cgutil

Documentation

Index

• Constants
• Variables
• func CgroupScope(allocID, task string) string
• func ConfigureBasicCgroups(config *lcc.Config) error
• func CopyCpuset(source, destination string) error
• func FindCgroupMountpointDir() (string, error)
• func GetCPUsFromCgroup(group string) ([]uint16, error)
• func GetCgroupParent(parent string) string
• func GetCgroupPathHelperV1(subsystem, cgroup string) (string, error)
• func SplitPath(p string) (string, string)
• type CgroupPathGetter
• type CpusetManager
  • func CreateCPUSetManager(parent string, logger hclog.Logger) CpusetManager
  • func NewCpusetManagerV1(cgroupParent string, logger hclog.Logger) CpusetManager
  • func NewCpusetManagerV2(parent string, logger hclog.Logger) CpusetManager
• type GroupKiller
  • func NewGroupKiller(logger hclog.Logger, pid int) GroupKiller
• type NoopCpusetManager
  • func (n NoopCpusetManager) AddAlloc(alloc *structs.Allocation)
  • func (n NoopCpusetManager) CgroupPathFor(allocID, task string) CgroupPathGetter
  • func (n NoopCpusetManager) Init([]uint16) error
  • func (n NoopCpusetManager) RemoveAlloc(allocID string)
• type TaskCgroupInfo

Constants

const (
	DefaultCgroupV1Parent    = "/nomad"
	SharedCpusetCgroupName   = "shared"
	ReservedCpusetCgroupName = "reserved"
)

const (
	// CreationPID is a special PID in libcontainer used to denote a cgroup
	// should be created, but with no process added.
	//
	// https://github.com/opencontainers/runc/blob/v1.0.3/libcontainer/cgroups/utils.go#L372
	CreationPID = -1

	// DefaultCgroupParentV2 is the name of Nomad's default parent cgroup, under which
	// all other cgroups are managed. This can be changed with client configuration
	// in case for e.g. Nomad tasks should be further constrained by an externally
	// configured systemd cgroup.
	DefaultCgroupParentV2 = "nomad.slice"
)

const (
	// CgroupRoot is hard-coded in the cgroups specification.
	// It only applies to linux but helpers have references to it in driver(s).
	CgroupRoot = "/sys/fs/cgroup"
)

Variables

var UseV2 = cgroups.IsCgroup2UnifiedMode()

UseV2 indicates whether only cgroups.v2 is enabled. If cgroups.v2 is not enabled or is running in hybrid mode with cgroups.v1, Nomad will make use of cgroups.v1

This is a read-only value.

Functions

func CgroupScope

func CgroupScope(allocID, task string) string

CgroupScope returns the name of the scope for Nomad's managed cgroups for the given allocID and task.

e.g. "<allocID>.<task>.scope"

Only useful for v2.

func ConfigureBasicCgroups

func ConfigureBasicCgroups(config *lcc.Config) error

ConfigureBasicCgroups will initialize a cgroup and modify config to contain a reference to its path.

v1: creates a random "freezer" cgroup which can later be used for cleanup of processes. v2: does nothing.

func CopyCpuset

func CopyCpuset(source, destination string) error

CopyCpuset copies the cpuset.cpus value from source into destination.

func FindCgroupMountpointDir

func FindCgroupMountpointDir() (string, error)

FindCgroupMountpointDir is used to find the cgroup mount point on a Linux system.

Note that in cgroups.v1, this returns one of many subsystems that are mounted. e.g. a return value of "/sys/fs/cgroup/systemd" really implies the root is "/sys/fs/cgroup", which is interesting on hybrid systems where the 'unified' subsystem is mounted as if it were a subsystem, but the actual root is different. (i.e. /sys/fs/cgroup/unified).

As far as Nomad is concerned, UseV2 is the source of truth for which hierarchy to use, and that will only be a true value if cgroups.v2 is mounted on /sys/fs/cgroup (i.e. system is not in v1 or hybrid mode).

➜ mount -l | grep cgroup tmpfs on /sys/fs/cgroup type tmpfs (ro,nosuid,nodev,noexec,mode=755,inode64) cgroup2 on /sys/fs/cgroup/unified type cgroup2 (rw,nosuid,nodev,noexec,relatime,nsdelegate) cgroup on /sys/fs/cgroup/systemd type cgroup (rw,nosuid,nodev,noexec,relatime,xattr,name=systemd) cgroup on /sys/fs/cgroup/memory type cgroup (rw,nosuid,nodev,noexec,relatime,memory) (etc.)

func GetCPUsFromCgroup

func GetCPUsFromCgroup(group string) ([]uint16, error)

GetCPUsFromCgroup gets the effective cpuset value for the given cgroup.

func GetCgroupParent

func GetCgroupParent(parent string) string

GetCgroupParent returns the mount point under the root cgroup in which Nomad will create cgroups. If parent is not set, an appropriate name for the version of cgroups will be used.

func GetCgroupPathHelperV1

func GetCgroupPathHelperV1(subsystem, cgroup string) (string, error)

func SplitPath

func SplitPath(p string) (string, string)

SplitPath determines the parent and cgroup from p. p must contain at least 2 elements (parent + cgroup).

Handles the cgroup root if present.

Types

type CgroupPathGetter

type CgroupPathGetter func(context.Context) (path string, err error)

CgroupPathGetter is a function which returns the cgroup path and any error which occurred during cgroup initialization.

It should block until the cgroup has been created or an error is reported.

type CpusetManager

type CpusetManager interface {
	// Init should be called with the initial set of reservable cores before any
	// allocations are managed. Ensures the parent cgroup exists and proper permissions
	// are available for managing cgroups.
	Init([]uint16) error

	// AddAlloc adds an allocation to the manager
	AddAlloc(alloc *structs.Allocation)

	// RemoveAlloc removes an alloc by ID from the manager
	RemoveAlloc(allocID string)

	// CgroupPathFor returns a callback for getting the cgroup path and any error that may have occurred during
	// cgroup initialization. The callback will block if the cgroup has not been created
	CgroupPathFor(allocID, taskName string) CgroupPathGetter
}

CpusetManager is used to setup cpuset cgroups for each task.

func CreateCPUSetManager

func CreateCPUSetManager(parent string, logger hclog.Logger) CpusetManager

CreateCPUSetManager creates a V1 or V2 CpusetManager depending on system configuration.

func NewCpusetManagerV1

func NewCpusetManagerV1(cgroupParent string, logger hclog.Logger) CpusetManager

NewCpusetManagerV1 creates a CpusetManager compatible with cgroups.v1

func NewCpusetManagerV2

func NewCpusetManagerV2(parent string, logger hclog.Logger) CpusetManager

type GroupKiller

type GroupKiller interface {
	KillGroup(cgroup *configs.Cgroup) error
}

GroupKiller is used for SIGKILL-ing the process tree[s] of a cgroup by leveraging the freezer cgroup subsystem.

func NewGroupKiller

func NewGroupKiller(logger hclog.Logger, pid int) GroupKiller

NewGroupKiller creates a GroupKiller with executor PID pid.

type NoopCpusetManager

type NoopCpusetManager struct{}

func (NoopCpusetManager) AddAlloc

func (n NoopCpusetManager) AddAlloc(alloc *structs.Allocation)

func (NoopCpusetManager) CgroupPathFor

func (n NoopCpusetManager) CgroupPathFor(allocID, task string) CgroupPathGetter

func (NoopCpusetManager) Init

func (n NoopCpusetManager) Init([]uint16) error

func (NoopCpusetManager) RemoveAlloc

func (n NoopCpusetManager) RemoveAlloc(allocID string)

type TaskCgroupInfo

type TaskCgroupInfo struct {
	CgroupPath         string
	RelativeCgroupPath string
	Cpuset             cpuset.CPUSet
	Error              error
}