I wonder if this may work better as a PR in libkrun first, which we can then expose here in krunkit after. There's a possible scenario where you create the pidfile, something fails in krun_start_enter, and there's that short period of time where there's a pidfile that exists even though the guest hasn't started yet and we haven't had a chance to remove it on krun_start_enter failure.

I wonder if this may work better as a PR in libkrun first, which we can then expose here in krunkit after. There's a possible scenario where you create the pidfile, something fails in krun_start_enter, and there's that short period of time where there's a pidfile that exists even though the guest hasn't started yet and we haven't had a chance to remove it on krun_start_enter failure.

This is very unlikely and the window is small. I think creating it before starting is good enough and it matches what vfkit does.

Programs running krunkit should check if the process terminated when they wait until the vm is reachable. I already do this in vment-helper test script, implementing this in minikube is easy.