I also have this problem, and I currently have not solved it. I thought it is a problem of gearmanmanager, thats my “daemon script” that starts many worker scripts and if one of them dies, it will be restarted.

gearmanmanager uses pcntl_fork to create child processes, and I think this is the root of all evil in this case. I have to check that again and see if it is so easy to reestablish a new connection the way to propose.