configuration

Documentation

Index

• Constants
• func SchedulingConfigValidation(sl validator.StructLevel)
• type Configuration
  • func (c Configuration) Validate() error
• type FailureEstimatorConfig
• type HistogramConfig
• type HttpConfig
• type LeaderConfig
• type LegacyMetricsConfig
• type MetricsConfig
• type NodeQuarantinerConfig
• type PrometheusSummaryConfig
• type QueueQuarantinerConfig
• type ResourceType
• type SchedulerMetricsConfig
• type SchedulingConfig
• type WellKnownNodeType

Constants

const (
	DuplicateWellKnownNodeTypeErrorMessage     = "duplicate well-known node type name"
	AwayNodeTypesWithoutPreemptionErrorMessage = "priority class has away node types but is not preemptible"
	UnknownWellKnownNodeTypeErrorMessage       = "priority class refers to unknown well-known node type"
)

const (
	// NodeIdLabel maps to a unique id associated with each node.
	// This label is automatically added to nodes within the NodeDb.
	NodeIdLabel = "armadaproject.io/nodeId"
)

Functions

func SchedulingConfigValidation

func SchedulingConfigValidation(sl validator.StructLevel)

Types

type Configuration

type Configuration struct {
	// Database configuration
	Postgres configuration.PostgresConfig
	// Redis Config
	Redis config.RedisConfig
	// Armada Api Connection.  Used to fetch queues.
	ArmadaApi client.ApiConnectionDetails
	// General Pulsar configuration
	Pulsar configuration.PulsarConfig
	// Configuration controlling leader election
	Leader LeaderConfig
	// Configuration controlling metrics
	Metrics LegacyMetricsConfig
	// Configuration for new scheduler metrics.
	// Due to replace metrics configured via the above entry.
	SchedulerMetrics MetricsConfig
	// Scheduler configuration (this is shared with the old scheduler)
	Scheduling SchedulingConfig
	Auth       authconfig.AuthConfig
	Grpc       grpcconfig.GrpcConfig
	Http       HttpConfig
	// If non-nil, net/http/pprof endpoints are exposed on localhost on this port.
	PprofPort *uint16
	// Maximum number of strings that should be cached at any one time
	InternedStringsCacheSize uint32 `validate:"required"`
	// How often the scheduling cycle should run
	CyclePeriod time.Duration `validate:"required"`
	// How often the job scheduling should run
	// This is expected to be a greater value than CyclePeriod as we don't need to schedule every cycle
	// This keeps the system more responsive as other operations happen in each cycle - such as state changes
	SchedulePeriod time.Duration `validate:"required"`
	// The maximum time allowed for a job scheduling round
	MaxSchedulingDuration time.Duration `validate:"required"`
	// How long after a heartbeat an executor will be considered lost
	ExecutorTimeout time.Duration `validate:"required"`
	// Maximum number of rows to fetch in a given query
	DatabaseFetchSize int `validate:"required"`
	// Timeout to use when sending messages to pulsar
	PulsarSendTimeout time.Duration `validate:"required"`
	// Frequency at which queues will be fetched from the API
	QueueRefreshPeriod time.Duration `validate:"required"`
	// If true then submit checks will be skipped
	DisableSubmitCheck bool
}

func (Configuration) Validate

func (c Configuration) Validate() error

type FailureEstimatorConfig

type FailureEstimatorConfig struct {
	Disabled                           bool
	NumInnerIterations                 int     `validate:"gt=0"`
	InnerOptimiserStepSize             float64 `validate:"gt=0"`
	OuterOptimiserStepSize             float64 `validate:"gt=0"`
	OuterOptimiserNesterovAcceleration float64 `validate:"gte=0"`
}

FailureEstimatorConfig controls node and queue success probability estimation. See internal/scheduler/failureestimator.go for details.

type HistogramConfig

type HistogramConfig struct {
	Start  float64
	Factor float64
	Count  int
}

type HttpConfig

type HttpConfig struct {
	Port int `validate:"required"`
}

type LeaderConfig

type LeaderConfig struct {
	// Valid modes are "standalone" or "kubernetes"
	Mode string `validate:"required"`
	// Name of the K8s Lock Object
	LeaseLockName string
	// Namespace of the K8s Lock Object
	LeaseLockNamespace string
	// The name of the pod
	PodName string
	// How long the lease is held for.
	// Non leaders much wait this long before trying to acquire the lease
	LeaseDuration time.Duration
	// RenewDeadline is the duration that the acting leader will retry refreshing leadership before giving up.
	RenewDeadline time.Duration
	// RetryPeriod is the duration the LeaderElector clients should waite between tries of actions.
	RetryPeriod time.Duration
	// Connection details to the leader
	LeaderConnection client.ApiConnectionDetails
}

type LegacyMetricsConfig

type LegacyMetricsConfig struct {
	Port            uint16
	RefreshInterval time.Duration
	Metrics         SchedulerMetricsConfig
}

TODO: ALl this needs to be unified with MetricsConfig

type MetricsConfig

type MetricsConfig struct {
	// If true, disable metric collection and publishing.
	Disabled bool
	// Regexes used for job error categorisation.
	// Specifically, the subCategory label for job failure counters is the first regex that matches the job error.
	// If no regex matches, the subCategory label is the empty string.
	TrackedErrorRegexes []string
	// Metrics are exported for these resources.
	TrackedResourceNames []v1.ResourceName
	// Optionally rename resources in exported metrics.
	// E.g., if ResourceRenaming["nvidia.com/gpu"] = "gpu", then metrics for resource "nvidia.com/gpu" use resource name "gpu" instead.
	// This can be used to avoid illegal Prometheus metric names (e.g., for "nvidia.com/gpu" as "/" is not allowed).
	// Allowed characters in resource names are [a-zA-Z_:][a-zA-Z0-9_:]*
	// It can also be used to track multiple resources within the same metric, e.g., "nvidia.com/gpu" and "amd.com/gpu".
	ResourceRenaming map[v1.ResourceName]string
	// Controls the cycle time metrics.
	// TODO(albin): Not used yet.
	CycleTimeConfig PrometheusSummaryConfig
	// The first matching regex of each error message is cached in an LRU cache.
	// This setting controls the cache size.
	MatchedRegexIndexByErrorMessageCacheSize uint64
	// Reset metrics this often. Resetting periodically ensures inactive time series are garbage-collected.
	ResetInterval time.Duration
}

type NodeQuarantinerConfig

type NodeQuarantinerConfig struct {
	FailureProbabilityQuarantineThreshold float64       `validate:"gte=0,lte=1"`
	FailureProbabilityEstimateTimeout     time.Duration `validate:"gte=0"`
}

NodeQuarantinerConfig controls how nodes are quarantined, i.e., removed from consideration when scheduling new jobs. See internal/scheduler/quarantine/node_quarantiner.go for details.

type PrometheusSummaryConfig

type PrometheusSummaryConfig struct {
	// Objectives defines the quantile rank estimates with their respective
	// absolute error. If Objectives[q] = e, then the value reported for q
	// will be the φ-quantile value for some φ between q-e and q+e.  The
	// default value is an empty map, resulting in a summary without
	// quantiles.
	Objectives map[float64]float64

	// MaxAge defines the duration for which an observation stays relevant
	// for the summary. Only applies to pre-calculated quantiles, does not
	// apply to _sum and _count. Must be positive. The default value is
	// DefMaxAge.
	MaxAge time.Duration
}

PrometheusSummaryConfig contains the relevant config for a prometheus.Summary.

type QueueQuarantinerConfig

type QueueQuarantinerConfig struct {
	QuarantineFactorMultiplier        float64       `validate:"gte=0,lte=1"`
	FailureProbabilityEstimateTimeout time.Duration `validate:"gte=0"`
}

QueueQuarantinerConfig controls how scheduling from misbehaving queues is rate-limited. See internal/scheduler/quarantine/queue_quarantiner.go for details.

type ResourceType

type ResourceType struct {
	// Resource name, e.g., "cpu", "memory", or "nvidia.com/gpu".
	Name string
	// Resolution with which Armada tracks this resource; larger values indicate lower resolution.
	Resolution resource.Quantity
}

ResourceType represents a resource the scheduler indexes for efficient lookup.

type SchedulerMetricsConfig

type SchedulerMetricsConfig struct {
	ScheduleCycleTimeHistogramSettings  HistogramConfig
	ReconcileCycleTimeHistogramSettings HistogramConfig
}

type SchedulingConfig

type SchedulingConfig struct {
	// Set to true to disable scheduling
	DisableScheduling bool
	// Set to true to enable scheduler assertions. This results in some performance loss.
	EnableAssertions bool
	// If using PreemptToFairShare,
	// the probability of evicting jobs on a node to balance resource usage.
	// TODO(albin): Remove.
	NodeEvictionProbability float64
	// If using PreemptToFairShare,
	// the probability of evicting jobs on oversubscribed nodes, i.e.,
	// nodes on which the total resource requests are greater than the available resources.
	// TODO(albin): Remove.
	NodeOversubscriptionEvictionProbability float64
	// Only queues allocated more than this fraction of their fair share are considered for preemption.
	ProtectedFractionOfFairShare float64 `validate:"gte=0"`
	// Armada adds a node selector term to every scheduled pod using this label with the node name as value.
	// This to force kube-scheduler to schedule pods on the node chosen by Armada.
	// For example, if NodeIdLabel is "kubernetes.io/hostname" and armada schedules a pod on node "myNode",
	// then Armada adds "kubernetes.io/hostname": "myNode" to the pod node selector before sending it to the executor.
	NodeIdLabel string `validate:"required"`
	// Map from priority class names to priority classes.
	// Must be consistent with Kubernetes priority classes.
	// I.e., priority classes defined here must be defined in all executor clusters and should map to the same priority.
	PriorityClasses map[string]types.PriorityClass `validate:"dive"`
	// Jobs with no priority class are assigned this priority class when ingested by the scheduler.
	// Must be a key in the PriorityClasses map above.
	DefaultPriorityClassName string
	// If set, override the priority class name of pods with this value when sending to an executor.
	PriorityClassNameOverride *string
	// Number of jobs to load from the database at a time.
	MaxQueueLookback uint
	// In each invocation of the scheduler, no more jobs are scheduled once this limit has been exceeded.
	// Note that the total scheduled resources may be greater than this limit.
	MaximumResourceFractionToSchedule map[string]float64
	// Overrides MaximalClusterFractionToSchedule if set for the current pool.
	MaximumResourceFractionToScheduleByPool map[string]map[string]float64
	// The rate at which Armada schedules jobs is rate-limited using a token bucket approach.
	// Specifically, there is a token bucket that persists between scheduling rounds.
	// The bucket fills up at a rate of MaximumSchedulingRate tokens per second and has capacity MaximumSchedulingBurst.
	// A token is removed from the bucket when a scheduling a job and scheduling stops while the bucket is empty.
	//
	// Hence, MaximumSchedulingRate controls the maximum number of jobs scheduled per second in steady-state,
	// i.e., once the burst capacity has been exhausted.
	//
	// Rate-limiting is based on the number of tokens available at the start of each scheduling round,
	// i.e., tokens accumulated while scheduling become available at the start of the next scheduling round.
	//
	// For more information about the rate-limiter, see:
	// https://pkg.go.dev/golang.org/x/time/rate#Limiter
	MaximumSchedulingRate float64 `validate:"gt=0"`
	// MaximumSchedulingBurst controls the burst capacity of the rate-limiter.
	//
	// There are two important implications:
	// - Armada will never schedule more than MaximumSchedulingBurst jobs per scheduling round.
	// - Gang jobs with cardinality greater than MaximumSchedulingBurst can never be scheduled.
	MaximumSchedulingBurst int `validate:"gt=0"`
	// In addition to the global rate-limiter, there is a separate rate-limiter for each queue.
	// These work the same as the global rate-limiter, except they apply only to jobs scheduled from a specific queue.
	//
	// Per-queue version of MaximumSchedulingRate.
	MaximumPerQueueSchedulingRate float64 `validate:"gt=0"`
	// Per-queue version of MaximumSchedulingBurst.
	MaximumPerQueueSchedulingBurst int `validate:"gt=0"`
	// Maximum number of times a job is retried before considered failed.
	MaxRetries uint
	// List of resource names, e.g., []string{"cpu", "memory"}, to consider when computing DominantResourceFairness.
	DominantResourceFairnessResourcesToConsider []string
	// Once a node has been found on which a pod can be scheduled,
	// the scheduler will consider up to the next maxExtraNodesToConsider nodes.
	// The scheduler selects the node with the best score out of the considered nodes.
	// In particular, the score expresses whether preemption is necessary to schedule a pod.
	// Hence, a larger MaxExtraNodesToConsider would reduce the expected number of preemptions.
	// TODO(albin): Remove. It's unused.
	MaxExtraNodesToConsider uint
	// Resource types (e.g. memory or nvidia.com/gpu) that the scheduler keeps track of.
	// Resource types not on this list will be ignored if seen on a node, and any jobs requesting them will fail.
	SupportedResourceTypes []ResourceType
	// Resources, e.g., "cpu", "memory", and "nvidia.com/gpu", for which the scheduler creates indexes for efficient lookup.
	// This list must contain at least one resource. Adding more than one resource is not required, but may speed up scheduling.
	// Ideally, this list contains all resources that frequently constrain which nodes a job can be scheduled onto.
	//
	// In particular, the allocatable resources on each node are rounded to a multiple of the resolution.
	// Lower resolution speeds up scheduling by improving node lookup speed but may prevent scheduling jobs,
	// since the allocatable resources may be rounded down to be a multiple of the resolution.
	//
	// See NodeDb docs for more details.
	IndexedResources []ResourceType
	// Node labels that the scheduler creates indexes for efficient lookup of.
	// Should include node labels frequently used by node selectors on submitted jobs.
	//
	// If not set, no labels are indexed.
	IndexedNodeLabels []string
	// Taint keys that the scheduler creates indexes for efficient lookup of.
	// Should include keys of taints frequently used in tolerations on submitted jobs.
	//
	// If not set, all taints are indexed.
	IndexedTaints []string
	// WellKnownNodeTypes defines a set of well-known node types used to define "home" and "away" nodes for a given priority class.
	WellKnownNodeTypes []WellKnownNodeType `validate:"dive"`
	// Executor that haven't heartbeated in this time period are considered stale.
	// No new jobs are scheduled onto stale executors.
	ExecutorTimeout time.Duration
	// Maximum number of jobs that can be assigned to a executor but not yet acknowledged, before
	// the scheduler is excluded from consideration by the scheduler.
	MaxUnacknowledgedJobsPerExecutor uint
	// If true, do not during scheduling skip jobs with requirements known to be impossible to meet.
	AlwaysAttemptScheduling bool
	// The frequency at which the scheduler updates the cluster state.
	ExecutorUpdateFrequency time.Duration
	// Controls node and queue success probability estimation.
	FailureProbabilityEstimation FailureEstimatorConfig
	// Controls node quarantining, i.e., removing from consideration for scheduling misbehaving nodes.
	NodeQuarantining NodeQuarantinerConfig
	// Controls queue quarantining, i.e., rate-limiting scheduling from misbehaving queues.
	QueueQuarantining QueueQuarantinerConfig
	// Defines the order in which pools will be scheduled. Higher priority pools will be scheduled first
	PoolSchedulePriority map[string]int
	// Default priority for pools that are not in the above list
	DefaultPoolSchedulePriority int
}

SchedulingConfig contains config controlling the Armada scheduler.

The Armada scheduler is in charge of assigning pods to cluster and nodes. The Armada scheduler is part of the Armada control plane.

Features: 1. Queuing and fairly dividing resources between users. 2. Fair preemption, including between jobs of equal priority to balance resource allocation. 3. Gang scheduling, optional across clusters, and with lower and upper bounds on the number of jobs scheduled.

Note that Armada still relies on kube-scheduler for binding of pods to nodes. This is achieved by adding to each pod created by Armada a node selector that matches only the intended node.

type WellKnownNodeType

type WellKnownNodeType struct {
	// Name is the unique identifier for this node type.
	Name string `validate:"required"`
	// Taints is the set of taints that characterizes this node type; a node is
	// part of this node type if and only if it has all of these taints.
	Taints []v1.Taint
}

A WellKnownNodeType defines a set of nodes; see AwayNodeType.