[kolla][kolla-ansible] distro mixing on hosts
Hi, We have been pretty consistent in deploying CentOS 7 on physical hardware where Kolla Ansible deploys its containers. However, we're going to be switching to Debian 10.8. It looks like Kolla Ansible likely won't care which distro is installed on the physical hardware, and so some nodes could have CentOS 7 and others have Debian 10.8. Long-term, we'll replace the CentOS 7 machines with Debian 10.8 and re-deploy. Kolla Container images will still be CentOS 7, but we will be recreating Kolla images with Debian and deploying those long term too. Is this supported? Any obvious issues we should be aware of? We will have the same interface names, so we will be pretty consistent in terms of the environment - but wanted to see if there was any gotchas that we didn't think of. Also, I guess I should ask if there are any specific Debian 10.8 issues we should be aware of? Has anybody else done this? :) Thanks! Eric
On Wed, 3 Mar 2021 at 22:38, Eric K. Miller <emiller@genesishosting.com> wrote:
Hi,
We have been pretty consistent in deploying CentOS 7 on physical hardware where Kolla Ansible deploys its containers. However, we're going to be switching to Debian 10.8.
It looks like Kolla Ansible likely won't care which distro is installed on the physical hardware, and so some nodes could have CentOS 7 and others have Debian 10.8. Long-term, we'll replace the CentOS 7 machines with Debian 10.8 and re-deploy.
Kolla Container images will still be CentOS 7, but we will be recreating Kolla images with Debian and deploying those long term too.
Is this supported? Any obvious issues we should be aware of? We will have the same interface names, so we will be pretty consistent in terms of the environment - but wanted to see if there was any gotchas that we didn't think of.
Hi Eric, While this is possible, it is not something we "support" upstream. We already have many combinations of OS distro and binary/source. I know that people have tried it in the past, and while it should mostly just work, occasionally it causes issues. A classic issue would be using an ansible fact about the host OS to infer something about the container. While we'll likely accept reasonable fixes for these issues upstream, mixing host and container OS may cause people to pay less attention to bugs raised. If you want to migrate from CentOS 7 to Debian, I would suggest keeping the host and container OS the same, by basing the kolla_base_distro variable on groups. Of course this will also be an untested configuration, and I suggest that you test it thoroughly, in particular things like migration between different versions of libvirt/qemu. Mark
Also, I guess I should ask if there are any specific Debian 10.8 issues we should be aware of?
Has anybody else done this? :)
Thanks!
Eric
While this is possible, it is not something we "support" upstream. We already have many combinations of OS distro and binary/source. I know that people have tried it in the past, and while it should mostly just work, occasionally it causes issues. A classic issue would be using an ansible fact about the host OS to infer something about the container. While we'll likely accept reasonable fixes for these issues upstream, mixing host and container OS may cause people to pay less attention to bugs raised.
Understood. If we run into issues, I'll report regardless of whether there is attention, just so someone knows what was tried, and what failed.
If you want to migrate from CentOS 7 to Debian, I would suggest keeping the host and container OS the same, by basing the kolla_base_distro variable on groups. Of course this will also be an untested configuration, and I suggest that you test it thoroughly, in particular things like migration between different versions of libvirt/qemu.
Ah, that's a good idea, setting kolla_base_distro in groups as opposed to globally. Thanks! If we can do that, then we'll definitely keep both host and container OS the same. We're testing in a VM environment first, so will try to identify problems before we have to do this on live systems. Thanks again for the quick responses as always Mark! Eric
participants (2)
-
Eric K. Miller
-
Mark Goddard