From day 1 rkt's approach always seemed like the right approach to me intuitively. Their process is for many reasons much more professional. Docker on the other hand consistently overpromises and underwhelms.
I tried to get docker to work for a CI pipeline and it failed miserably. Consistent lock-ups and just generically slow operation when spinning up and deleting 100s of containers per hour forced me to move to lxc which works extremely well for a CI type workload. Not that surprising considering TravisCI and CircleCI behind the scenes are using lxc as well. Volume mounts, networking, sandboxing and just general stability all work without surprises in lxc but to this day I have no idea what the proper patterns are for sharing host level mounts properly inside docker containers.
I really hope rkt becomes the default container substrate. From first principles and fundamentals perspective rkt is the right solution.
The bad experiences we have had with docker have been from devicemapper bugs. Since devicemapper is part of the linux kernel I can't exactly blame docker. We really need a (limited) multi-host solution and have been using rancher.
I tried to get docker to work for a CI pipeline and it failed miserably. Consistent lock-ups and just generically slow operation when spinning up and deleting 100s of containers per hour forced me to move to lxc which works extremely well for a CI type workload. Not that surprising considering TravisCI and CircleCI behind the scenes are using lxc as well. Volume mounts, networking, sandboxing and just general stability all work without surprises in lxc but to this day I have no idea what the proper patterns are for sharing host level mounts properly inside docker containers.
I really hope rkt becomes the default container substrate. From first principles and fundamentals perspective rkt is the right solution.