I'm just trying to understand the prefect 2.0 architecture from a scaling perspective. I'm just trying to picture it in my mind.. I understand 3 of the components needed to self host are: • Orion Server • UI • API Server All these 3 components are part of Orion. So to scale do we have multiple replicas start with API Server HA:

--scheduler       --no-scheduler
│ --ui              --no-ui

UI Server HA:

--scheduler       --no-scheduler

Or am I thing through this in the wrong fashion

Hi Vishy, You can scale the front-end however you’d like - I believe it’s a ReplicaSet, which is fronted by a service, so UI / API can be scaled easily. The agent, is entirely up to you - it can be one agent. or many agents with different tags. The database is the harder piece to scale properly, because if you are self-hosting prefect server, your best option is to scale vertically

@Christopher Boyd I agree, the agents is upto us and that's a great feature with Prefect 🎉 So If I get you right, the scheduler part of Orion paired with the database SHOULD be scaled vertically as a guidance. So technically if our DB is HA, we could let the scheduler failover to another scheduler (dormant) in case of DR needs

yep, I think that would be the best solution.