CPU Manager for Kubernetes

Overview

This project provides basic core affinity for NFV-style workloads on top of vanilla Kubernetes v1.5+.

This project ships a single multi-use command-line program to perform various functions for host configuration, managing groups of CPUs, and constraining workloads to specific CPUs.

Concepts

Term	Meaning
Pool	A named group of CPU lists. A pool can be either exclusive or shared. In an exclusive pool, only one task may be allocated to each CPU list simultaneously.
CPU list	A group of logical CPUs, identified by ID as reported by the operating system. CPU lists conform to the Linux cpuset CPU list format.
Task list	A list of Linux process IDs.
Isolation	Steps required to set up a process environment so that it runs only on a desired subset of the available CPUs.
Reconciliation	The process of resolving state between the CMK configuration directory and the Linux procfs.

Usage summary

cmk.

Usage:
  cmk (-h | --help)
  cmk --version
  cmk cluster-init (--host-list=<list>|--all-hosts) [--cmk-cmd-list=<list>]
                   [--cmk-img=<img>] [--cmk-img-pol=<pol>] [--conf-dir=<dir>]
                   [--install-dir=<dir>] [--num-dp-cores=<num>]
                   [--num-cp-cores=<num>] [--pull-secret=<name>]
                   [--saname=<name>]
  cmk init [--conf-dir=<dir>] [--num-dp-cores=<num>] [--num-cp-cores=<num>]
  cmk discover [--conf-dir=<dir>]
  cmk describe [--conf-dir=<dir>]
  cmk reconcile [--conf-dir=<dir>] [--publish] [--interval=<seconds>]
  cmk isolate [--conf-dir=<dir>] --pool=<pool> <command> [-- <args> ...]
              [--no-affinity]
  cmk install [--install-dir=<dir>]
  cmk node-report [--conf-dir=<dir>] [--publish] [--interval=<seconds>]
  cmk uninstall [--install-dir=<dir>] [--conf-dir=<dir>]

Options:
  -h --help             Show this screen.
  --version             Show version.
  --host-list=<list>    Comma seperated list of Kubernetes nodes to prepare
                        for CMK software.
  --all-hosts           Prepare all Kubernetes nodes for the CMK software.
  --cmk-cmd-list=<list> Comma seperated list of CMK sub-commands to run on
                        each host
                        [default: init,reconcile,install,discover,nodereport].
  --cmk-img=<img>       CMK Docker image [default: cmk:v1.2.2].
  --cmk-img-pol=<pol>   Image pull policy for the CMK Docker image
                        [default: IfNotPresent].
  --conf-dir=<dir>      CMK configuration directory [default: /etc/cmk].
  --install-dir=<dir>   CMK install directory [default: /opt/bin].
  --interval=<seconds>  Number of seconds to wait between rerunning.
                        If set to 0, will only run once. [default: 0]
  --num-dp-cores=<num>  Number of data plane cores [default: 4].
  --num-cp-cores=<num>  Number of control plane cores [default: 1].
  --pool=<pool>         Pool name: either infra, controlplane or dataplane.
  --publish             Whether to publish reports to the Kubernetes
                        API server.
  --pull-secret=<name>  Name of secret used for pulling Docker images from
                        restricted Docker registry.
  --saname=<name>       ServiceAccount name to pass
                        [default: cmk-serviceaccount].
  --no-affinity         Do not set cpu affinity before forking the child
                        command. In this mode the user program is responsible
                        for reading the `CMK_CPUS_ASSIGNED` environment
                        variable and moving a subset of its own processes
                        and/or tasks to the assigned CPUs.

For detailed usage information about each subcommand, see Using the cmk command-line tool.

Further Reading

• Architecture
• Building cmk
• Operator manual
• User manual
• Using the cmk command-line tool
• The cmk configuration directory
• Releasing