Oracle Cloud Infrastructure Documentation

Model Store Back Up and Retention

The Model Store uses a dedicated OSS bucket in each region for storing model artifacts.

Important

The back up functionality is only available in commercial realms where Data Science has several regions.
OCI's OSS service ensures high durability by redundantly storing each object across several availability domains or fault domains, depending on the region's infrastructure. It monitors and maintains data integrity, repairing any detected corruption. This system autonomously handles any loss in data redundancy, ensuring no impact on users.

Data Replication plays a crucial role in protecting against the impacts of natural disasters and is essential for disaster recovery and data redundancy compliance. In sectors where compliance and record-keeping regulations are strict, the effective archiving and purging of inactive models become vital. Such practices are key elements of sustainable data management, aiding in controlling storage expenses and upholding data integrity.

With this new feature, you can configure model retention policies to automatically archive older models and save storage. This helps you take control of models in the model catalog. You can also configure model backup, to increase models' redundancy and potential disaster recovery (DR) by automatically backing up models to other regions.

You set these features when Creating and Saving a Model to the Model Catalog.

Model Backup

Some features of model back-up are:

  • You can opt-in to the backup at the time of model creation or during model update operation. When this setting is applied then model catalog service takes the backup of the model in the selected region
  • Backup is per region, to support redundancy.
  • The backup is persisted as long as model isn't deleted or you explicitly opt out from the backup.
  • The backup settings can be applied to new models and existing models.
  • You can opt-out from the model backup at any time. This operation deletes the backed up model artifact.
  • You can delete the model. Upon deleting the model the backup is deleted.

Model Archive and Delete

Some features of model archive and delete are:
  • you can apply the model archival and model delete settings at the time of model creation or during the model update operation.
  • For the model create operation, a minimum of 30 days must be specified for model archival. Model deletion happens after the model is archived. The default value for model deletion is 30 days after model archival setting expires.
  • The model delete setting can only be applied with model archive settings.
  • These settings can be applied to existing models. The number of archival days is to be calculated from the model create time. Archive Model Properties:
    • An Archived model is represented as Lifecycle state → InActive and Lifecycle Details → Archived.
    • No updates are allowed on an Archived model.
    • When a model is archived it can't be transitioned to ACTIVE state.
    • You can delete the archived model.
    • Archived model artifact can be downloaded. To download the artifact, first restore it using the Restore Option and then download the artifact.