Autoscaler Setup Procedure

1. Navigate to the Orchestration menu option on the left hand side of the ClearML landing page.
2. In the Orchestration view, click on “Autoscalers” at the top, and select “GCP Autoscaler”.
3. Right click on the running autoscaler instance (the one with a green play button symbol) that is called "GCP Autoscaler" or "GCP Autoscaler CPU Only" depending on whether you need a GPU or not, and select “Clone”.
4. In the config that opens up after selecting “Clone”, under each “Compute Resources” section, modify the “Resource Name” to include the name of the queue you’re pointing the resource to e.g. a2-highgpu-1g-production where a2-highgpu-1g is the resource and production is the queue name. Also in each “Compute Resources” section, change “Monitored Queue” to point to the desired queue. Finally, append the queue name to the Autoscaler Instance Name e.g. GCP Autoscaler Production.
5. Click “Launch” at the bottom of the config to launch the autoscaler.
6. When you no longer need the autoscaler, right click on the autoscaler and select “Stop”. But keep the original autoscaler instance that you cloned running because it is used for Serval E2E tests.