Minor version update

To get developer understanding of minor updates, first read the operator docs for minor updates and perhaps try to go through the update as an operator would, to get the basic idea.

Assuming operator-level familiarity with the minor updates, let’s look at individual pieces in more detail.

How update commands work

The following subsections describe the individual update commands:

You might also find it helpful to consult this high-level diagram as you read:

Minor update workflow diagram

openstack overcloud update prepare

The update prepare command performs a Heat stack update, mapping some resources to OS::Heat::None in order to prevent the usual deployment config management tasks being performed (running Puppet, starting containers, running external installers like ceph-ansible). See the update prepare environment file.

The purpose of this stack update is to regenerate fresh outputs of the Heat stack. These outputs contain Ansible playbooks and task lists which are then used in the later in the update run phase.

openstack overcloud update run

The update run command utilizes the previously generated Heat stack outputs. It downloads the playbook yamls and their included task list yaml via the config-download mechanisms, and executes the update steps playbook.

The command accepts --nodes or --roles argument to limit which nodes will be targeted during a particular update run execution. Even if the limit matches multiple nodes (e.g. all nodes within one role), the play is executed with serial: 1, meaning that all actions are finished on one node before starting the update on another.

The play first executes update_steps_tasks.yaml which are tasks collected from the update_tasks entry in composable services.

After the update tasks are finished, deployment workflow is is performed on the node being updated. That means reusing host_prep_tasks.yaml and common_deploy_steps_tasks.yaml, which are executed like on a fresh deployment, except during minor update they’re within a play with the aforementioned serial: 1 limiting.

Finally, post_update_tasks are executed. They are utilized by services which need to perform something after deployment workflow during the minor update. The update of the node is complete and the Ansible play continues to update another node.

openstack overcloud external-update run

The external-update run command is used to update the services whose deployment (and update) procedure is not tied to execution on particular overcloud nodes. The deployment/update procedures are thus executed from the undercloud, even though a full overcloud inventory is available for use.

The external update playbook first executes external_update_tasks and then external_deploy_tasks. The execution happens within the same Ansible play, so facts from external_update_tasks are carried over to external_deploy_tasks. This is a mechanism which will allow you to amend what your deploy tasks do based on whether an update is being run or not.

Often it’s not desirable to run the tasks for all services at the same time, so external-update run supports --tags argument to limit which tasks are run.

The mechanisms of external-upgrade and external-update commands and Ansible tasks are the same, but two commands and task hooks are provided because generally in OpenStack we distinguish minor update vs. major upgrade workflows. If your service only has one type of upgrade, you can make the external_update_tasks the same as external_upgrade_tasks by using YAML anchors and references.

openstack overcloud update converge

The update converge command performs a Heat stack update, reverting the previous OS::Heat::None resource mappings back to the values used for regular deployments and configuration updates, and potentially also resets some parameter values. For environments with Ceph, majority of this already happened on ceph-upgrade run, so the final update converge effectively just resets the CephAnsiblePlaybook parameter.

See the update converge environment file.

The purpose of this stack update is to re-run config management mechanisms and assert that the overcloud state matches what is provided by the templates and environment files.

Writing update logic for a service

Simple config/image replacement

If the service is managed by Paunch, it may be that there’s no need to write any update tasks. Paunch can automatically handle simple updates: change in configuration or change of container image URL triggers automatic removal of the old container and creation of new one with latest config and latest image. If that’s all the service needs for updates, you don’t need to create any update_tasks.

Custom tasks during updates

If the service is not managed by Paunch, or if the simple container replacement done by Paunch is not sufficient for the service update, you will need to include custom update logic. This is done via providing these outputs in your composable service template:

  • update_tasks – these are executed before deployment tasks on the node being updated.
  • post_update_tasks – these are executed after deployment tasks on the node being updated.

Update tasks are generally meant to bring the service into a stopped state (sometimes with pre-fetched new images, this is necessary for services managed by Pacemaker). Then the same workflow as during deployment is used to bring the node back up into a running state, and the post-update tasks can then perform any actions needed after the deployment workflow.

Similarly as deployment tasks, the update tasks and post-update tasks are executed in steps.