Yeah, I'm willing to live with a small race condition here. This is an internal tool and the frequency with which the tool is used means that the frequency at which two servers would run at the same time is very low, and thus the probability of the race condition become a concern is extremely low indeed.

Since this server is actually a c++ app and I have to communicate the port number by writing a new configuration file each time I don't want to muck with catching a failed startup, determine if it's because the port was in use, write a new config file and try again.