It doesn't seem that these items are configurable, except manually after launch:

• Labels. For example, browser tests with WebGL require g2 instances. Such instances would be labelled with "gpu", and the test job would be configured to run on "gpu" nodes. This would be set per-fleet.

• Usage. "Utilize this node as much as possible" vs "Only build jobs with label restrictions matching this node". This would be set per-fleet.

• Number of executors. Perhaps this could be based off the spot instance weight.