建立自訂機構政策限制

本頁說明如何使用 Organization Policy Service 自訂限制，限制對下列 Google Cloud 資源執行的特定作業：

  • aiplatform.googleapis.com/CustomJob
  • aiplatform.googleapis.com/HyperparameterTuningJob
  • aiplatform.googleapis.com/NasJob

如要進一步瞭解機構政策，請參閱「自訂機構政策」。

關於機構政策和限制

Google Cloud 機構政策服務可讓您透過程式以集中方式控管機構的資源。身為機構政策管理員，您可以定義機構政策，也就是一組稱為「限制」的限制，適用於Google Cloud 資源和Google Cloud 資源階層中這些資源的子系。您可以在機構、資料夾或專案層級強制執行機構政策。

機構政策提供各種 Google Cloud 服務的內建代管限制。不過，如要更精細地自訂機構政策中受限的特定欄位，也可以建立自訂限制，並在機構政策中使用這些自訂限制。

政策繼承

根據預設，機構政策會由您強制執行政策的資源子系繼承。舉例來說，如果您對資料夾強制執行政策， Google Cloud 系統會對該資料夾中的所有專案強制執行政策。如要進一步瞭解這項行為及如何變更，請參閱「階層評估規則」。

優點

您可以透過自訂機構政策，允許或拒絕 Vertex AI 訓練資源的特定值。舉例來說，如果建立自訂訓練工作的要求未通過組織政策設定的自訂限制驗證，要求就會失敗，並向呼叫端傳回錯誤。

限制

與所有機構政策限制一樣，政策變更不會回溯套用至現有資源。

  • 新政策不會影響現有的資源設定。
  • 除非您將資源設定中的值從符合規定變更為不符合規定，否則現有資源設定仍有效。

事前準備

  1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

  2. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

    Go to project selector

  3. Verify that billing is enabled for your Google Cloud project.

  4. Install the Google Cloud CLI.

  5. 如果您使用外部識別資訊提供者 (IdP)，請先 使用聯合身分登入 gcloud CLI。

  6. 如要初始化 gcloud CLI，請執行下列指令： 

    gcloud init

  7. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

    Go to project selector

  8. Verify that billing is enabled for your Google Cloud project.

  9. Install the Google Cloud CLI.

  10. 如果您使用外部識別資訊提供者 (IdP)，請先 使用聯合身分登入 gcloud CLI。

  11. 如要初始化 gcloud CLI，請執行下列指令： 

    gcloud init
  12. 請確認您知道機構 ID。

    13. 必要的角色

    如要取得管理自訂機構政策所需的權限，請要求管理員為您授予機構資源的機構政策管理員 (roles/orgpolicy.policyAdmin) IAM 角色。如要進一步瞭解如何授予角色，請參閱「管理專案、資料夾和機構的存取權」。

    您或許還可透過自訂角色或其他預先定義的角色取得必要權限。

    建立自訂限制

    自訂限制是在 YAML 檔案中定義，當中包含您要強制執行機構政策的服務所支援的資源、方法、條件和動作。自訂限制的條件是使用一般運算語言 (CEL) 定義。如要進一步瞭解如何使用 CEL 在自訂限制中建構條件，請參閱「建立及管理自訂限制」一文的 CEL 一節。

    如要建立自訂限制，請使用下列格式建立 YAML 檔案：

    name: organizations/ORGANIZATION_ID/customConstraints/CONSTRAINT_NAME
resourceTypes:
- RESOURCE_NAME
methodTypes:
- CREATE
- UPDATE
condition: "CONDITION"
actionType: ACTION
displayName: DISPLAY_NAME
description: DESCRIPTION

    更改下列內容：

    • ORGANIZATION_ID：您的機構 ID，例如 123456789。

    • CONSTRAINT_NAME：新自訂限制的名稱。自訂限制條件必須以 custom. 開頭，且只能包含大寫英文字母、小寫英文字母或數字。例如，custom.restrictMachineType。這個欄位的長度上限為 70 個字元。

    • RESOURCE_NAME：包含要限制物件和欄位的Google Cloud 資源完整名稱。例如：aiplatform.googleapis.com/CustomJob。

    • CONDITION：針對支援服務資源的代表項目編寫的 CEL 條件。這個欄位的長度上限為 1000 個字元。如要進一步瞭解可編寫條件的資源，請參閱「支援的資源」。例如："resource.jobSpec.workerPoolSpecs.exists(spec, spec.machineSpec.machineType != \"n1-standard-4\")"。

    • ACTION：如果符合 condition，應採取的動作。可能的值為 ALLOW 和 DENY。

    • DISPLAY_NAME：限制條件的易記名稱。這個欄位的長度上限為 200 個字元。

    • DESCRIPTION：違反政策時，要以錯誤訊息形式顯示的限制說明。這個欄位的長度上限為 2000 個字元。

    如要進一步瞭解如何建立自訂限制，請參閱「定義自訂限制」。

    設定自訂限制

    為新的自訂限制建立 YAML 檔案後，您必須進行設定，才能在貴機構的機構政策中使用該檔案。如要設定自訂限制，請使用 gcloud org-policies set-custom-constraint 指令： 
    gcloud org-policies set-custom-constraint CONSTRAINT_PATH
     將 CONSTRAINT_PATH 替換為自訂限制檔案的完整路徑。例如：/home/user/customconstraint.yaml。 完成後，自訂限制就會顯示在 Google Cloud 機構政策清單中，做為機構政策使用。如要確認自訂限制存在，請使用 gcloud org-policies list-custom-constraints 指令：
    gcloud org-policies list-custom-constraints --organization=ORGANIZATION_ID
     將 ORGANIZATION_ID 替換為機構資源的 ID。 詳情請參閱「查看組織政策」。

    強制執行自訂機構政策

    如要強制執行限制，請建立參照該限制的機構政策，然後將該政策套用至 Google Cloud 資源。

    控制台

    1. 前往 Google Cloud 控制台的「Organization policies」(組織政策) 頁面。

      前往「機構政策」

    2. 在專案選擇工具中，選取要設定機構政策的專案。
    3. 在「Organization policies」(機構政策) 頁面上的清單中選取限制條件，即可查看該限制條件的「Policy details」(政策詳情) 頁面。
    4. 如要設定這項資源的機構政策，請按一下「管理政策」。
    5. 在「編輯政策」頁面中，選取「覆寫上層政策」。
    6. 按一下「新增規則」。
    7. 在「Enforcement」(強制執行) 區段中，選取是否要強制執行這項機構政策。
    8. 選用：如要根據標記設定機構政策條件，請按一下「新增條件」。請注意，如果為組織政策新增條件式規則，您必須至少新增一項無條件規則，否則無法儲存政策。詳情請參閱「使用標記設定組織政策」。
    9. 按一下「測試變更」，模擬機構政策的影響。舊版受管理限制不支援政策模擬。詳情請參閱「 使用 Policy Simulator 測試組織政策變更」。
    10. 如要完成並套用機構政策，請按一下「設定政策」。這項政策最多需要 15 分鐘才會生效。

    gcloud

    如要建立含有布林值規則的機構政策，請建立參照限制的政策 YAML 檔案：

          name: projects/PROJECT_ID/policies/CONSTRAINT_NAME
      spec:
        rules:
        - enforce: true

    取代下列項目：

    • PROJECT_ID：要強制執行限制的專案。
    • CONSTRAINT_NAME：您為自訂限制定義的名稱。例如：custom.restrictMachineType。

    如要強制執行包含限制的機構政策，請執行下列指令： 

        gcloud org-policies set-policy POLICY_PATH

    將 POLICY_PATH 替換為機構政策 YAML 檔案的完整路徑。這項政策最多需要 15 分鐘才會生效。

    測試自訂機構政策

    以下範例會建立自訂限制和政策，限制機器類型。

    開始之前，請務必瞭解下列事項：

    • 您的機構 ID
    • 專案 ID

    建立限制

    1. 將下列檔案儲存為 constraint-custom-job.yaml：

      name: organizations/ORGANIZATION_ID/customConstraints/custom.restrictMachineType
resourceTypes:
- aiplatform.googleapis.com/CustomJob
methodTypes:
- CREATE
condition: "resource.jobSpec.workerPoolSpecs.exists(spec, spec.machineSpec.machineType != \"n1-standard-4\")"
actionType: DENY
displayName: Restrict machine type custom training jobs
description: All new custom training jobs must use n1-standard-4 machines.

      這項限制會規定所有新的自訂訓練工作都必須使用 n1-standard-4 機器類型。如果自訂訓練工作未使用此機器類型，系統會拒絕建立該工作。

    2. 套用限制：

      gcloud org-policies set-custom-constraint ~/constraint-custom-job.yaml

    3. 確認限制條件是否存在：

      gcloud org-policies list-custom-constraints --organization=ORGANIZATION_ID

      輸出結果會與下列內容相似：

      CUSTOM_CONSTRAINT                            ACTION_TYPE  METHOD_TYPES   RESOURCE_TYPES                                     DISPLAY_NAME
custom.restrictMachineType                   DENY         CREATE         aiplatform.googleapis.com/CustomJob                Restrict machine type custom training jobs
...

    建立政策

    1. 將下列檔案儲存為 policy-deny-custom-job.yaml：

      name: projects/PROJECT_ID/policies/custom.restrictMachineType
spec:
  rules:
  - enforce: true

      將 PROJECT_ID 替換為您的專案 ID。

    2. 套用政策：

      gcloud org-policies set-policy ~/policy-deny-custom-job.yaml

    3. 確認政策存在：

      gcloud org-policies list --project=PROJECT_ID

      輸出結果會與下列內容相似：

      CONSTRAINT                          LIST_POLICY  BOOLEAN_POLICY        ETAG
custom.restrictMachineType          -            SET                   CLj9zMIGEIiS3K4D-

    套用政策後，請等待約兩分鐘， Google Cloud 就會開始強制執行政策。

    測試政策

    嘗試使用受限的機器類型建立 Vertex AI 自訂訓練工作：

    gcloud ai custom-jobs create \
  --region=LOCATION \
  --display-name=JOB_NAME \
  --worker-pool-spec=machine-type=n1-standard-8,replica-count=REPLICA_COUNT,container-image-uri=CUSTOM_CONTAINER_IMAGE_URI

    輸出內容如下：

    Operation denied by org policy on resource 'projects/PROJECT_ID/locations/LOCATION': ["customConstraints/custom.restrictMachineType": "All new custom training jobs must use n1-standard-4 machines."]

    常見用途的自訂機構政策範例

    下表提供一些常見用途的自訂限制語法：

    說明 限制語法
    限制 Vertex AI 自訂訓練工作的機器類型 
          name: organizations/ORGANIZATION_ID/customConstraints/custom.restrictMachineType
      resourceTypes:
      - aiplatform.googleapis.com/CustomJob
      methodTypes:
      - CREATE
      condition: "resource.jobSpec.workerPoolSpecs.exists(spec, spec.machineSpec.machineType != "n1-standard-4")"
      actionType: DENY
      displayName: Restrict machine type custom training jobs
      description: All new custom training jobs must use n1-standard-4 machines.

    Vertex AI 支援的資源

    下表列出可在自訂限制中參照的 Vertex AI 資源。

    資源 欄位
    aiplatform.googleapis.com/CustomJob resource.displayName
    resource.encryptionSpec.kmsKeyName
    resource.jobSpec.baseOutputDirectory.outputUriPrefix
    resource.jobSpec.enableDashboardAccess
    resource.jobSpec.enableWebAccess
    resource.jobSpec.experiment
    resource.jobSpec.experimentRun
    resource.jobSpec.models
    resource.jobSpec.network
    resource.jobSpec.persistentResourceId
    resource.jobSpec.protectedArtifactLocationId
    resource.jobSpec.pscInterfaceConfig.networkAttachment
    resource.jobSpec.reservedIpRanges
    resource.jobSpec.scheduling.disableRetries
    resource.jobSpec.scheduling.maxWaitDuration
    resource.jobSpec.scheduling.restartJobOnWorkerRestart
    resource.jobSpec.scheduling.strategy
    resource.jobSpec.scheduling.timeout
    resource.jobSpec.serviceAccount
    resource.jobSpec.tensorboard
    resource.jobSpec.workerPoolSpecs.containerSpec.args
    resource.jobSpec.workerPoolSpecs.containerSpec.command
    resource.jobSpec.workerPoolSpecs.containerSpec.imageUri
    resource.jobSpec.workerPoolSpecs.diskSpec.bootDiskSizeGb
    resource.jobSpec.workerPoolSpecs.diskSpec.bootDiskType
    resource.jobSpec.workerPoolSpecs.machineSpec.acceleratorCount
    resource.jobSpec.workerPoolSpecs.machineSpec.acceleratorType
    resource.jobSpec.workerPoolSpecs.machineSpec.machineType
    resource.jobSpec.workerPoolSpecs.machineSpec.reservationAffinity.key
    resource.jobSpec.workerPoolSpecs.machineSpec.reservationAffinity.reservationAffinityType
    resource.jobSpec.workerPoolSpecs.machineSpec.reservationAffinity.values
    resource.jobSpec.workerPoolSpecs.machineSpec.tpuTopology
    resource.jobSpec.workerPoolSpecs.nfsMounts.mountPoint
    resource.jobSpec.workerPoolSpecs.nfsMounts.path
    resource.jobSpec.workerPoolSpecs.nfsMounts.server
    resource.jobSpec.workerPoolSpecs.pythonPackageSpec.args
    resource.jobSpec.workerPoolSpecs.pythonPackageSpec.executorImageUri
    resource.jobSpec.workerPoolSpecs.pythonPackageSpec.packageUris
    resource.jobSpec.workerPoolSpecs.pythonPackageSpec.pythonModule
    resource.jobSpec.workerPoolSpecs.replicaCount
    aiplatform.googleapis.com/HyperparameterTuningJob resource.displayName
    resource.encryptionSpec.kmsKeyName
    resource.maxFailedTrialCount
    resource.maxTrialCount
    resource.parallelTrialCount
    resource.studySpec.algorithm
    resource.studySpec.convexAutomatedStoppingSpec.learningRateParameterName
    resource.studySpec.convexAutomatedStoppingSpec.maxStepCount
    resource.studySpec.convexAutomatedStoppingSpec.minMeasurementCount
    resource.studySpec.convexAutomatedStoppingSpec.minStepCount
    resource.studySpec.convexAutomatedStoppingSpec.updateAllStoppedTrials
    resource.studySpec.convexAutomatedStoppingSpec.useElapsedDuration
    resource.studySpec.decayCurveStoppingSpec.useElapsedDuration
    resource.studySpec.measurementSelectionType
    resource.studySpec.medianAutomatedStoppingSpec.useElapsedDuration
    resource.studySpec.metrics.goal
    resource.studySpec.metrics.metricId
    resource.studySpec.metrics.safetyConfig.desiredMinSafeTrialsFraction
    resource.studySpec.metrics.safetyConfig.safetyThreshold
    resource.studySpec.observationNoise
    resource.studySpec.parameters.categoricalValueSpec.defaultValue
    resource.studySpec.parameters.categoricalValueSpec.values
    resource.studySpec.parameters.conditionalParameterSpecs.parentCategoricalValues.values
    resource.studySpec.parameters.conditionalParameterSpecs.parentDiscreteValues.values
    resource.studySpec.parameters.conditionalParameterSpecs.parentIntValues.values
    resource.studySpec.parameters.discreteValueSpec.defaultValue
    resource.studySpec.parameters.discreteValueSpec.values
    resource.studySpec.parameters.doubleValueSpec.defaultValue
    resource.studySpec.parameters.doubleValueSpec.maxValue
    resource.studySpec.parameters.doubleValueSpec.minValue
    resource.studySpec.parameters.integerValueSpec.defaultValue
    resource.studySpec.parameters.integerValueSpec.maxValue
    resource.studySpec.parameters.integerValueSpec.minValue
    resource.studySpec.parameters.parameterId
    resource.studySpec.parameters.scaleType
    resource.studySpec.studyStoppingConfig.maxDurationNoProgress
    resource.studySpec.studyStoppingConfig.maximumRuntimeConstraint.endTime
    resource.studySpec.studyStoppingConfig.maximumRuntimeConstraint.maxDuration
    resource.studySpec.studyStoppingConfig.maxNumTrials
    resource.studySpec.studyStoppingConfig.maxNumTrialsNoProgress
    resource.studySpec.studyStoppingConfig.minimumRuntimeConstraint.endTime
    resource.studySpec.studyStoppingConfig.minimumRuntimeConstraint.maxDuration
    resource.studySpec.studyStoppingConfig.minNumTrials
    resource.studySpec.studyStoppingConfig.shouldStopAsap
    resource.trialJobSpec.baseOutputDirectory.outputUriPrefix
    resource.trialJobSpec.enableDashboardAccess
    resource.trialJobSpec.enableWebAccess
    resource.trialJobSpec.experiment
    resource.trialJobSpec.experimentRun
    resource.trialJobSpec.models
    resource.trialJobSpec.network
    resource.trialJobSpec.persistentResourceId
    resource.trialJobSpec.protectedArtifactLocationId
    resource.trialJobSpec.pscInterfaceConfig.networkAttachment
    resource.trialJobSpec.reservedIpRanges
    resource.trialJobSpec.scheduling.disableRetries
    resource.trialJobSpec.scheduling.maxWaitDuration
    resource.trialJobSpec.scheduling.restartJobOnWorkerRestart
    resource.trialJobSpec.scheduling.strategy
    resource.trialJobSpec.scheduling.timeout
    resource.trialJobSpec.serviceAccount
    resource.trialJobSpec.tensorboard
    resource.trialJobSpec.workerPoolSpecs.containerSpec.args
    resource.trialJobSpec.workerPoolSpecs.containerSpec.command
    resource.trialJobSpec.workerPoolSpecs.containerSpec.imageUri
    resource.trialJobSpec.workerPoolSpecs.diskSpec.bootDiskSizeGb
    resource.trialJobSpec.workerPoolSpecs.diskSpec.bootDiskType
    resource.trialJobSpec.workerPoolSpecs.machineSpec.acceleratorCount
    resource.trialJobSpec.workerPoolSpecs.machineSpec.acceleratorType
    resource.trialJobSpec.workerPoolSpecs.machineSpec.machineType
    resource.trialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.key
    resource.trialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.reservationAffinityType
    resource.trialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.values
    resource.trialJobSpec.workerPoolSpecs.machineSpec.tpuTopology
    resource.trialJobSpec.workerPoolSpecs.nfsMounts.mountPoint
    resource.trialJobSpec.workerPoolSpecs.nfsMounts.path
    resource.trialJobSpec.workerPoolSpecs.nfsMounts.server
    resource.trialJobSpec.workerPoolSpecs.pythonPackageSpec.args
    resource.trialJobSpec.workerPoolSpecs.pythonPackageSpec.executorImageUri
    resource.trialJobSpec.workerPoolSpecs.pythonPackageSpec.packageUris
    resource.trialJobSpec.workerPoolSpecs.pythonPackageSpec.pythonModule
    resource.trialJobSpec.workerPoolSpecs.replicaCount
    aiplatform.googleapis.com/NasJob resource.displayName
    resource.encryptionSpec.kmsKeyName
    resource.nasJobSpec.multiTrialAlgorithmSpec.metric.goal
    resource.nasJobSpec.multiTrialAlgorithmSpec.metric.metricId
    resource.nasJobSpec.multiTrialAlgorithmSpec.multiTrialAlgorithm
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.maxFailedTrialCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.maxParallelTrialCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.maxTrialCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.baseOutputDirectory.outputUriPrefix
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.enableDashboardAccess
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.enableWebAccess
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.experiment
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.experimentRun
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.models
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.network
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.persistentResourceId
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.protectedArtifactLocationId
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.pscInterfaceConfig.networkAttachment
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.reservedIpRanges
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.scheduling.disableRetries
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.scheduling.maxWaitDuration
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.scheduling.restartJobOnWorkerRestart
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.scheduling.strategy
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.scheduling.timeout
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.serviceAccount
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.tensorboard
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.containerSpec.args
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.containerSpec.command
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.containerSpec.imageUri
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.diskSpec.bootDiskSizeGb
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.diskSpec.bootDiskType
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.acceleratorCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.acceleratorType
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.machineType
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.key
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.reservationAffinityType
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.values
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.machineSpec.tpuTopology
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.nfsMounts.mountPoint
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.nfsMounts.path
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.nfsMounts.server
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.pythonPackageSpec.args
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.pythonPackageSpec.executorImageUri
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.pythonPackageSpec.packageUris
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.pythonPackageSpec.pythonModule
    resource.nasJobSpec.multiTrialAlgorithmSpec.searchTrialSpec.searchTrialJobSpec.workerPoolSpecs.replicaCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.frequency
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.maxParallelTrialCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.baseOutputDirectory.outputUriPrefix
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.enableDashboardAccess
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.enableWebAccess
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.experiment
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.experimentRun
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.models
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.network
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.persistentResourceId
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.protectedArtifactLocationId
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.pscInterfaceConfig.networkAttachment
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.reservedIpRanges
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.scheduling.disableRetries
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.scheduling.maxWaitDuration
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.scheduling.restartJobOnWorkerRestart
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.scheduling.strategy
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.scheduling.timeout
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.serviceAccount
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.tensorboard
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.containerSpec.args
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.containerSpec.command
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.containerSpec.imageUri
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.diskSpec.bootDiskSizeGb
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.diskSpec.bootDiskType
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.acceleratorCount
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.acceleratorType
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.machineType
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.key
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.reservationAffinityType
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.reservationAffinity.values
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.machineSpec.tpuTopology
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.nfsMounts.mountPoint
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.nfsMounts.path
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.nfsMounts.server
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.pythonPackageSpec.args
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.pythonPackageSpec.executorImageUri
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.pythonPackageSpec.packageUris
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.pythonPackageSpec.pythonModule
    resource.nasJobSpec.multiTrialAlgorithmSpec.trainTrialSpec.trainTrialJobSpec.workerPoolSpecs.replicaCount
    resource.nasJobSpec.resumeNasJobId
    resource.nasJobSpec.searchSpaceSpec

    後續步驟