tasks

package
v0.750.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Aug 19, 2024 License: Apache-2.0 Imports: 31 Imported by: 2

Documentation

Index

Constants

View Source 
const (

	// SingularityEntrypointWrapperScript is just the name of the singularity entrypoint wrapper.
	SingularityEntrypointWrapperScript = "singularity-entrypoint-wrapper.sh"

	// StartupHookScript contains the script to run on task startup filled in dynamically.
	StartupHookScript = "dynamic-tcd-startup-hook.sh"
)
View Source 
const (
	// DefaultWorkDir is the default workdir.
	DefaultWorkDir = "/run/determined/workdir"

	RunDir = "/run/determined"

	PasswdPath = "/run/determined/etc/passwd"
	ShadowPath = "/run/determined/etc/shadow"
	GroupPath  = "/run/determined/etc/group"

	// DtrainSSHPortBase is starting range for Dtrain ports.
	DtrainSSHPortBase = 12350
	// InterTrainProcessCommPort1Base is starting range for intertraincomm1 ports.
	InterTrainProcessCommPort1Base = 12360
	// InterTrainProcessCommPort2Base is starting range for intertraincomm2 ports.
	InterTrainProcessCommPort2Base = 12365
	// C10DPortBase is starting range for c10D ports.
	C10DPortBase = 29400
	// DTrainSSHPort is the name of a port.
	DTrainSSHPort = "DTRAIN_SSH_PORT"
	// InterTrainProcessCommPort1 is the name of a port.
	InterTrainProcessCommPort1 = "INTER_TRAIN_PROCESS_COMM_PORT_1"
	// InterTrainProcessCommPort2 is the name of a port.
	InterTrainProcessCommPort2 = "INTER_TRAIN_PROCESS_COMM_PORT_2"
	// C10DPort is the name of a port.
	C10DPort = "C10D_PORT"
)

File location constants.

View Source 
const (

	// ManifestName is the name used by DAI RM when creating HPC job manifests.
	ManifestName = "det"
)

Variables

This section is empty.

Functions

func ToDockerMounts 

func ToDockerMounts(bindMounts []expconf.BindMount, workDir string) []mount.Mount

ToDockerMounts converts expconf bind mounts to container mounts.

func TrialSpecProxyPorts 

func TrialSpecProxyPorts(
	taskSpec *TaskSpec,
	expConfig expconf.ExperimentConfigV0,
) expconf.ProxyPortsConfig

TrialSpecProxyPorts combines user-defined and system proxy configs. This static function is public because trial actor builds `TrialSpec` instances late.

func ValidatePbs 

func ValidatePbs(pbsOptions []string) []error

ValidatePbs checks that the specified PBS options are allowed. If any are not messages are returned in an array of errors.

func ValidateSlurm 

func ValidateSlurm(slurmOptions []string) []error

ValidateSlurm checks that the specified slurm options are allowed. If any are not messages are returned in an array of errors.

Types

type GCCkptSpec 

type GCCkptSpec struct {
	Base TaskSpec

	ExperimentID int
	LegacyConfig expconf.LegacyConfig
	ToDelete     string
	// If len(CheckpointGlobs) == 0 then we won't delete any checkpoint files
	// and just refresh the state of the checkpoint.
	CheckpointGlobs    []string
	DeleteTensorboards bool
}

GCCkptSpec is a description of a task for running checkpoint GC.

func (GCCkptSpec) ToTaskSpec 

func (g GCCkptSpec) ToTaskSpec() TaskSpec

ToTaskSpec generates a TaskSpec.

type GenericCommandSpec 

type GenericCommandSpec struct {
	Base TaskSpec

	CommandID string

	Config model.CommandConfig

	// Deprecated: kept so we can still marshal to this.
	// Please use command.CreateGeneric.modelDef instead.
	UserFiles       archive.Archive
	AdditionalFiles archive.Archive

	Metadata genericCommandSpecMetadata

	Keys *ssh.PrivateAndPublicKeys

	WatchProxyIdleTimeout  bool
	WatchRunnerIdleTimeout bool

	TaskType model.TaskType
}

GenericCommandSpec is a description of a task for running a command.

func (*GenericCommandSpec) MakeEnvPorts 

func (s *GenericCommandSpec) MakeEnvPorts()

MakeEnvPorts fills in `Environment.Ports` i.e. exposed ports for container config.

func (*GenericCommandSpec) ProxyPorts 

func (s *GenericCommandSpec) ProxyPorts() expconf.ProxyPortsConfig

ProxyPorts combines user-defined and system proxy configs.

func (GenericCommandSpec) ToTaskSpec 

func (s GenericCommandSpec) ToTaskSpec() TaskSpec

ToTaskSpec generates a TaskSpec.

type GenericTaskSpec 

type GenericTaskSpec struct {
	Base           TaskSpec
	ProjectID      int
	WorkspaceID    int
	RegisteredTime time.Time
	JobID          model.JobID

	GenericTaskConfig model.GenericTaskConfig
}

GenericTaskSpec is the generic task spec.

func (GenericTaskSpec) ResourcePool 

func (s GenericTaskSpec) ResourcePool() string

ResourcePool - returns resource pool.

func (GenericTaskSpec) SetJobPriority 

func (s GenericTaskSpec) SetJobPriority(priority int) error

SetJobPriority todo.

func (GenericTaskSpec) SetResourcePool 

func (s GenericTaskSpec) SetResourcePool(resourcePool string) error

SetResourcePool todo.

func (GenericTaskSpec) SetWeight 

func (s GenericTaskSpec) SetWeight(weight float64) error

SetWeight todo.

func (GenericTaskSpec) ToTaskSpec 

func (s GenericTaskSpec) ToTaskSpec() TaskSpec

ToTaskSpec converts the generic task spec to the common task spec.

func (GenericTaskSpec) ToV1Job 

func (s GenericTaskSpec) ToV1Job() (*jobv1.Job, error)

ToV1Job todo.

type TaskSpec 

type TaskSpec struct {
	// Fields that are only for task logics.
	Description string
	// LoggingFields are fields to include in each record of structured logging.
	LoggingFields map[string]string
	// LogRetentionDays is the number of days to retain logs for.
	LogRetentionDays *int16

	// Fields that are set on the cluster level.
	ClusterID   string
	HarnessPath string
	MasterCert  []byte
	SSHRsaSize  int

	SegmentEnabled bool
	SegmentAPIKey  string

	// Fields that are set on the per-request basis.
	// TaskContainerDefaults should be removed from TaskSpec once we move to using the same
	// schema for the cluster-level defaults and the request-level configuration.
	TaskContainerDefaults model.TaskContainerDefaultsConfig
	Environment           expconf.EnvironmentConfig
	ResourcesConfig       expconf.ResourcesConfig
	WorkDir               string
	Owner                 *model.User
	AgentUserGroup        *model.AgentUserGroup
	ExtraArchives         []cproto.RunArchive
	ExtraEnvVars          map[string]string
	ExtraPodLabels        map[string]string
	Entrypoint            []string
	Mounts                []mount.Mount
	// UseHostMode is whether host mode networking would be desirable for this task.
	// This is used by Docker only.
	UseHostMode bool
	ShmSize     int64

	// The parent task of an allocation.
	TaskID string

	// Fields that are set on per-resources basis.
	AllocationID           string
	AllocationSessionToken string
	ResourcesID            string
	ContainerID            string
	Devices                []device.Device

	UserSessionToken string
	TaskType         model.TaskType
	SlurmConfig      expconf.SlurmConfig
	PbsConfig        expconf.PbsConfig

	ExtraProxyPorts expconf.ProxyPortsConfig

	Workspace string
	Project   string
	Labels    []string
	// Ports required by trial or commands and their respective base port values.
	UniqueExposedPortRequests map[string]int

	// For testing only.
	DontShipLogs bool
}

TaskSpec defines the spec of a task.

func (*TaskSpec) Archives 

func (t *TaskSpec) Archives() ([]cproto.RunArchive, []cproto.RunArchive)

Archives returns all the archives.

func (*TaskSpec) Clone 

func (t *TaskSpec) Clone() (*TaskSpec, error)

Clone deep copies a taskSpec.

func (TaskSpec) EnvVars 

func (t TaskSpec) EnvVars() map[string]string

EnvVars returns all the environment variables.

func (*TaskSpec) LogShipperWrappedEntrypoint 

func (t *TaskSpec) LogShipperWrappedEntrypoint() []string

LogShipperWrappedEntrypoint returns the configured Entrypoint wrapped with ship_logs.py.

func (*TaskSpec) ResolveWorkDir 

func (t *TaskSpec) ResolveWorkDir()

ResolveWorkDir resolves the work dir.

func (*TaskSpec) ToDispatcherManifest 

func (t *TaskSpec) ToDispatcherManifest(
	syslog *logrus.Entry,
	allocationID string,
	tlsEnabled bool,
	masterHost string,
	masterPort int,
	certificateName string,
	numSlots int,
	slotType device.Type,
	slurmPartition string,
	tresSupported bool,
	gresSupported bool,
	containerRunType string,
	isPbsLauncher bool,
	labelMode *string,
	disabledNodes []string,
) (*launcher.Manifest, string, string, error)

ToDispatcherManifest creates the manifest that will be ultimately sent to the launcher. Returns: 

Manifest, launchingUserName, PayloadName, err

Note: Cannot pass "req *sproto.AllocateRequest" as an argument, as it requires import of "github.com/determined-ai/determined/master/internal/sproto", which results in an "import cycle not allowed" error.

func (*TaskSpec) ToDockerSpec 

func (t *TaskSpec) ToDockerSpec() cproto.Spec

ToDockerSpec converts a task spec to a docker container spec.

func (*TaskSpec) WarnUnsupportedOptions 

func (t *TaskSpec) WarnUnsupportedOptions(
	userConfiguredPriority bool,
	containerRunType string,
) string

WarnUnsupportedOptions gives warnings for user configurations that are not supported by HPC launcher.

type TaskSpecifier 

type TaskSpecifier interface {
	ToTaskSpec() TaskSpec
}

TaskSpecifier creates a TaskSpec. ToTaskSpec must only be called once per specifier.

type TrialSpec 

type TrialSpec struct {
	Base TaskSpec

	ExperimentID     int
	TrialID          int
	TrialRunID       int
	ExperimentConfig expconf.ExperimentConfig
	HParams          map[string]interface{}
	TrialSeed        uint32
	LatestCheckpoint *model.Checkpoint
	StepsCompleted   int

	Keys ssh.PrivateAndPublicKeys
}

TrialSpec is a description of a task for running a trial container.

func (*TrialSpec) MakeEnvPorts 

func (s *TrialSpec) MakeEnvPorts() expconf.EnvironmentConfigV0

MakeEnvPorts fills in `Environment.Ports` i.e. exposed ports for container config.

func (*TrialSpec) ProxyPorts 

func (s *TrialSpec) ProxyPorts() expconf.ProxyPortsConfig

ProxyPorts combines user-defined and system proxy configs.

func (TrialSpec) ToTaskSpec 

func (s TrialSpec) ToTaskSpec() TaskSpec

ToTaskSpec generates a TaskSpec.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.