Install controller computehci on 3 nodes with tripleO
Hi, Is it technically possible to install openstack tripleO on 3 nodes, such that all 3 nodes play the roles of controller compute and ceph storage (hci) at the same time, with replication and HA and all? Regards.
On Mon, Sep 20, 2021 at 11:46 AM wodel youchi <wodel.youchi@gmail.com> wrote:
Hi,
Is it technically possible to install openstack tripleO on 3 nodes, such that all 3 nodes play the roles of controller compute and ceph storage (hci) at the same time, with replication and HA and all?
It's technically possible for a POC (just compose a role [1] accordingly) but it's not a recommended configuration because the controllers use pacemaker with power fencing which you don't want running on your OSD hosts. E.g. pacemaker could take out 1/3rd of your OSDs and make the cluster backfill. If you're short on nodes then one option is to use four. One as a basic KVM hypervisor to host VMs including a virtual undercloud and three virtual controllers. Then use three physical HCI nodes which host the VMs and OSDs. John [1] https://docs.openstack.org/project-deploy-guide/tripleo-docs/latest/features...
Hi, Thanks, your answer about pacemaker raised another question, VM (instances) high availability is managed by peacemaker in TripleO, could this be a problem when using HCI deployment? Regards. Le lun. 20 sept. 2021 à 16:59, John Fulton <johfulto@redhat.com> a écrit :
On Mon, Sep 20, 2021 at 11:46 AM wodel youchi <wodel.youchi@gmail.com> wrote:
Hi,
Is it technically possible to install openstack tripleO on 3 nodes, such
that all 3 nodes play the roles of controller compute and ceph storage (hci) at the same time, with replication and HA and all?
It's technically possible for a POC (just compose a role [1] accordingly) but it's not a recommended configuration because the controllers use pacemaker with power fencing which you don't want running on your OSD hosts. E.g. pacemaker could take out 1/3rd of your OSDs and make the cluster backfill. If you're short on nodes then one option is to use four. One as a basic KVM hypervisor to host VMs including a virtual undercloud and three virtual controllers. Then use three physical HCI nodes which host the VMs and OSDs.
John
[1] https://docs.openstack.org/project-deploy-guide/tripleo-docs/latest/features...
On Mon, Sep 20, 2021 at 12:18 PM wodel youchi <wodel.youchi@gmail.com> wrote:
Hi,
Thanks, your answer about pacemaker raised another question, VM (instances) high availability is managed by peacemaker in TripleO, could this be a problem when using HCI deployment?
Yes, it's not a recommended combination for the same reasons. John
Regards.
Le lun. 20 sept. 2021 à 16:59, John Fulton <johfulto@redhat.com> a écrit :
On Mon, Sep 20, 2021 at 11:46 AM wodel youchi <wodel.youchi@gmail.com> wrote:
Hi,
Is it technically possible to install openstack tripleO on 3 nodes, such that all 3 nodes play the roles of controller compute and ceph storage (hci) at the same time, with replication and HA and all?
It's technically possible for a POC (just compose a role [1] accordingly) but it's not a recommended configuration because the controllers use pacemaker with power fencing which you don't want running on your OSD hosts. E.g. pacemaker could take out 1/3rd of your OSDs and make the cluster backfill. If you're short on nodes then one option is to use four. One as a basic KVM hypervisor to host VMs including a virtual undercloud and three virtual controllers. Then use three physical HCI nodes which host the VMs and OSDs.
John
[1] https://docs.openstack.org/project-deploy-guide/tripleo-docs/latest/features...
On 9/20/21 17:45, wodel youchi wrote:
Hi,
Is it technically possible to install openstack tripleO on 3 nodes, such that all 3 nodes play the roles of controller compute and ceph storage (hci) at the same time, with replication and HA and all?
I think it could be done by creating a custom role which merges the Controller and ComputeHCI services together but it has important implications; for example, should pacemaker fence a node, that would kill the workload *and* the osds hosted on that node this can be acceptable or not, depending on expectations and other possible mitigations I guess -- Giulio Fidente GPG KEY: 08D733BA
participants (3)
-
Giulio Fidente
-
John Fulton
-
wodel youchi