Minimal RHEL 7 support in MOS as a compute node
- As a Cloud owner I want to introduce pre-provisioned RHEL hosts into my MOS
cloud as Compute nodes so that I could run my RHEL-certified workloads on it
A Minimal Viable Product is to provide detailed documentation for enabling
minimal RHEL 7 support in MOS as a compute node to work under Ubuntu-based
control plane. This documentation should consist of the following parts:
- How to validate that RHEL node is ready to be used as a compute node
(validate network interfaces, disks).
- How to add MOS repositories to this node.
- How to install a minimal required set of packages on top of the
pre-provisioned RHEL 7 node to have a possibility to run puppet granular
tasks from Fuel Library.
- How to prepare configuration in astute.yaml for Fuel Library tasks.
- Desription of the procedure (with automated/scripted steps where it’s
possible) to deploy MOS packages on a pre-provisioned RHEL node and
configure Compute services using granular puppet tasks from Fuel Library.
RHEL 7 Compute nodes will not be integrated with Fuel (no web UI, nailgun and
etc changes are required).
The feature is targeted at the Fuel 8.0 release, so each environment which was
deployed by Fuel 8.0 version supports adding of RHEL 7 compute node to it.
Upon successfully delivering MVP for this feature in 8.0, we can continue with
this feature in 9.0 - full track of RHEL packages, automate deployment and
support RHEL computes at scale. Also other roles can be supported.
Fuel Library may need to be updated to work on RHEL (Fuel common tasks,
l23network, neutron-l2-agent and nova-compute related manifests).
This changes will be RHEL-specific (will not affect Ubuntu deployments).
Implement full support of RHEL compute nodes managed by Fuel:
- Adapt Fuel-related package and source code (mcollective, nailgun-agent
and etc) for RHEL.
- Implement provisioning of RHEL nodes.
- Support RHEL subscription (allow user to provide RHEL license).
- Integrate RHEL nodes with Fuel Architecture.
- Allow Fuel to deploy environments with multiple operating systems.
If an existing cluster was upgraded to MOS 8.0, then it’s possible
to add RHEL computes to it.
This feature requires separate hardware server where we will run new
SWARM & BVT tests. We also need to prepare a qcow2 image for compute nodes
and upload this image to all hardware servers where we will execute
new SWARM tests.
We need to configure CI jobs per commit for all changes in RPM packages
and add new tests to SWARM test suite to execute them on daily basis.
All implementation steps should be fully documented and contain examples
of commands and templates for configuration files.
The following workflow should be considered:
User creates base MOS cloud (Fuel Master + Controllers) using standard Fuel
flow (UI or CLI). The target environment configuration to be considered is:
- Ubuntu on Controllers
- Neutron+OVS (VXLAN, VLAN) on networking layer
- Ceph storage for Nova, Glance, Cinder (managed by Fuel, host OS is Ubuntu)
User validates base requirements for Compute node (disks, network interfaces)
User validates connection to MOS repositories and configures them
on RHEL node
User installs base packages for Fuel Library granular tasks and adds
all required puppet modules on RHEL node
User prepares astute.yaml for puppet tasks based on template
which is provided in documentation
User connects pre-provisioned (provisioned and configured without Fuel)
RHEL node to MOS Management network using Fuel Library netconfig task
User configures installed MOS packages to have services up and running
managed by puppet granular tasks from Fuel Library
User validates RHEL compute functionality performing next steps:
- On controller node check that:
- RHEL node is in up state in nova hypervisor list
- RHEL node’s compute service is in up state in nova services list
- RHEL node’s OVS agent is in alive state in neutron agents list
- Test of instance launch on RHEL node
- Test of network connectivity of instance launched on RHEL node
- Validate functionality after RHEL node reboot by repeating steps 1-3
- Primary assignee:
- Ivan Berezovskiy
- Other contributors:
- Sergey Kolekonov
- Mandatory design review:
- Sergii Golovatiuk
- QA engineer:
- Timur Nurlygayanov
- Documentation with instructions which describes how to deploy RHEL compute
- Manual testing of RHEL 7 nodes in a certain lab environment
- Automate testing on CI using RHEL 7 based computes. Some new
tests will be added to existing SWARM test suite by MOS QA team.
- Perfomance testing on real hardware with RHEL 7
- Cloud Deployment engineer can validate whether RHEL node is ready
to be introduced into MOS cloud as a compute node:
- Network interfaces are sufficient and can be configured to operate with
- Disk size and partitioning is sufficient/feasible to enable basic
MOS services on Compute node.
- Packages are successfully installed without any conflicts with official
- Cloud Deployment engineer can deploy MOS packages on existing RHEL node
and standup+configure MOS services to operate under control of Ubuntu-based
- Cloud Deployment engineer can validate that newly introduced RHEL node
is operational and can be introduced into an operating MOS cloud
- There is the way to test all changes to OpenStack components (for Nova and
Neutron projects in 8.0-liberty branch) and their dependencies
under RHEL as well as Ubuntu (automatically or manually)