Adding new Roles and Services

Adding new Roles and Services

If you would like to contribute towards a role to introduce an OpenStack service or an infrastructure service to support an OpenStack deployment, the OpenStack-Ansible project would welcome that contribution and your assistance in maintaining it.

Writing a new role

Here are the steps to write the role:

  1. You can review roles which may be currently in development by checking our specs repository and unmerged specs on review.openstack.org. If you do not find a spec for the role, propose a blueprint/spec (see also the spec template) outlining the new Role. By proposing a draft spec you can help the OpenStack-Ansible community keep track of what roles are being developed and perhaps connect you with others who may be interested and able to help you in the process.

  2. Create a source repository (e.g. on Github) to start your work on the Role.

  3. Generate the reference directory structure for an Ansible role which is the necessary subset of the documented Best Practice. You might use Ansible Galaxy tools to do this for you (e.g. ansible-galaxy init). You may additionally want to include directories such as docs and examples and tests for your role.

  4. Generate a meta/main.yml right away. This file is important to Ansible to ensure your dependent roles are installed and available and provides others with the information they will need to understand the purpose of your role.

  5. Develop task files for each of the install stages in turn, creating any handlers and templates as needed. Ensure that you notify handlers after any task which impacts the way the service would run (such as configuration file modifications). Also take care that file ownership and permissions are appropriate.

    Hint

    Fill in variable defaults, libraries, and prerequisites as you discover a need for them. You can also develop documentation for your role at the same time.

  6. Add tests to the role.

  7. Ensuring the role deploys with an AIO.

Writing tasks in a role

Most OpenStack services will follow a common series of stages to install or update a service deployment. This is apparent when you review tasks/main.yml for existing roles.

  1. pre-install: prepare the service user group and filesystem directory paths on the host or container
  2. install: install system packages, prepare the (optional) service virtual environment, install service and requirements (into a virtual environment)
  3. post-install: apply all configuration files
  4. service add: register the service (each of: service type, service project, service user, and endpoints) within Keystone’s service catalog.
  5. service setup: install a service-startup script (init, upstart, systemd, etc.) so that the service will start up when the container or host next starts.
  6. service init/startup: signal to the host or container to start the services, make sure the service runs on boot.

There may be other specialized steps required by some services but most of the roles will perform all of these at a minimum. Begin by reviewing a role for a service that has something in common with your service and think about how you can fit most of the common service setup and configuration steps into that model.

Hint

Following the patterns you find in other roles can help ensure your role is easier to use and maintain.

Keep in mind a role candidate for inclusion should respect our Ansible Style Guide.

Adding tests to a role

Each of the role tests is in its tests/ folder.

This folder contains at least the following files:

  1. test.yml (“super” playbook acting as test router to sub-playbooks)
  2. <role name>-overrides.yml. This var file is automatically loaded by our shell script in our tests repository.
  3. inventory. A static inventory for role testing. It’s possible some roles have multiple inventories. See for example the neutron role with its lxb_inventory, calico_inventory.
  4. group_vars and host_vars. These folders will hold override the necessary files for testing. For example, this is where you override the IP addresses, IP ranges, and ansible connection details.
  5. ansible-role-requirements.yml. This should be fairly straightforward: this file contains all the roles to clone before running your role. The roles’ relative playbooks will have to be listed in the test.yml file. However, keep in mind to NOT re-invent the wheel. For example, if your role needs keystone, you don’t need to create your own keystone install playbook, because we have a generic keystone install playbook in the tests repository.

Deploying the role

  1. Include your role on the deploy host. See also Adding Galaxy roles.

  2. Perform any other host preparation (such as the tasks performed by the bootstrap-aio.yml playbook). This includes any preparation tasks that are particular to your service.

  3. Generate files to include your service in the Ansible inventory using env.d and conf.d files for use on your deploy host.

    Hint

    You can follow examples from other roles, making the appropriate modifications being sure that group labels in env.d and conf.d files are consistent.

  4. Generate secrets, if any, as described in the Configure Service Credentials. You can append your keys to an existing user_secrets.yml file or add a new file to the openstack_deploy directory to contain them. Provide overrides for any other variables you will need at this time as well, either in user_variables.yml or another file. This is explained in more depth under Extending OpenStack-Ansible.

  5. If your service is installed from source or relies on python packages which need to be installed from source, specify a repository for the source code of each requirement by adding a file to your deploy host under playbooks/defaults/repo_packages in the OpenStack-Ansible source repository and following the pattern of files currently in that directory. You could also simply add an entry to an existing file there. Be sure to run the repo-build.yml play later so that wheels for your packages will be included in the repository infrastructure.

  6. Make any required adjustments to the load balancer configuration (e.g. modify playbooks/vars/configs/haproxy_config.yml in the OpenStack-Ansible source repository on your deploy host) so that your service can be reached through a load balancer, if appropriate, and be sure to run the haproxy-install.yml play later so your changes will be applied.

  7. Put together a service install playbook file for your role. This can also be modeled from any existing service playbook that has similar dependencies to your service (database, messaging, storage drivers, container mount points, etc.). A common place to keep playbook files in a Galaxy role is in an examples directory off the root of the role.

Hint

If you adhere to the pattern of isolating your role’s extra deployment requirements (secrets and var files, HAProxy yml fragments, repo_package files, etc.) in their own files it makes it easy for you to automate these additional steps when testing your role.

Role development maturity

In order to include a role into the integrated build implemented by the playbooks in the openstack/openstack-ansible repository, it needs to meet an appropriate level of maturity requirements. Developers are encouraged to observe the below-mentioned patterns in the existing roles.

Note

Many of the existing roles may not fully implement all the patterns just yet. It is important to look through multiple roles to get an idea of the common patterns rather than to fixate on a single role’s pattern.

The development of a role will usually go through the following stages:

  1. Initial role development

    • Include base scaffolding. To facilitate development and the tests implemented across all OpenStack-Ansible roles, a base set of folders and files need to be implemented. A base set of configuration and test facilitation scripts must include at least the following:

      • tox.ini: The lint testing, documentation build, release notes build and functional build execution process for the role’s gate tests are all defined in this file.
      • test-requirements.txt: The python requirements which must be installed when executing the tests.
      • other-requirements.txt: The binary requirements which must be installed on the host the tests are executed on for the python requirements and the tox execution to work.
      • setup.cfg and setup.py: Information about the repository which is used during the build of any artifacts.
      • run_tests.sh: A convenient script for developers to execute all standard tests on a suitable host.
      • Vagrantfile: A convenient configuration file to allow a developer to easily create a test virtual machine using Vagrant. This must automatically execute run_tests.sh.
      • README.rst, LICENSE, CONTRIBUTING.rst: A set of standard files which have content describing their purpose.
      • .gitignore: A standard git configuration file for the repository which should be pretty uniform across all the repositories.
      • .gitreview: A standard file configured for the project to inform the git-review plugin where to find the upstream gerrit remote for the repository.
    • The role development should initially be focused on implementing a set of tasks and a test playbook which converge. The convergence must:

      • Implement developer_mode to build from a git source into a Python venv.
      • Deploy the applicable configuration files in the right places.
      • Ensure that the service starts.

      The convergence may involve consuming other OpenStack-Ansible roles (For example: galera_server, galera_client, rabbitmq_server) in order to ensure that the appropriate infrastructure is in place. Reuse of existing roles in OpenStack-Ansible or Ansible Galaxy is strongly encouraged.

    • The role must support Ubuntu 16.04 LTS. It should ideally also support CentOS7 but this is not required at this time. The patterns to achieve this include:

      • The separation of platform specific variables into role vars files.
      • The detection and handling of different init systems (init.d, SystemD).
      • The detection and handling of different package managers (apt, yum).
      • The detection and handling of different network configuration methods.

      There are several examples of these patterns implemented across many of the OpenStack-Ansible roles. Developers are advised to inspect the established patterns and either implement or improve upon them.

    • The role implementation should be done in such a way that it is agnostic with regards to whether it is implemented in a container, or on a physical host. The test infrastructure may make use of LXC containers for the separation of services, but if a role is used by a playbook that targets a host, it must work regardless of whether that host is a container, a virtual server, or a physical server. The use of LXC containers for role tests is not required but it may be useful in order to simulate a multi-node build out as part of the testing infrastructure.

    • Any secrets (For example: passwords) should not be provided with default values in the tasks, role vars, or role defaults. The tasks should be implemented in such a way that any secrets required, but not provided, should result in the task execution failure. It is important for a secure-by-default implementation to ensure that an environment is not vulnerable due to the production use of default secrets. Deployers must be forced to properly provide their own secret variable values.

    • Once the initial convergence is working and the services are running, the role development should focus on implementing some level of functional testing. Ideally, the functional tests for an OpenStack role should make use of Tempest to execute the functional tests. The ideal tests to execute are scenario tests as they test the functions that the service is expected to do in a production deployment. In the absence of any scenario tests for the service a fallback option is to implement the smoke tests instead.

    • The role must include documentation. The Documentation and Release Note Guidelines provide specific guidelines with regards to style and conventions. The documentation must include a description of the mandatory infrastructure (For example: a database and a message queue are required), variables (For example: the database name and credentials) and group names (For example: The role expects a group named foo_all to be present and it expects the host to be a member of it) for the role’s execution to succeed.

  2. Integration development

    Once the role has implemented the above requirements, work can begin on integrating the role into the integrated build. This involves the preparation of the following items:

    • Host and container group configuration

      This is implemented into the dynamic inventory through the definition of content in an env.d file. A description of how these work can be found in Appendix C of the Deployment Guide.

    • Load balancer configuration

      OpenStack-Ansible deploys services in a highly available configuration by default, so all API services must be configured for implementation behind HAProxy. This is done through the modification of playbooks/vars/configs/haproxy_config.yml.

    • Install playbook

      In order to implement the role in the appropriate way, an os-<service>-install.yml playbook must be created and targeted at the appropriate group defined in the service env.d file. The playbook should also ensure that the database(s), database user(s), rabbitmq vhost and rabbitmq user are setup for the service. It is crucial that the implementation of the service is optional and that the deployer must opt-in to the deployment through the population of a host in the applicable host group. If the host group has no hosts, Ansible skips the playbook’s tasks automatically.

    • Secrets

      Any secrets required for the role to work must be noted in the etc/openstack_deploy/user_secrets.yml file.

    • Group vars

      Any variables needed by other roles to connect to the new role, or by the new role to connect to other roles, should be implemented in playbooks/inventory/group_vars. The group vars are essentially the glue which playbooks use to ensure that all roles are given the appropriate information. When group vars are implemented it should be a minimum set to achieve the goal of integrating the new role into the integrated build.

    • Documentation

      Content must be added to the Installation Guide to describe how to implement the new service in an integrated environment. This content must adhere to the Documentation and Release Note Guidelines. Until the role has integrated functional testing implemented, the documentation must make it clear that the service inclusion in OpenStack-Ansible is experimental and is not fully tested by OpenStack-Ansible in an integrated build.

    • Release note

      A feature release note must be added to announce the new service availability and to refer to the Installation Guide entry and the role documentation for further details. This content must adhere to the Documentation and Release Note Guidelines.

    • Integration test

      It must be possible to execute a functional, integrated test which executes a deployment in the same way as a production environment. The test must execute a set of functional tests using Tempest. This is the required last step before a service can remove the experimental warning from the documentation.

Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.