Architectural Blueprint for True Multi-Region - Single Management Plane

[dahampere]

We are expanding our single-site Apache CloudStack deployment (2x Management Nodes, 3-node Galera cluster) across multiple geographically distant datacenters.

We want to maintain a single management plane, but every standard approach seems to have a critical flaw:

1. Centralized DB: Keeping the DB in Site A makes it a massive single point of failure and forces Site B hypervisors to traverse WAN/cross-connects for chatty management traffic.
2. Stretched Galera: Stretching synchronous replication across the WAN destroys write performance due to latency inflation, backing up management worker threads.
3. Async DB + Standby Nodes: Site B management nodes can't run (unless again pointed towards the cross-site active DB nodes) because CloudStack requires a writable DB and doesn't support read-only backends.
4. Native "Regions": It’s a loose UI wrapper. Offerings, global settings, and accounts don't actually sync, forcing us to build heavy external automation (Ansible/Terraform/SAML) just to keep them identical.

I can't be the first person to hit this wall.

• What are my options?
• Does CloudStack have a roadmap to solve this?
• Or am I misunderstanding CloudStack and using it for something it was not intended for?

[ingox]

It is mostly a question of the latency between the datacenters.

High latency:

• You'll need to go with option 4.

Medium latency:

• You'll go with option 3. Make sure active management server and writable database are in the same location. For DR purposes you'll have an offline management server and a async replication of your database. Latency between your hosts and the management are not that critical than between management and database.

Low latency:

• You can place management servers and databases across your locations.