admin_email             => 'admin_email',
  admin_password          => 'admin_password',
  keystone_admin_token    => 'keystone_admin_token',
  glance_user_password    => 'glance_user_password',
  nova_user_password      => 'nova_user_password',
  rabbit_password         => 'rabbit_password',
  rabbit_user             => 'rabbit_user',
}

For more information on the parameters, check out the inline documentation in the manifest:

<module_path>/openstack/manifests/controller.pp

openstack::compute

The Openstack compute class is used to manage the underlying hypervisor. A typical multi-host Openstack installation would consist of a single openstack::controller node and multiple openstack::compute nodes (based on the amount of resources being virtualized)

The openstack::compute class deploys the following services:

• nova
  • compute service (libvirt backend)
  • optionally, the nova network service (if multi_host is enabled)
  • optionally, the nova api service (if multi_host is enabled)
  • optionally, the nova volume service if it is enabled

Usage Example:

An openstack compute class can be configured as follows:

class { 'openstack::compute':
  private_interface   => 'eth1',
  internal_address    => $ipaddress_eth0,
  libvirt_type        => 'kvm',
  fixed_range         => '10.0.0.0/24',
  network_manager     => 'nova.network.manager.FlatDHCPManager',
  multi_host          => false,
  database_connection => 'mysql://nova:nova_db_passwd@192.168.101.10/nova',
  rabbit_host         => '192.168.101.10',
  glance_api_servers  => '192.168.101.10:9292',
  vncproxy_host       => '192.168.101.10',
  vnc_enabled         => true,
  manage_volumes      => true,
}

For more information on the parameters, check out the inline documentation in the manifest:

<module_path>/openstack/manifests/compute.pp

Creating your deployment scenario

So far, classes have been discussed as configuration interfaces used to deploy the openstack roles. This section explains how to apply these roles to actual nodes using a puppet site manifest.

The default file name for the site manifest is site.pp. This file should be contained in the puppetmaster's manifestdir:

• open source puppet - /etc/puppet/manifests/site.pp
• Puppet Enterprise - /etc/puppetlabs/puppet/manifests/site.pp

Node blocks are used to map a node's certificate name to the classes that should be assigned to it.

Node blocks can match specific hosts:

node my_explicit_host {...}

Or they can use regular expression to match sets of hosts

node /my_similar_hosts/ {...}

Inside the site.pp file, Puppet resources declared within node blocks are applied to those specified nodes. Resources specified at top-scope are applied to all nodes.

Deploying an Openstack all-in-one environment

The easiest way to get started with the openstack::all class is to use the file

<module_dir>/openstack/examples/site.pp

There is a node entry for

node /openstack_all/ {...}

that can be used to deploy a simple nova all-in-one environment.

You can explicitly target this node entry by specifying a matching certname and targeting the manifest explicitly with:

puppet apply /etc/puppet/modules/openstack/examples/site.pp --certname openstack_all

You could also update site.pp with the hostname of the node on which you wish to perform an all-in-one installation:

node /<my_node>/ {...}

If you wish to provision an all-in-one host from a remote puppetmaster, you can run the following command:

puppet agent -td

Deploying an Openstack multi-node environment

A Puppet Master should be used when deploying multi-node environments.

The example modules and site.pp should be installed on the Master.

This file contains entries for:

node /openstack_controller/ {...}

node /openstack_compute/ {...}

Which can be used to assign the respective roles.

(As above, you can replace these default certificate names with the hostnames of your nodes)

The first step for building out a multi-node deployment scenario is to choose the IP address of the controller node.

Both nodes will need this configuration parameter.

In the example site.pp, replace the following line:

$controller_node_address = <your_node_ip>

with the IP address of your controller.

It is also possible to use store configs in order for the compute hosts to automatically discover the address of the controller host. Documentation for this may not be available until a later release of the openstack modules.

Once everything is configured on the master, you can configure the nodes using:

puppet agent -t <--certname ROLE_CERTNAME>

It is recommended that you first configure the controller before configuring your compute nodes:

openstack_controller> puppet agent -t --certname openstack_controller
openstack_compute1>   puppet agent -t --certname openstack_compute1
openstack_compute2>   puppet agent -t --certname openstack_compute2

Verifying an OpenStack deployment

Once you have installed openstack using Puppet (and assuming you experience no errors), the next step is to verify the installation:

openstack::auth_file

The optionstack::auth_file class creates the file:

/root/openrc

which stores environment variables that can be used for authentication of openstack command line utilities.

Usage Example:

class { 'openstack::auth_file':
  admin_password       => 'my_admin_password',
  controller_node      => 'my_controller_node',
  keystone_admin_token => 'my_admin_token',
}

Verification Process

1. Ensure that your authentication information is stored in /root/openrc. This assumes that the class openstack::auth_file had been applied to this node.

2. Ensure that your authenthication information is in the user's environment.

   source /root/openrc

3. Verify that all of the services for nova are operational:

   nova-manage service list Binary Host Zone Status State Updated_At nova-volume <your_host> nova enabled :-) 2012-06-06 22:30:05 nova-consoleauth <your_host> nova enabled :-) 2012-06-06 22:30:04 nova-scheduler <your_host> nova enabled :-) 2012-06-06 22:30:05 nova-compute <your_host> nova enabled :-) 2012-06-06 22:30:02 nova-network <your_host> nova enabled :-) 2012-06-06 22:30:07 nova-cert <your_host> nova enabled :-) 2012-06-06 22:30:04

4. Ensure that the test script has been deployed to the node.

   file { '/tmp/test_nova.sh': source => 'puppet:///modules/openstack/nova_test.sh', }

5. Run the test script.

   bash /tmp/test_nova.sh

This script will verify that an image can be inserted into glance, and that
that image can be used to fire up a virtual machine instance.

6. Log into horizon on port 80 of your controller node and walk through a few operations:

   • fire up a VM
   • create a volume
   • attach that volume to the VM
   • allocate a floating IP address to a VM instance.
   • verify that voluem is actually attached to the VM and that it is reachable by its floating ip address (which will require some security groups)

Building your own custom deployment scenario for Openstack

The classes included in the Openstack module are implemented using a number of other modules. These modules can be used directly to create a customized openstack deployment.

A list of the modules used by puppetlabs-openstack and the source locations for those modules can be found in other_repos.yaml in the openstack module folder.

other_repos.yaml

These building block modules have been written to support a wide variety of specific configuration and deployment use cases. They also provide a lot of configuration options not available with the more constrained puppetlabs-openstack modules.

The manifests in the Openstack module can serve as an example of how to use these base building block to compose custom deployments.

<module_path>/openstack/manifests/{all,controller,compute}.pp

These files contain examples of how to deploy the following services:

• nova
  • api
  • scheduler
  • volumes
  • compute
  • network
• keystone
• glance
  • api
  • registry
• horizon
• database
  • examples only exist for Mysql and Sqlite (there is work underway for postgresql)
• message queue
  • examples currently only exist for rabbitmq

Once you have selected which services need to be combined on which nodes, you should review the modules for all of these services and figure out how you can configure things like the pipelines and back-ends for these individual services.

This information should then be used to compose your own custom site.pp

Deploying swift

In order to deploy swift, you should use the example manifest that comes with the swift modules (examples/site.pp)

In this example, the following nodes are specified:

• swift_proxy
  • used as the ringbuilder + proxy node
• swift_storage_1
  • used as a storage node
• swift_storage_2
  • used as a storage node
• swift_storage_3
  • used as a storage node

This swift configuration requires both a puppetmaster with storeconfigs enabled.

To fully configure a Swift environment, the nodes must be configured in the following order:

• First the storage nodes need to be configured. This creates the storage services (object, container, account) and exports all of the storage endpoints for the ring builder into storeconfigs. (The replicator service fails to start in this initial configuration)
• Next, the ringbuild and swift proxy must be configured. The ringbuilder needs to collect the storage endpoints and create the ring database before the proxy can be installed. It also sets up an rsync server which is used to host the ring database. Resources are exported that are used to rsync the ring database from this server.
• Finally, the storage nodes should be run again so that they can rsync the ring databases.

This configuration of rsync create two loopback devices on every node. For more realistic scenarios, users should deploy their own volumes in combination with the other classes.

Better examples of this will be provided in a future version of the module.

Participating

Need a feature? Found a bug? Let me know!

We are extremely interested in growing a community of OpenStack experts and users around these modules so they can serve as an example of consolidated best practices of how to deploy openstack.

The best way to get help with this set of modules is to email the group associated with this project:

puppet-openstack@puppetlabs.com

Issues should be opened here:

https://projects.puppetlabs.com/projects/openstack

The process for contributing code is as follows:

• fork the projects in github
• submit pull requests to the projects containing code contributions
  • rspec tests are preferred but not required with initial pull requests. I am happy to work with folks to help them get then up and going with rspec-puppet.

Future features:

efforts are underway to implement the following additional features:

• Validate module on Fedora 17 and RHEL
• monitoring (basic system and Openstack application monitoring support with Nagios/Ganglia and/or sensu)
• Redundancy/HA - implementation of modules to support highly available and redundant Openstack deployments.
• These modules are currently intended to be classified and data-fied in a site.pp. Starting in version 3.0, it is possible to populate class parameters explicitly using puppet data bindings (which use hiera as the back-end). The decision not to use hiera was primarily based on the fact that it requires explicit function calls in 2.7.x
• Implement provisioning automation that can be used to fully provision an entire environment from scratch
• Integrate with PuppetDB to allow service auto-discovery to simplify the configuration of service association