kolla-ansible/doc/source/admin/deployment-philosophy.rst
wu.chunyang 39933699bb Add cpu_mode into deployment-philosophy.rst
With libvirt 2.0, when using qemu cpu mode is not properly
evaluated and need to be set to cpu_mode = none.

Add this option when kvm is not supported, otherwise user
will still facing errors when launching instances over qemu.

Change-Id: I1e25fc5429b92f77ce87f537467f97b510fa154a
2018-03-26 01:55:28 +00:00

3.9 KiB

Kolla's Deployment Philosophy

Overview

Kolla has an objective to replace the inflexible, painful, resource-intensive deployment process of OpenStack with a flexible, painless, inexpensive deployment process. Often to deploy OpenStack at the 100+ nodes scale, small businesses may require building a team of OpenStack professionals to maintain and manage the OpenStack deployment. Finding people experienced in OpenStack deployment is very difficult and expensive, resulting in a big barrier for OpenStack adoption. Kolla seeks to remedy this set of problems by simplifying the deployment process while enabling flexible deployment models.

Kolla is a highly opinionated deployment tool out of the box. This permits Kolla to be deployable with the simple configuration of three key/value pairs. As an operator's experience with OpenStack grows and the desire to customize OpenStack services increases, Kolla offers full capability to override every OpenStack service configuration option in the deployment.

Why not Template Customization?

The Kolla upstream community does not want to place key/value pairs in the Ansible playbook configuration options that are not essential to obtaining a functional deployment. If the Kolla upstream starts down the path of templating configuration options, the Ansible configuration could conceivably grow to hundreds of configuration key/value pairs which is unmanageable. Further, as new versions of Kolla are released, there would be independent customization available for different versions creating an unsupportable and difficult to document environment. Finally, adding key/value pairs for configuration options creates a situation in which development and release cycles are required in order to successfully add new customizations. Essentially templating in configuration options is not a scalable solution and would result in an inability of the project to execute its mission.

Kolla's Solution to Customization

Rather than deal with the customization madness of templating configuration options in Kolla's Ansible playbooks, Kolla eliminates all the inefficiencies of existing deployment tools through a simple, tidy design: custom configuration sections.

During deployment of an OpenStack service, a basic set of default configuration options are merged with and overridden by custom ini configuration sections. Kolla deployment customization is that simple! This does create a situation in which the Operator must reference the upstream documentation if a customization is desired in the OpenStack deployment. Fortunately the configuration options documentation is extremely mature and well-formulated.

As an example, consider running Kolla in a virtual machine. In order to launch virtual machines from Nova in a virtual environment, it is necessary to use the QEMU hypervisor, rather than the KVM hypervisor. To achieve this result, simply mkdir -p /etc/kolla/config and modify the file /etc/kolla/config/nova.conf with the contents

[libvirt]
virt_type=qemu
cpu_mode = none

After this change Kolla will use an emulated hypervisor with lower performance. Kolla could have templated this commonly modified configuration option. If Kolla starts down this path, the Kolla project could end with hundreds of config options all of which would have to be subjectively evaluated for inclusion or exclusion in the source tree.

Kolla's approach yields a solution which enables complete customization without any upstream maintenance burden. Operators don't have to rely on a subjective approval process for configuration options nor rely on a development/test/release cycle to obtain a desired customization. Instead operators have ultimate freedom to make desired deployment choices immediately without the approval of a third party.