[libvirt] [PATCH 00/18] Re-arrange the way cgroups are setup

[Daniel P. Berrange]

This is a greatly expanded version of a previous series I posted

  https://www.redhat.com/archives/libvir-list/2013-March/msg01373.html

Currently libvirt creates a cgroups hiearchy at 

   $LOCATION-OF-LIBVIRTD/libvirt/{qemu,lxc}/$GUEST-NAME

eg

/sys/fs/cgroup
├── blkio
│   └── libvirt
│       ├── lxc
│       │   └── busy
│       └── qemu
│           └── vm1
├── cpu,cpuacct
│   ├── libvirt
│   │   ├── lxc
│   │   │   └── busy
│   │   └── qemu
│   │       └── vm1
│   │           ├── emulator
│   │           └── vcpu0
│   └── system
│       ├── abrtd.service
│       ....snip....
│       └── upower.service
├── cpuset
│   └── libvirt
│       ├── lxc
│       │   └── busy
│       └── qemu
│           └── vm1
│               ├── emulator
│               └── vcpu0
├── devices
│   └── libvirt
│       ├── lxc
│       │   └── busy
│       └── qemu
│           └── vm1
├── freezer
│   └── libvirt
│       ├── lxc
│       │   └── busy
│       └── qemu
│           └── vm1
├── memory
│   └── libvirt
│       ├── lxc
│       │   └── busy
│       └── qemu
│           └── vm1
├── net_cls
├── perf_event


This series changes it so that libvirt creates cgroups at

     /system/$VMNAME.{qemu,lxc}.libvirt

and allows configuration of the "resource partition" (ie the
"/system" bit) via the XML. So we get a layout like this:

/sys/fs/cgroup
├── blkio
│   └── system
│       ├── demo.lxc.libvirt
│       └── vm1.qemu.libvirt
├── cpu,cpuacct
│   └── system
│       ├── abrtd.service
│       ....snip....
│       ├── demo.lxc.libvirt
│       ....snip....
│       └── vm1.qemu.libvirt
│           ├── emulator
│           └── vcpu0
├── cpuset
│   └── system
│       ├── demo.lxc.libvirt
│       └── vm1.qemu.libvirt
│           ├── emulator
│           └── vcpu0
├── devices
│   └── system
│       ├── demo.lxc.libvirt
│       └── vm1.qemu.libvirt
├── freezer
│   └── system
│       ├── demo.lxc.libvirt
│       └── vm1.qemu.libvirt
├── memory
│   └── system
│       ├── demo.lxc.libvirt
│       └── vm1.qemu.libvirt
├── net_cls
├── perf_event


Flattening out the libvirt created hiearchy has serious
performance wins, due to poor kernel scalability with
deep hierarchies. It also makes it easier to configure
system wide policy for resource usage across system
services and virtual machines / containers, since they
all live at the top level in comon resource partitions.