question

No, the way our hosting works is that we launch the executable you've specified in the configuration. Only when that executable exits do we copy off the log and output files and mark that instance as "free" (and then, if there are no instances running on that host machine, and taking it away won't put you below the min slots setting, we shut down the server host).

AWS does have DDoS protections, but that actually wouldn't be relevant to your case unless you're doing something really unusual like flooding the server with HTTP calls. What you have is a server where each game instance is allocated a port. Clients connect, usually via TCP, to custom game servers on that port. The theoretical limit on connected players is extremely high, but in practical terms, you need to determine what your actual resource usage is like. There's a small amount of memory used per connected player (the TCP Control Block), but the actual traffic is then also a factor. What's best is to do some testing locally, so that you have direct access to the machine and can check everything about it - available memory, CPU usage, etc. - as you load on 1, 10, or 100 test users. My suspicion, if you're not seeing any errors in your log files and you're confident that you're logging all errors, is that you're probably running out of memory, but a quick local scale test should be able to confirm that.

For the instances exiting just after they start there isn't any more that we can add apart from what your logs show - all we do is run the executable provided and then capture the log and output files when it's done. I'm not seeing anything in the EC2 reports that would indicate any kind of systemic issue with the servers being allocated.