Skip to content

Latest commit

 

History

History
459 lines (446 loc) · 17.4 KB

sagemaker.md

File metadata and controls

459 lines (446 loc) · 17.4 KB

SageMaker

AWS's sagemaker service · portable interface driver.Service · AWS index

Operations (207)

Operation Description
AddAssociation
AddTags
AssociateTrialComponent
BatchDescribeModelPackage
CreateAction
CreateApp
CreateAppImageConfig
CreateArtifact
CreateAutoMLJobV2
CreateCluster
CreateCodeRepository
CreateCompilationJob
CreateContext
CreateDomain
CreateEndpoint
CreateEndpointConfig
CreateExperiment
CreateFeatureGroup
CreateHyperParameterTuningJob
CreateImage
CreateImageVersion
CreateInferenceComponent
CreateLabelingJob
CreateModel
CreateModelPackage
CreateModelPackageGroup
CreateMonitoringJobDefinition
CreateMonitoringSchedule
CreateNotebookInstance
CreateNotebookInstanceLifecycleConfig
CreatePipeline
CreateProcessingJob
CreateProject
CreateSpace
CreateTrainingJob
CreateTransformJob
CreateTrial
CreateTrialComponent
CreateUserProfile
CreateWorkforce
CreateWorkteam
DeleteAction
DeleteApp
DeleteAppImageConfig
DeleteArtifact
DeleteAssociation
DeleteCluster
DeleteCodeRepository
DeleteContext
DeleteDomain
DeleteEndpoint
DeleteEndpointConfig
DeleteExperiment
DeleteFeatureGroup
DeleteImage
DeleteImageVersion
DeleteInferenceComponent
DeleteModel
DeleteModelPackage
DeleteModelPackageGroup
DeleteModelPackageGroupPolicy
DeleteMonitoringJobDefinition
DeleteMonitoringSchedule
DeleteNotebookInstance
DeleteNotebookInstanceLifecycleConfig
DeletePipeline
DeleteProject
DeleteRecord
DeleteSpace
DeleteTags
DeleteTrial
DeleteTrialComponent
DeleteUserProfile
DeleteWorkforce
DeleteWorkteam
DescribeAction
DescribeApp
DescribeAppImageConfig
DescribeArtifact
DescribeAutoMLJobV2
DescribeCluster
DescribeClusterNode
DescribeCodeRepository
DescribeCompilationJob
DescribeContext
DescribeDomain
DescribeEndpoint
DescribeEndpointConfig
DescribeExperiment
DescribeFeatureGroup
DescribeHyperParameterTuningJob
DescribeImage
DescribeImageVersion
DescribeInferenceComponent
DescribeLabelingJob
DescribeLineageGroup
DescribeModel
DescribeModelPackage
DescribeModelPackageGroup
DescribeMonitoringJobDefinition
DescribeMonitoringSchedule
DescribeNotebookInstance
DescribeNotebookInstanceLifecycleConfig
DescribePipeline
DescribePipelineExecution
DescribeProcessingJob
DescribeProject
DescribeSpace
DescribeSubscribedWorkteam
DescribeTrainingJob
DescribeTransformJob
DescribeTrial
DescribeTrialComponent
DescribeUserProfile
DescribeWorkforce
DescribeWorkteam
DisassociateTrialComponent
GetLineageGroupPolicy
GetModelPackageGroupPolicy
GetRecord
GetSearchSuggestions
InvokeEndpoint
InvokeEndpointAsync
ListActions
ListAliases
ListAppImageConfigs
ListApps
ListArtifacts
ListAssociations ListAssociations filters by opts.Filters keys AssociationType, SourceArn,
ListAutoMLJobs
ListClusterNodes
ListClusters
ListCodeRepositories
ListCompilationJobs
ListContexts
ListDomains
ListEndpointConfigs
ListEndpoints
ListExperiments
ListFeatureGroups
ListHyperParameterTuningJobs
ListImageVersions
ListImages
ListInferenceComponents
ListLabelingJobs
ListLabelingJobsForWorkteam
ListLineageGroups
ListModelPackageGroups
ListModelPackages
ListModels
ListMonitoringAlertHistory
ListMonitoringAlerts
ListMonitoringExecutions
ListMonitoringJobDefinitions
ListMonitoringSchedules
ListNotebookInstanceLifecycleConfigs
ListNotebookInstances
ListPipelineExecutions
ListPipelines
ListProcessingJobs
ListProjects
ListSpaces
ListSubscribedWorkteams
ListTags
ListTrainingJobs
ListTransformJobs
ListTrialComponents ListTrialComponents filters by opts.Filters keys ExperimentName, TrialName
ListTrials
ListUserProfiles
ListWorkforces
ListWorkteams
PutModelPackageGroupPolicy
PutRecord
QueryLineage
Search
StartMonitoringSchedule
StartNotebookInstance
StartPipelineExecution
StopAutoMLJob
StopCompilationJob
StopHyperParameterTuningJob
StopLabelingJob
StopMonitoringSchedule
StopNotebookInstance
StopPipelineExecution
StopProcessingJob
StopTrainingJob
StopTransformJob
UpdateAction
UpdateAppImageConfig
UpdateArtifact
UpdateCluster
UpdateContext
UpdateEndpoint
UpdateEndpointWeightsAndCapacities
UpdateImage
UpdateImageVersion
UpdateModelPackage
UpdateMonitoringAlert
UpdateMonitoringSchedule
UpdateNotebookInstance
UpdateNotebookInstanceLifecycleConfig
UpdatePipeline
UpdateProject
UpdateTrialComponent
UpdateWorkforce
UpdateWorkteam

Optional capabilities

Discovered by type assertion; only some providers implement these.

Runtime

Runtime is the SageMaker inference runtime (the sagemaker-runtime service):

Operation Description
InvokeEndpoint
InvokeEndpointAsync

SageMaker

SageMaker is the interface that SageMaker AI control-plane implementations

Operation Description
AddAssociation
AddTags
AssociateTrialComponent
BatchDescribeModelPackage
CreateAction
CreateApp
CreateAppImageConfig
CreateArtifact
CreateAutoMLJobV2
CreateCluster
CreateCodeRepository
CreateCompilationJob
CreateContext
CreateDomain
CreateEndpoint
CreateEndpointConfig
CreateExperiment
CreateFeatureGroup
CreateHyperParameterTuningJob
CreateImage
CreateImageVersion
CreateInferenceComponent
CreateLabelingJob
CreateModel
CreateModelPackage
CreateModelPackageGroup
CreateMonitoringJobDefinition
CreateMonitoringSchedule
CreateNotebookInstance
CreateNotebookInstanceLifecycleConfig
CreatePipeline
CreateProcessingJob
CreateProject
CreateSpace
CreateTrainingJob
CreateTransformJob
CreateTrial
CreateTrialComponent
CreateUserProfile
CreateWorkforce
CreateWorkteam
DeleteAction
DeleteApp
DeleteAppImageConfig
DeleteArtifact
DeleteAssociation
DeleteCluster
DeleteCodeRepository
DeleteContext
DeleteDomain
DeleteEndpoint
DeleteEndpointConfig
DeleteExperiment
DeleteFeatureGroup
DeleteImage
DeleteImageVersion
DeleteInferenceComponent
DeleteModel
DeleteModelPackage
DeleteModelPackageGroup
DeleteModelPackageGroupPolicy
DeleteMonitoringJobDefinition
DeleteMonitoringSchedule
DeleteNotebookInstance
DeleteNotebookInstanceLifecycleConfig
DeletePipeline
DeleteProject
DeleteRecord
DeleteSpace
DeleteTags
DeleteTrial
DeleteTrialComponent
DeleteUserProfile
DeleteWorkforce
DeleteWorkteam
DescribeAction
DescribeApp
DescribeAppImageConfig
DescribeArtifact
DescribeAutoMLJobV2
DescribeCluster
DescribeClusterNode
DescribeCodeRepository
DescribeCompilationJob
DescribeContext
DescribeDomain
DescribeEndpoint
DescribeEndpointConfig
DescribeExperiment
DescribeFeatureGroup
DescribeHyperParameterTuningJob
DescribeImage
DescribeImageVersion
DescribeInferenceComponent
DescribeLabelingJob
DescribeLineageGroup
DescribeModel
DescribeModelPackage
DescribeModelPackageGroup
DescribeMonitoringJobDefinition
DescribeMonitoringSchedule
DescribeNotebookInstance
DescribeNotebookInstanceLifecycleConfig
DescribePipeline
DescribePipelineExecution
DescribeProcessingJob
DescribeProject
DescribeSpace
DescribeSubscribedWorkteam
DescribeTrainingJob
DescribeTransformJob
DescribeTrial
DescribeTrialComponent
DescribeUserProfile
DescribeWorkforce
DescribeWorkteam
DisassociateTrialComponent
GetLineageGroupPolicy
GetModelPackageGroupPolicy
GetRecord
GetSearchSuggestions
ListActions
ListAliases
ListAppImageConfigs
ListApps
ListArtifacts
ListAssociations ListAssociations filters by opts.Filters keys AssociationType, SourceArn,
ListAutoMLJobs
ListClusterNodes
ListClusters
ListCodeRepositories
ListCompilationJobs
ListContexts
ListDomains
ListEndpointConfigs
ListEndpoints
ListExperiments
ListFeatureGroups
ListHyperParameterTuningJobs
ListImageVersions
ListImages
ListInferenceComponents
ListLabelingJobs
ListLabelingJobsForWorkteam
ListLineageGroups
ListModelPackageGroups
ListModelPackages
ListModels
ListMonitoringAlertHistory
ListMonitoringAlerts
ListMonitoringExecutions
ListMonitoringJobDefinitions
ListMonitoringSchedules
ListNotebookInstanceLifecycleConfigs
ListNotebookInstances
ListPipelineExecutions
ListPipelines
ListProcessingJobs
ListProjects
ListSpaces
ListSubscribedWorkteams
ListTags
ListTrainingJobs
ListTransformJobs
ListTrialComponents ListTrialComponents filters by opts.Filters keys ExperimentName, TrialName
ListTrials
ListUserProfiles
ListWorkforces
ListWorkteams
PutModelPackageGroupPolicy
PutRecord
QueryLineage
Search
StartMonitoringSchedule
StartNotebookInstance
StartPipelineExecution
StopAutoMLJob
StopCompilationJob
StopHyperParameterTuningJob
StopLabelingJob
StopMonitoringSchedule
StopNotebookInstance
StopPipelineExecution
StopProcessingJob
StopTrainingJob
StopTransformJob
UpdateAction
UpdateAppImageConfig
UpdateArtifact
UpdateCluster
UpdateContext
UpdateEndpoint
UpdateEndpointWeightsAndCapacities
UpdateImage
UpdateImageVersion
UpdateModelPackage
UpdateMonitoringAlert
UpdateMonitoringSchedule
UpdateNotebookInstance
UpdateNotebookInstanceLifecycleConfig
UpdatePipeline
UpdateProject
UpdateTrialComponent
UpdateWorkforce
UpdateWorkteam

Not in scope

  • Nothing runs: training, processing, transform, tuning, AutoML, labeling and compilation jobs, pipeline executions and monitoring executions are records that move through the real status values, never containers. InvokeEndpoint echoes the request body, and an endpoint's container is never started or health-checked.
  • Streaming and batch runtime calls that are not served: InvokeEndpointWithResponseStream, and the feature-store runtime BatchGetRecord.
  • Model Monitor executions are generated on a schedule's cron cadence and are always Completed (a seeded status can be set from Go); monitoring alerts cannot be created through the API (AWS creates them from Model Dashboard violations), so ListMonitoringAlerts is empty when the emulator runs as a server; the seeding hooks (alerts, a seeded model failure, a seeded execution status) are test seams and are not part of the production surface, so the model-failure branch of InvokeEndpoint (HTTP 424, Invocation5XXErrors) runs only from Go tests, and the nested job-definition configuration is stored and echoed verbatim without being validated.
  • Search covers the resource types the emulator stores (training jobs, tuning jobs, endpoints, models, model packages and groups, pipelines and their executions, experiments, trials, trial components, feature groups, projects); a valid resource type it does not store (images and image versions, feature metadata, model cards, pipeline versions, Job) returns an empty result rather than an error. GetSearchSuggestions suggests the Tags., HyperParameters. and Parameters. property names the stored resources carry.
  • Metrics carry nominal constant values (the real service publishes measured latency and utilization): per-job host utilization is published when a job is created and is skipped under --async-settle, and Invocation5XXErrors and InvocationModelErrors appear only when a model failure is seeded from a Go test or the caller names a variant the endpoint lacks (4XX); a successful invocation does not publish a zero datapoint, so an error-rate average over successes alone reads as 1 instead of 0.
  • EventBridge events are published for training, processing, transform and tuning jobs, endpoints, model packages and pipeline executions. There are no Step Status Change, Model, Feature Group, Image, HyperPod or Endpoint Deployment events, and under --async-settle the first state of a change is published at call time and the settled state (IN_SERVICE, Completed, Stopped, Failed) once, the first time the resource is read after it settles, because the mock has no scheduler; state restored from a persisted snapshot can publish its settled state again on its first read. The endpoint status in the event detail follows the upper-snake form in the AWS example (IN_SERVICE), which AWS does not document further.
  • Workforces: one workforce per account, Active immediately, and no sign-in portal or labeling UI is served; ListSubscribedWorkteams is always empty (no Marketplace vendors).
  • Studio custom images are registered by reference: the base image is never pulled or built, so a version is CREATED at once and ContainerImage is the base image.
  • Projects are provisioned by reference: the Service Catalog product is never launched, so a project is CreateCompleted at once with no provisioned products or pipelines behind it.
  • Limits and message texts that AWS documents only as "see Service Quotas" (the 500-project and 25-workteam caps) are the emulator's choice; the exception names are the documented ones.
  • Target-group-style list defaults follow the documented ones (SortOrder Descending for ListEndpoints, ListModels, ListEndpointConfigs and ListTransformJobs; SortBy Name for ListCodeRepositories and ListNotebookInstances). A filter on a field an operation does not have is a ValidationException, never an empty result.
  • Known fidelity gaps left for follow-up: error message wording differs from AWS (AWS quotes ARNs and uses Cannot create already existing ...); ARNs keep the caller's name casing where AWS lowercases the name segment; lineage ARNs use a shorter id than AWS; DeployedImages.ResolvedImage is a copy of the image tag, not a digest; notebook instances skip Pending under --async-settle; the 1 MiB control-plane body cap rejects a maximum-size PipelineDefinition once JSON-escaped; and event payloads carry the tags the resource had at creation.
  • UpdateCodeRepository, UpdateUserProfile and UpdateSpace are not implemented yet (AWS has them), so LastModifiedTime on code repositories, user profiles and spaces equals the creation time; a Terraform aws_sagemaker_code_repository update gets UnknownOperationException.
  • More follow-ups: DeleteWorkteam on a missing team returns {"Success": false} (whether AWS raises ValidationException there is unconfirmed); DeleteDomain checks for children and then deletes without a lock, so a concurrent child create can leave an orphan; and the settled-event bookkeeping keeps one small map entry per settled job for the life of the process.
  • Library API change: driver.ProductionVariant.InitialVariantWeight is a *float64 (nil means omitted, and the provider applies SageMaker's default of 1.0), so Go callers that wrote InitialVariantWeight: 1 must pass a pointer.