configuration

Documentation

Index

• Constants
• func IsSchedulingAnnotation(annotation string) bool
• type ArmadaConfig
• type PostgresConfig
• type PulsarConfig
• type QueryApiConfig
• type SubmissionConfig

Constants

const (
	// GangIdAnnotation maps to a unique id of the gang the job is part of; jobs with equal value make up a gang.
	// All jobs in a gang are guaranteed to be scheduled onto the same cluster at the same time.
	GangIdAnnotation = "armadaproject.io/gangId"
	// GangCardinalityAnnotation All jobs in a gang must specify the total number of jobs in the gang via this annotation.
	// The cardinality should be expressed as a positive integer, e.g., "3".
	GangCardinalityAnnotation = "armadaproject.io/gangCardinality"
	// The jobs that make up a gang may be constrained to be scheduled across a set of uniform nodes.
	// Specifically, if provided, all gang jobs are scheduled onto nodes for which the value of the provided label is equal.
	// Used to ensure, e.g., that all gang jobs are scheduled onto the same cluster or rack.
	GangNodeUniformityLabelAnnotation = "armadaproject.io/gangNodeUniformityLabel"
	// GangNumJobsScheduledAnnotation is set by the scheduler and indicates how many gang jobs were scheduled.
	// FailFastAnnotation, if set to true, ensures Armada does not re-schedule jobs that fail to start.
	// Instead, the job the pod is part of fails immediately.
	FailFastAnnotation = "armadaproject.io/failFast"
)

Functions

func IsSchedulingAnnotation

func IsSchedulingAnnotation(annotation string) bool

Types

type ArmadaConfig

type ArmadaConfig struct {
	Auth authconfig.AuthConfig

	GrpcPort    uint16
	HttpPort    uint16
	MetricsPort uint16
	Profiling   *profilingconfig.ProfilingConfig

	CorsAllowedOrigins []string
	GrpcGatewayPath    string

	Grpc grpcconfig.GrpcConfig

	SchedulerApiConnection client.ApiConnectionDetails

	EventsApiRedis redis.UniversalOptions
	Pulsar         PulsarConfig
	Postgres       PostgresConfig // Needs to point to the lookout db
	QueryApi       QueryApiConfig

	// Period At which the Queue cache will be refreshed
	QueueCacheRefreshPeriod time.Duration

	// Config relating to job submission.
	Submission SubmissionConfig
}

type PostgresConfig

type PostgresConfig struct {
	Connection map[string]string
}

TODO: we can probably just typedef this to map[string]string

type PulsarConfig

type PulsarConfig struct {
	// Pulsar URL
	URL string `validate:"required"`
	// Path to the trusted TLS certificate file (must exist)
	TLSTrustCertsFilePath string
	// Whether Pulsar client accept untrusted TLS certificate from broker
	TLSAllowInsecureConnection bool
	// Whether the Pulsar client will validate the hostname in the broker's TLS Cert matches the actual hostname.
	TLSValidateHostname bool
	// Max number of connections to a single broker that will be kept in the pool. (Default: 1 connection)
	MaxConnectionsPerBroker int
	// Whether Pulsar authentication is enabled
	AuthenticationEnabled bool
	// Authentication type. For now only "JWT" auth is valid
	AuthenticationType string
	// Path to the JWT token (must exist). This must be set if AuthenticationType is "JWT"
	JwtTokenPath string
	// The pulsar topic that Jobset Events will be published to
	JobsetEventsTopic string
	// Compression to use.  Valid values are "None", "LZ4", "Zlib", "Zstd".  Default is "None"
	CompressionType pulsar.CompressionType
	// Compression Level to use.  Valid values are "Default", "Better", "Faster".  Default is "Default"
	CompressionLevel pulsar.CompressionLevel
	// Maximum allowed Events per message
	MaxAllowedEventsPerMessage int `validate:"gte=0"`
	// Maximum allowed message size in bytes
	MaxAllowedMessageSize uint
	// Timeout when sending messages asynchronously
	SendTimeout time.Duration `validate:"required"`
	// Backoff from polling when Pulsar returns an error
	BackoffTime time.Duration
	// Number of pulsar messages that will be queued by the pulsar consumer.
	ReceiverQueueSize int
}

type QueryApiConfig

type QueryApiConfig struct {
	MaxQueryItems int
}

type SubmissionConfig

type SubmissionConfig struct {
	// The priorityClassName field on submitted pod must be either empty or in this list.
	// These names should correspond to priority classes defined in schedulingConfig.
	AllowedPriorityClassNames map[string]bool
	// Priority class name assigned to pods that do not specify one.
	// Must be an entry in PriorityClasses above.
	DefaultPriorityClassName string
	// Default job resource limits added to pods.
	DefaultJobLimits armadaresource.ComputeResources
	// Tolerations added to all submitted pods.
	DefaultJobTolerations []v1.Toleration
	// Tolerations added to all submitted pods of a given priority class.
	DefaultJobTolerationsByPriorityClass map[string][]v1.Toleration
	// Tolerations added to all submitted pods requesting a non-zero amount of some resource.
	DefaultJobTolerationsByResourceRequest map[string][]v1.Toleration
	// Pods of size greater than this are rejected at submission.
	MaxPodSpecSizeBytes uint
	// Jobs requesting less than this amount of resources are rejected at submission.
	MinJobResources v1.ResourceList
	// Default value of GangNodeUniformityLabelAnnotation if not set on submitted jobs.
	// TODO(albin): We should add a label to nodes in the nodeDb indicating which cluster it came from.
	//              If we do, we can default to that label if the uniformity label is empty.
	DefaultGangNodeUniformityLabel string
	// Minimum allowed termination grace period for pods submitted to Armada.
	// Should normally be set to a positive value, e.g., "10m".
	// Since a zero grace period causes Kubernetes to force delete pods, which may causes issues with container resource cleanup.
	//
	// The grace period of pods that either
	// - do not set a grace period, or
	// - explicitly set a grace period of 0 seconds,
	// is automatically set to MinTerminationGracePeriod.
	MinTerminationGracePeriod time.Duration
	// Max allowed grace period.
	// Should normally not be set greater than single-digit minutes,
	// since cancellation and preemption may need to wait for this amount of time.
	MaxTerminationGracePeriod time.Duration
	// Default activeDeadline for all pods that don't explicitly set activeDeadlineSeconds.
	// Is trumped by DefaultActiveDeadlineByResourceRequest.
	DefaultActiveDeadline time.Duration
	// Default activeDeadline for pods with at least one container requesting a given resource.
	// For example, if
	// DefaultActiveDeadlineByResourceRequest: map[string]time.Duration{"gpu": time.Second},
	// then all pods requesting a non-zero amount of gpu and don't explicitly set activeDeadlineSeconds
	// will have activeDeadlineSeconds set to 1.
	// Trumps DefaultActiveDeadline.
	DefaultActiveDeadlineByResourceRequest map[string]time.Duration
	// Maximum ratio of limits:requests per resource. Jobs who have a higher limits:resource ratio than this will be rejected.
	// Any resource type missing from this map will default to 1.0.
	MaxOversubscriptionByResourceRequest map[string]float64
	// Enforce that an init containers requestion non-integer cpu. This is due to https://github.com/kubernetes/kubernetes/issues/112228
	AssertInitContainersRequestFractionalCpu bool
}

SubmissionConfig contains config relating to job submission.