I've also run into the daemon crashing on production systems and taking down all of the other containers on that host. I end up just killing the host and resta

I've also run into the daemon crashing on production systems and taking down all of the other containers on that host. I end up just killing the host and restarting the jobs on a newly provisioned one. There's a lot of overhead. If they could get it to run without requiring a persistent (and root) shared daemon that would be a big improvement.