Transition to kolla
Hello, is there anyone who already did transition on existing cluster, from package-version to kolla-ansible managed installation? Looking for guides, tips, your experience on that. BR, crazik
Hello, I haven’t done such migration but in theory it should be possible. It depends on how your storage and networking looks like on the current env. Have you thought about creating a new env and migrating workload in stages from the old one to the new one? You can start with a small new env and after freeing compute and storage hardware after first migration you can move them to the new env to make more space for a next migration stage. BR, Piotr On Wed, 17 May 2023 at 11:02 Ryszard Mielcarek <crazy+kolla@geeks.pl> wrote:
Hello, is there anyone who already did transition on existing cluster, from package-version to kolla-ansible managed installation? Looking for guides, tips, your experience on that.
BR, crazik
Dnia 2023-05-17, o godz. 11:53:11 Piotr Misiak <piotrmisiak1984@gmail.com> napisał(a):
Hello,
I haven’t done such migration but in theory it should be possible. It depends on how your storage and networking looks like on the current env. Have you thought about creating a new env and migrating workload in stages from the old one to the new one? You can start with a small new env and after freeing compute and storage hardware after first migration you can move them to the new env to make more space for a next migration stage.
Hey Piotr, thanks for the idea, but my priority is minimal impact for the VMs (I want only live-migrate them during takeover). With new env I will end up with downtime for each VM and IP change. BR, crazik
BR, Piotr
On Wed, 17 May 2023 at 11:02 Ryszard Mielcarek <crazy+kolla@geeks.pl> wrote:
Hello, is there anyone who already did transition on existing cluster, from package-version to kolla-ansible managed installation? Looking for guides, tips, your experience on that.
BR, crazik
Hey, We did such migration from RDO to the OSA deployment and it was quite resilient for the end users. But we've played with LB quite a lot, and had new hardware for net nodes and control plane, but other then that - it's completely possible to achieve. I think as a first step we've spawned new rabbit and mariadb cluster, though it was easier for us as we ran galera before, so we were able to just scale up galera cluster and point LB to the new IP. And then it was just about setup of new services with osa and pointing haproxy to the new set of the backends. As long as you're deploying the same major version it should be fine to do so. When it comes to net nodes, we migrated l3 and DHCP agents to the freshly provisioned nodes with quite simple script, that list all agents for agent A, removes agent it from A and adds to B afterwards. For compute nodes, I think we didn't do even live migrates - just stopped old service and provisioned new in place. As hostname remains the same, it was just picked up and placement wasn't complaining. But yeah, live migration should just work as long as you have same os version. I think this all should work nicely for kolla as well. ср, 17 мая 2023 г., 12:12 Ryszard Mielcarek <crazy+kolla@geeks.pl>:
Dnia 2023-05-17, o godz. 11:53:11 Piotr Misiak <piotrmisiak1984@gmail.com> napisał(a):
Hello,
I haven’t done such migration but in theory it should be possible. It depends on how your storage and networking looks like on the current env. Have you thought about creating a new env and migrating workload in stages from the old one to the new one? You can start with a small new env and after freeing compute and storage hardware after first migration you can move them to the new env to make more space for a next migration stage.
Hey Piotr, thanks for the idea, but my priority is minimal impact for the VMs (I want only live-migrate them during takeover). With new env I will end up with downtime for each VM and IP change.
BR, crazik
BR, Piotr
On Wed, 17 May 2023 at 11:02 Ryszard Mielcarek <crazy+kolla@geeks.pl> wrote:
Hello, is there anyone who already did transition on existing cluster, from package-version to kolla-ansible managed installation? Looking for guides, tips, your experience on that.
BR, crazik
Hi, On 17/5/23 17:01, Ryszard Mielcarek wrote:
is there anyone who already did transition on existing cluster, from package-version to kolla-ansible managed installation? Looking for guides, tips, your experience on that.
We have just completed doing exactly that, from Puppet OpenStack (train) with each service in its own VM, to a straightforward Kolla Ansible (train-eol) config on 3 control plane nodes and same compute nodes with staged migration of the instances. This page gave us great inspiration and structure to the approach: https://www.stackhpc.com/migrating-to-kolla.html Our only outage was to migrate databases, haproxy, keepalived and rabbitmq. It may have been possible to avoid that outage, but we decided to simplify those steps a little. Compute nodes ended up having trouble with mismatching vcpu features and security drivers, so we scheduled and cold migrated the instances across to kolla compute nodes. I am very pleased with the experience so far, and grateful to the Kolla Ansible community past and present. Next up, upgrades! Greg.
participants (4)
-
Dmitriy Rabotyagov
-
Gregory Orange
-
Piotr Misiak
-
Ryszard Mielcarek