We use a similar approach to deploy the 'service' region first, which contains of Keystone+Horizon+MariaDB+Memcached and HA (keepalived, haproxy, galera) components on 3 separate controller nodes. Then we deploy KVM regions (Compute, Ironic, etc) and add them into the 'service' region, also integrate each with external CEPH backend (for Cinder, Glance, etc). With HA configuration and correct architecture each region shouldn't go down.

On Tue, May 23, 2023 at 5:10 PM Danny Webb Danny.Webb@thehutgroup.com wrote:

We tested this out here at THG (a bit too late as we'd already deployed multiple regions). What we did was install a stretched galera cluster across multiple DCs with clustered proxysql fronting it so we could force writes to a single node. We then deployed keystone in each regions on top of this. We put keysone behind a geo-loadbalancer so each region would prefer talking to their local keystones and proxy sql would force writes to a single node in the stretched cluster avoiding issues with commit latency and deadlocks.

Horizon doesn't need to exist in a single region, it can exist in all regions as long as it has api connectivity to all regions endponts and the "global" keystone.

If you are using kolla we found it easier to have a specific deployment for keystone only and then each region had it's normal deployment but with a remote keystone.

---

*From:* wodel youchi wodel.youchi@gmail.com *Sent:* 23 May 2023 14:02 *To:* OpenStack Discuss openstack-discuss@lists.openstack.org *Subject:* multi-region deployment

• CAUTION: This email originates from outside THG *

---

Hi,

I have been reading about multi-region in openstack.

From my understanding, that keystone and horizon are shared and deployed only in the first region.

Is this correct? If yes, what happens if the first region goes down?

Regards.