Hide Forgot
Description of problem: Officially, RHEL OSP is currently (as of OSP 9) only officially supported on bare metal, however there are support exceptions in place and many proof-of-concept or dev environments running virtual nodes. The minimum requirements for virtual environments should be well documented. Actual results: The official Installation and Usage guide for OSP only includes instructions for using bare metal hosts. Expected results: In practice, there have been many installations which used virtual machines for at least the Undercloud, if not at least some of the Overcloud nodes. Even though virtual machines are not currently supported without a support exception for overcloud nodes, we have some support exceptions in place, and at least one partner (I think at least two) who recommend using virtualization. The installation, configuration, and minimum requirements for virtualization should be available for those customers with a support exception, or for proof-of-concept environments which do not receive production support. Furthermore, since much of the testing of RHEL OSP is done in virtual machines, this will provide a guide for setting up virtual environments for OSP. Additional info: If we have a large number of customers doing installations on virtual machines without documentation, some are bound to run into problems. For instance, certain hypervisors work better than others for OpenStack virtualization. For instance, VMware doesn't play nice with stock Neutron because of the way Neutron self-generates MAC addresses. In order for Neutron to work properly on VMware, you must enable promiscuous mode, forged retransmits, and MAC address spoofing. This has the side-effect of greatly reduced performance. RHEV does not suffer from this same behavior, making it a higher-performing choice as a hypervisor for emulating bare-metal OpenStack nodes. Furthermore, networking can be difficult to set up when using virtual machines as bare-metal OpenStack nodes. We should fully document the procedure for setting up the proper bridges.
Adding Keith (assuming all overcloud nodes here, undercloud is handled in other requests) I'd suggest that we should either generate kbase articles with appropriate not-supported disclaimers or instead work to generally document and support overcloud controller services on virtual hardware. I expect that we'll never support compute services on virtual hardware, so I suspect that kbase articles would be the right path here.
Hi Dan, The OSP 10 Director Installation and Usage guide now includes a section about virtualization support and requirements, which covers many (if not all) of the points you mention in this bug: https://access.redhat.com/documentation/en/red-hat-openstack-platform/10/single/director-installation-and-usage/#virtualization_support If there are further requirements for the official docs, please advise. If not, I'll close this bug.
No further requirements have been identified. Closing this bug. Please reopen if needed, with a clear description of what is missing.