Independent Role Repositories

date:

2015-08-17 14:00

tags:

roles

In order to improve the ability to independently consume the roles produced by openstack-ansible in different reference use-case deployments and allow independent development of each role by different projects, this specification proposes that:

  1. New roles be registered in separate repositories named openstack/openstack-ansible-<role>.

  2. Existing roles can, through an independent blueprint/spec process, be split into their own repositories.

This provides the following benefits:

  • Other projects (eg: DevStack, Kolla, Compass, RPC, etc) will be able to consume the roles using their own playbooks. This increases the opportunity for other projects to collaborate with openstack-ansible.

  • The roles will be more easily consumed in different reference architectures. Currently the playbooks and roles are specifically geared for deployment in LXC containers. Making the roles independent entities will allow them to be consumed for completely different architectures. eg: a deployment without containers, a deployment with VM’s instead of containers, a deployment with a different container technology.

  • Each role can be developed at its own pace and versioned independently. This will make openstack-ansible more like the OpenStack “Big Tent”.

  • Each role can be independently gate checked with checks that are specific and relevant to the role. This will provide quicker developer feedback and allow a quicker turnaround for development.

  • The roles can be registered in Ansible Galaxy. This will provide greater awareness of the roles and may attract more contributors. This is especially useful for the infrastructure roles which may have a broader application than just for use in an OpenStack deployment. (eg: haproxy, MariaDB, etc)

  • Role separation will simplify the mission of openstack-ansible’s playbooks and scripts to be for the purpose of providing examples of how to consume the roles and to implement gate testing for integrated deployment verification of the use-cases that matter to the community. This prevents the situation where the playbooks have to cater for every possible combination use-case that may be thrown at them.

Problem description

A detailed description of the problem:

  • Currently the openstack-ansible roles are tightly coupled with the playbooks that consume them. While the roles can technically be consumed using different playbooks, this is not immediately obvious to downstream consumers.

  • When consumers to try to consume the roles with different playbooks for a different architecture, they are forced to implement many workarounds for the tight coupling that we have. Even when it is possible to do, it is hard for a deployer to see how to do it due to the deluge of variables that need to be set and code that needs to be read.

  • There has been some interest in creating roles for other services (eg: rally, congress, etc). Implementing the strategy of a repo per role allows these fledgling roles to get into the open and get collaborated on far more quickly. Once they’re at the point where they’re ready to be integrated into an integrated use-case with integration gate tests, then they can tag a version and implement the openstack-ansible playbooks and scripts to test the appropriate use-case.

  • The infrastructure roles implemented as part of the project are not getting much attention as they are more like a second-class citizen within the role structure. This results in the configurations deployed often not lining up to best-practices.

  • Any small changes to roles require the execution of a full integration gate test which is slow and prone to error. It is difficult to isolate the error in the current monolithic stack.

Proposed change

The following documentation should be developed:

  1. The primary use-cases/implementations tested by the project.

  2. How to apply to add a new use-cases for integration testing.

  3. How to register a new role within openstack-ansible’s umbrella.

  4. First steps for building a new role.

  5. Update the README for each role to describe how to use it independently, whether using a static inventory or the dynamic inventory. It should also cover what options are available for its use, whether it relies on any other roles, how upgrades are handled, any known issues, etc.

The following process should be followed for registering new roles:

  1. A blueprint must be registered and a spec for the implementation provided for review, with specific attention paid to any changes that would need to be made in the current openstack-ansible playbooks and roles to integrate the new role.

  2. Once the spec is approved, the review to register the new repository should be registered upstream by the openstack-ansible PTL or a nominated openstack-ansible-core team member.

  3. Once the new repository has been created, work can commence on the new role.

  4. Before the first tag is set for the role, comprehensive testing for the role must already be in place.

The following process should be followed for breaking out existing roles:

  1. For each role targeted for breaking out, a separate blueprint must be registered and a spec provided for the high level changes that would be required in the current openstack-ansible playbooks and roles to accommodate this change.

  2. Once the spec is approved, the review to register the new repository should be registered upstream by the openstack-ansible PTL or a nominated openstack-ansible-core team member.

  3. Once the new repository has been created, work can commence on extracting the role as planned in the spec for the role’s migration.

  4. Once independent gate checks on the role repository confirm that it is in working order and the work is done to prepare the role for usage in the integrated use-cases, tag the initial version of the role and implement the openstack-ansible playbook, script and role-requirements changes to consume the new role. The changes in openstack-ansible will need to pass the integrated gate checks before they can merge.

Alternatives

Leave everything as it is and continue to merge any new roles proposed into the same monolithic repository.

Playbook/Role impact

The impact to playbooks should be fairly incremental, but will need to be determined on a role by role basis. This must be described in the spec on a per role basis.

It is clear that the libraries, filters and plugins will need to be broken out into its own repository and each role that consumes them will need to use a submodule reference in the role. This ensures that all the roles use a common set of libraries, filters and plugins.

Upgrade impact

There are two aspects of upgrade impact to be considered:

  1. The ability for the role itself to handle upgrades from previous interations of itself.

  2. The ability for the integrated build use-cases that consume the role to be upgradable.

Each role should be developed with upgradability in mind and conform to the upgrade framework which is being developed for the Kilo -> Liberty upgrade process.

Security impact

n/a

Performance impact

n/a

End user impact

End users will not know of any differences.

Deployer impact

The structure and placement of roles on the deployment host will be different. The changes will need to be documented for support purposes. However, the configuration and execution of the existing playbooks for downstream consumers should be targeted to be exactly the same to minimise disruption.

Any role-specific impacts will need to be defined on a per-role basis.

Developer impact

The biggest negative developer impact will be the difficulty working between the two different structures - master being split, with liberty and kilo being consolidated. It may be worth considering ways to make them all work the same way once the conversions are done for master.

The positive impact has been outlined in the introduction.

Dependencies

This should only be implemented after liberty has been released.

Implementation

Assignee(s)

Primary assignee:

https://launchpad.net/~jesse-pretorius odyssey4me https://launchpad.net/~kevin-carter cloudnull

Other contributors:

TBD

Work items

See proposed change section.

Testing

The testing impact will need to be described on a per-role basis.

As a standard, the following tests are expected to be implemented as a standard for each role:

  • bashate tests for all shell scripts

  • pep8 tests for all python files

  • ansible syntax checks for all ansible task files

  • ansible-lint tests for all ansible task files

  • functional tests for the service

Integration tests are expected to be implemented in the openstack-ansible repository and executed whenever the role versions are incremented. This ensures that a role tag increment is only accepted for an integrated release once it passes a full set of integration tests.

Documentation impact

While the placement of the role files on the deployment host will be different, the configuration and execution of the deployment should remain the same, resulting in minimal documentation impact.

See the proposed change section for developer reference documentation to be developed.

References

n/a