Create load generator for Coder Connect

[ibetitsmike]

From the plan:

Goal:

• To observe how Coder handles Coder Desktop usage at scale.

User target	Workspaces	Clients	Regions	coderd replicas	Provisioners
1k	300	165	3	3	60
2k	600	330	3	3	120
3k	900	495	3	4	180
10k	3000	1650	3	10	600

Infrastructure:

• Kubernetes cluster per region, each cluster has:
  • 1/3 of workspaces.
  • 1/3 of provisioners.

Implementation:

• Tools: scaletest create-workspaces + new custom Tailnet clients.
• Build concurrency equal to provisioners, to minimize build load.

Scenarios:

1. Half of the workspaces will be owned by users who own 10 workspaces. The other half will be owned by users with 1 workspace.
   • Most users will own one workspace, but we want to ensure those who own lots of workspaces don’t degrade the experience for others.

Run Sequence:

1. Create users.
2. All user clients connect to api/v2/tailnet.
3. Staggered start workspaces (via build concurrency = provisioners).
4. Wait until all updates have been received, or a reasonable timeout.
5. Stop all running workspaces.
6. Repeat step 3.

Key Results:

• Latency from build start to corresponding workspace update receipt.
• DB utilization/load during update fan out.
• Missed updates (each client expects minimum two updates).

Other results:

• Coder deployment + database resource utilization (incl. pprofs, Grafana Pyroscope).
• Workspace creation success rate.
• API response times.