Description of problem: I think i've come across an overall design flaw in the implementation of containers and we need to introduce the fourth/fith slice type called application.slice. The reason for this is make a clear distinction between OS containers ( machine.slice ) applications containers ( application.slice ) system containers ( system.slice ) and user containers ( user.slice ) and allow administrators to assign different resources to the application container and the system container via the higher level cgroup options. The overall design scheme would be changed to this /sys/fs/cgroup/ `-- systemd |-- machine.slice | `-- machine-f20.scope | |-- application.slice | | |--httpd.service | |-- system.slice | | |-- console-getty.service | | |-- systemd-journald.service | | | `-- user.slice |--application.slice | |-- ntpd.service |-- system.slice | |-- systemd-journald.service | |-- systemd-udevd.service | | `-- user.slice |-- user-1000.slice | |-- systemd-user | | |-- firefox.service Version-Release number of selected component (if applicable): How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: Expected results: Additional info:
You are asking for two different slices for the OS itself vs. stuff the admin actually installed on top of the OS? i.e. you want all systemd services to be in one slice and all the apache instances in another? Note that for multi-instantiated services there's already an implicit slice created where they are all placed in for each service. i.e. all apache instances are implicitly placed in systems.slice/apache.slice/apache...
I was thinking resource assignment basically ensure you could reserve X resources to the system itself something like. I got 4 cores ( or 40% cpu ) assigned to an OS container within that os container I want to give 20% to the application container which contains apache ssh etc 10% to the system container ( new ) and 10% to the user sessions. same applies with other resource deployment memory bandwith etc. If it dont make sense to you just close this.
This doesn't seem to be going anywhere. If anything, it should be discussed upstream.