Managing Resources Using Control Groups

Control groups manage resource use through kernel resource controllers. A kernel resource controller represents a single resource, such as CPU time, memory, network bandwidth, or disk I/O.

To identify mounted resource controllers in the system, check the contents of the /procs/cgroups file, for example run:

less /proc/cgroups

#subsys_name    hierarchy       num_cgroups     enabled
cpuset  0       103     1
cpu     0       103     1
cpuacct 0       103     1
blkio   0       103     1
memory  0       103     1
devices 0       103     1
freezer 0       103     1
net_cls 0       103     1
perf_event      0       103     1
net_prio        0       103     1
hugetlb 0       103     1
pids    0       103     1
rdma    0       103     1
misc    0       103     1

For a detailed explanation of the kernel resource controllers of cgroups, see the cgroups(7) manual page.

cgroup functionality is mounted as a hierarchical file system in /sys/fs/cgroup.

The directory /sys/fs/cgroup is also called the root control group.

The root control group directory contents vary slightly depending on the mounted version, but on systems using cgroups v2 you typically see entries similar to the following:

ls /sys/fs/cgroup

cgroup.controllers      cpuset.mems.effective  memory.stat
cgroup.max.depth        cpu.stat               misc.capacity
cgroup.max.descendants  dev-hugepages.mount    sys-fs-fuse-connections.mount
cgroup.procs            dev-mqueue.mount       sys-kernel-config.mount
cgroup.stat             init.scope             sys-kernel-debug.mount
cgroup.subtree_control  io.pressure            sys-kernel-tracing.mount
cgroup.threads          io.stat                system.slice
cpu.pressure            memory.numa_stat       user.slice
cpuset.cpus.effective   memory.pressure

Each cgroup in the hierarchy contains the following files:

For a sample procedure that creates child groups where you can implement resource management for an application, see Setting CPU Weight to Regulate Distribution of CPU Time.

To remove a cgroup, ensure that the cgroup doesn't contain other child groups, and then remove the directory. For example, to remove child group /sys/fs/cgroup/MyGroups/cgroup1 you can run the following command:.

sudo rmdir /sys/fs/cgroup/MyGroups/cgroup1

The following distribution models provide you ways of implementing control or regulation in distributing resources for use by cgroups v2:

Weights

In this model, the weights of all the control groups are totaled. Each group receives a fraction of the resource based on the ratio of the group's weight against the total weight.

Consider 10 control groups, each with a weight of 100 for a combined total of 1000. In this case, each group can use a tenth of a specified resource.

Weight is typically used to distribute stateless resources. To apply this resource, the CPUWeight option is used.

Limits

In this model, a group can use up to the configured amount of a resource. If a resource such as memory usage for a process exceeds the limit, the kernel might stop the process with an out-of-memory (oom) message.

You can also overcommit resources so that the sum of the subgroups limits can exceed the limit of the parent group. Overcommitment assumes that resources in all subgroups aren't likely to all reach their limits at the same time.

To implement this distribution model, the MemoryMax option is often used.

Protections

In this model, a group is assigned a protected boundary. If the group's resource usage remains within the protected amount, the kernel can't deprive the group of the use of the resource in favor of other groups that are competing for the same resource. In this model, an overcommitment of resources is allowed.

To implement this model, the MemoryLow option is often used.

Allocations

In this model, a specific absolute amount is allocated for the use of finite type of resources, such as real-time budget.