There are some cases where a Modal web endpoint might not spin up and ACK in less than 5 seconds. To avoid that, you can put keep_warm on the stub.function so you don't hit cold boots, but then you lose autoscaling. You can also reduce the chance of by putting less work in the path of the request -- e.g. defining a separate stub.function to handle a run and using Function.spawn to launch it as a background task, which means you can get to the ACK faster.

I'll include that in my example, so no need to put it in here, just wanted to raise the issue.