Awesome article and demo, but I had a question about limiting scaling. How would you approach limiting the number of agents that would be alive at any time?

For example someone queues up 30 builds you could be spinning up 30 agents and really bog down a cluster quickly. How would you limit that to 15 at any one time etc.?

I was thinking about having an agent-less step make a rest call to some sort of “agent queuer” (basically do what you’re doing with the init agent but inside a pod). I wasn’t sure if it was the right approach, or even to go the route of some batch job scheduler.

Would be great to hear your thoughts, as were in the middle of this exact project and your article was a great find.

🙏🏼