Documentation
¶
Index ¶
- Constants
- Variables
- func Serve(dev DevicePlugin, logger log.Logger)
- type ContainerReservation
- type Device
- type DeviceGroup
- type DeviceGroupStats
- type DeviceLocality
- type DevicePlugin
- type DeviceSpec
- type DeviceStats
- type FingerprintFn
- type FingerprintResponse
- type MockDevicePlugin
- type Mount
- type PluginDevice
- type ReserveFn
- type StatsFn
- type StatsResponse
Constants ¶
const (
// ApiVersion010 is the initial API version for the device plugins
ApiVersion010 = "v0.1.0"
)
const (
// DeviceTypeGPU is a canonical device type for a GPU.
DeviceTypeGPU = "gpu"
)
Variables ¶
var ( // ErrPluginDisabled indicates that the device plugin is disabled ErrPluginDisabled = fmt.Errorf("device is not enabled") )
Functions ¶
Types ¶
type ContainerReservation ¶
type ContainerReservation struct {
// Envs are a set of environment variables to set for the task.
Envs map[string]string
// Mounts are used to mount host volumes into a container that may include
// libraries, etc.
Mounts []*Mount
// Devices are the set of devices to mount into the container.
Devices []*DeviceSpec
}
ContainerReservation describes how to mount a device into a container. A container is an isolated environment that shares the host's OS.
type Device ¶
type Device struct {
// ID is the identifier for the device.
ID string
// Healthy marks whether the device is healthy and can be used for
// scheduling.
Healthy bool
// HealthDesc describes why the device may be unhealthy.
HealthDesc string
// HwLocality captures hardware locality information for the device.
HwLocality *DeviceLocality
}
Device is an instance of a particular device.
type DeviceGroup ¶
type DeviceGroup struct {
// Vendor is the vendor providing the device (nvidia, intel, etc).
Vendor string
// Type is the type of the device (gpu, fpga, etc).
Type string
// Name is the devices model name.
Name string
// Devices is the set of device instances.
Devices []*Device
// Attributes are a set of attributes shared for all the devices.
Attributes map[string]*structs.Attribute
}
DeviceGroup is a grouping of devices that share a common vendor, device type and name.
func (*DeviceGroup) Validate ¶
func (d *DeviceGroup) Validate() error
Validate validates that the device group is valid
type DeviceGroupStats ¶
type DeviceGroupStats struct {
Vendor string
Type string
Name string
// InstanceStats is a mapping of each device ID to its statistics.
InstanceStats map[string]*DeviceStats
}
DeviceGroupStats contains statistics for each device of a particular device group, identified by the vendor, type and name of the device.
type DeviceLocality ¶
type DeviceLocality struct {
// PciBusID is the PCI bus ID of the device.
PciBusID string
}
DeviceLocality captures hardware locality information for a device.
type DevicePlugin ¶
type DevicePlugin interface {
base.BasePlugin
// Fingerprint returns a stream of devices that are detected.
Fingerprint(ctx context.Context) (<-chan *FingerprintResponse, error)
// Reserve is used to reserve a set of devices and retrieve mount
// instructions.
Reserve(deviceIDs []string) (*ContainerReservation, error)
// Stats returns a stream of statistics per device collected at the passed
// interval.
Stats(ctx context.Context, interval time.Duration) (<-chan *StatsResponse, error)
}
DevicePlugin is the interface for a plugin that can expose detected devices to Nomad and inform it how to mount them.
type DeviceSpec ¶
type DeviceSpec struct {
// TaskPath is the location to mount the device in the task's file system.
TaskPath string
// HostPath is the host location of the device.
HostPath string
// CgroupPerms defines the permissions to use when mounting the device.
CgroupPerms string
}
DeviceSpec captures how to mount a device into a container.
type DeviceStats ¶
type DeviceStats struct {
// Summary exposes a single summary metric that should be the most
// informative to users.
Summary *structs.StatValue
// Stats contains the verbose statistics for the device.
Stats *structs.StatObject
// Timestamp is the time the statistics were collected.
Timestamp time.Time
}
DeviceStats is the statistics for an individual device
type FingerprintFn ¶
type FingerprintFn func(context.Context) (<-chan *FingerprintResponse, error)
func ErrorChFingerprinter ¶
func ErrorChFingerprinter(err error) FingerprintFn
ErrorChFingerprinter returns an error fingerprinting over the channel
func StaticFingerprinter ¶
func StaticFingerprinter(devices []*DeviceGroup) FingerprintFn
StaticFingerprinter fingerprints the passed devices just once
type FingerprintResponse ¶
type FingerprintResponse struct {
// Devices is a set of devices that have been detected.
Devices []*DeviceGroup
// Error is populated when fingerprinting has failed.
Error error
}
FingerprintResponse includes a set of detected devices or an error in the process of fingerprinting.
func NewFingerprint ¶
func NewFingerprint(devices ...*DeviceGroup) *FingerprintResponse
NewFingerprint takes a set of device groups and returns a fingerprint response
func NewFingerprintError ¶
func NewFingerprintError(err error) *FingerprintResponse
NewFingerprintError takes an error and returns a fingerprint response
type MockDevicePlugin ¶
type MockDevicePlugin struct {
*base.MockPlugin
FingerprintF FingerprintFn
ReserveF ReserveFn
StatsF StatsFn
}
MockDevicePlugin is used for testing. Each function can be set as a closure to make assertions about how data is passed through the base plugin layer.
func (*MockDevicePlugin) Fingerprint ¶
func (p *MockDevicePlugin) Fingerprint(ctx context.Context) (<-chan *FingerprintResponse, error)
func (*MockDevicePlugin) Reserve ¶
func (p *MockDevicePlugin) Reserve(devices []string) (*ContainerReservation, error)
func (*MockDevicePlugin) Stats ¶
func (p *MockDevicePlugin) Stats(ctx context.Context, interval time.Duration) (<-chan *StatsResponse, error)
type Mount ¶
type Mount struct {
// TaskPath is the location in the task's file system to mount.
TaskPath string
// HostPath is the host directory path to mount.
HostPath string
// ReadOnly defines whether the mount should be read only to the task.
ReadOnly bool
}
Mount is used to mount a host directory into a container.
type PluginDevice ¶
type PluginDevice struct {
plugin.NetRPCUnsupportedPlugin
Impl DevicePlugin
}
PluginDevice is wraps a DevicePlugin and implements go-plugins GRPCPlugin interface to expose the interface over gRPC.
func (*PluginDevice) GRPCClient ¶
func (p *PluginDevice) GRPCClient(ctx context.Context, broker *plugin.GRPCBroker, c *grpc.ClientConn) (interface{}, error)
func (*PluginDevice) GRPCServer ¶
func (p *PluginDevice) GRPCServer(broker *plugin.GRPCBroker, s *grpc.Server) error
type ReserveFn ¶
type ReserveFn func([]string) (*ContainerReservation, error)
func StaticReserve ¶
func StaticReserve(out *ContainerReservation) ReserveFn
StaticReserve returns the passed container reservation
type StatsFn ¶
func ErrorChStats ¶
ErrorChStats returns an error collecting stats over the channel
func StaticStats ¶
func StaticStats(out []*DeviceGroupStats) StatsFn
StaticStats returns the passed statistics
type StatsResponse ¶
type StatsResponse struct {
// Groups contains statistics for each device group.
Groups []*DeviceGroupStats
// Error is populated when collecting statistics has failed.
Error error
}
StatsResponse returns statistics for each device group.
func NewStatsError ¶
func NewStatsError(err error) *StatsResponse
NewStatsError takes an error and returns a stats response