Maybe you can track it down a little bit by logging 'everything'.

I'd start by letting tcpdump log everything on input, everything on output into a file.

Once the server dies, the final parts of the logs might help finding out what is wrong.
Even if there is some bug to be found - I'd suspect some bad request - being able to reproduce the issue at will would be a huge help for debugging.

Also, monitor the server's size while running. Maybe it runs out of memory if there are some sort of memory leaks...?

Simply looking on the code and find what might be wrong is really hard, most of the time.