In a single workflow, the scale tests runner maintains a consistent load

distribution as follows:

- 80% of users open and utilize SSH connections.

- 25% of users connect to the workspace using the Web Terminal.

- 40% of users simulate traffic for workspace apps.

- 20% of users interact with the Coder UI via a headless browser.

This distribution closely mirrors natural user behavior, as observed among our

customers.

The basic setup of scale tests environment involves:

1. Scale tests runner: `c2d-standard-32` (32 vCPU, 128 GB RAM)

2. Coderd: 2 replicas (4 vCPU, 16 GB RAM)

3. Database: 1 replica (2 vCPU, 32 GB RAM)

4. Provisioner: 50 instances (0.5 vCPU, 512 MB RAM)

No pod restarts or internal errors were observed.

In our scale tests, we simulate activity from 2000 users, 2000 workspaces, and

2000 agents, with metadata being sent 2 x every 10 s. Here are the resulting

metrics:

Coder:

- coderd: Median CPU usage for coderd: 3 vCPU, peaking at 3.7 vCPU during

dashboard tests.

- provisionerd: Median CPU usage is 0.35 vCPU during workspace provisioning.

API:

- Median request rate: 350 req/s during dashboard tests, 250 req/s during Web

Terminal and workspace apps tests.

- 2000 agent connections with latency: p90 at 60 ms, p95 at 220 ms.

- on average 2400 websocket connections during dashboard tests.

Database:

- Median CPU utilization: 80%.

- Median memory utilization: 40%.

- Average write_ops_count per minute: between 400 and 500 operations.