Description
Overview
We need a loose, multi-cluster diagram for a geo-distributed 10k user architecture as the first step in executing a 10k user reference architecture. This diagram would not go directly into docs, so it can be a simple blueprint in TLDraw just to inform the shape of this test.
Loosely, this is the shape we've discussed informally:
Coder Deployment represented as regionally distributed clusters in (US-WEST, EU-WEST, AP-WEST). To match a realistic scenario of our ICP, it would be on a private network with VPC interconnect.
Each region's cluster contains the following:
- A dedicated organization
- A workspace proxy
- External provisioner to spin up workspaces
- ~20 nodes to facilitate ~3,300 workspaces in the region
- Pod to generate load, simulating 4 types of traffic for ~3.3k users
- SSH traffic over direct connections
- Workspace app traffic over WSProxies
- PTY (Terminal) traffic over WSProxies
- An idle browser window with the dashboard open (headless)
For the sake of clear communication, we could name the clusters from east to west: "Home", "In-between," "Remote."
Additionally, we need long-term storage for prometheus metrics collected during the test.
Doc Requirements
- Background on why and when to use geo-distribution
- Description and listing multi-region cluster requirements
- Placeholder estimates on demands for scaling up to 2k users
- Diagram of infrastructure