Copyright 2015 Hewlett-Packard Development Company, L.P. This work is licensed under a Creative Commons Attribution 3.0 Unported License. http://creativecommons.org/licenses/by/3.0/legalcode
With donated hardware and datacenter space, we can run an optimized semi-private cloud for the purpose of adding testing capacity and also with an eye for “dog fooding” OpenStack itself.
Currently all of the test resources that we use are provided by public clouds. This is very useful to us and is also a good demonstration of a cross-public-cloud OpenStack application. Some organizations are also able to provide hardware instead of (or in addition to) public cloud resources. By operating that hardware as a private cloud with only OpenStack Infrastructure as a tenant, we can expand our test capacity and also demonstrate a public-private hybrid OpenStack application.
Further, we can operate the cloud in a completely transparent manner as we do the rest of the project infrastructure and help bridge the gap between developers and operators.
This spec describes the process of standing up the initial infra-cloud, but intentionally does not delve into technical detail. Many of those decisions will need to be made and updated as the process unfolds, and also need to be recorded as system documentation. Therefore, most of the actual technical decisions and documentation will happen in the system-config repository in the doc/source/infra-cloud.rst file.
In order to accept a donation of hardware from an organization, we will also need to be provided a contact from the organization that can help us with any hands-on work needed to maintain the machines. Newly donated hardware will be inventoried and standardized, and then an infra-cloud region will be deployed on it, as described in system-config.
We have an initial hardware donation from HP in two data centers, which we will stand up as two clouds. Once these clouds are well established, we can consider adding new clouds based on further donations as needed. We may consider running a single centralized keystone in the future and combine the separate clouds into one cloud with multiple regions.
Infra-cloud is run like any other infra managed service. Puppet modules and Ansible do the bulk of configuring hosts, and Gerrit code review drives 99% of activities, with logins used only for debugging and repairing the service.
Our CI system itself is fault-tolerant across clouds, so no individual region of infra-cloud (or even infra-cloud as a whole) should be considered a critical piece of infrastructure. There is no uptime guarantee and in the case of any error, we should be content to operate the CI system without part or all of infra-cloud until the error can be corrected in due course.
In order to focus on our initial deployment goals, we are strictly limiting the scope of infra-cloud. In particular it is not a general purpose cloud to provide services to any user other than the project infrastructure, and it is not intended to provide a special test environment (e.g., bare metal) not otherwise provided by public clouds. It is also not intended as a test system for OpenStack itself; the update frequency of the version of OpenStack deployed on infra-cloud is not defined. We may deploy new versions of OpenStack as needed, or we may continue to run a stable version for a length of time.
Continue to use only externally provided clouds.
- Primary assignee:
Use Gerrit topic “infra-cloud” for all patches related to this spec.
git-review -t infra-cloud
- Normalize HP hardware
- Agree on initial deployment choices (in system-config)
- Write puppet implementation
- Begin use in nodepool
No new repos are currently anticipated.
Many, as specified in system-config documentation.
A DNS entry should be registered for each Keystone auth endpoint. Other individual servers may also get their own DNS entries.
The system should be documented from before implementation and kept up to date in system-config.
The only tenant will be OpenStack Infrastructure, and the tenant credentials will be managed in the normal way (using hiera) for CI clouds. The administrative credentials will be similarly managed. We would like to make a considerable amount of operational logging available publicly. We will need to be concerned about leaking credentials through that process.
If nodepool works with it, it’s good. We could run tempest or refstack against it as well if we want.
The technical decisions will be made in the system-config repository.