The unicorn workers are leaking memory. Any worker going above a limit will be restarted. Obviously it would be better to reduce the memory leaks further and we welcome help with that. To be clear, it is not a crash, all requests are handled without error to the end user.
