Atlas lets you restore data from selected databases or collections from Scheduled or On-Demand Snapshots and Continuous Cloud Backups.
Considerations
Atlas supports restoring from selected databases or collections captured by a given backup. This restore model enables the following:
Selectively restoring tenants in a multi-tenant database architecture where each database or collection maps to a specific tenant.
Targeted recovery of essential data
Selection of specific collections for analytics
Important
Collection- and Database-Level Restore jobs are not intended to reduce RTO and can take longer than full cluster restore jobs.
Atlas supports two write strategies for Collection- and Database-Level Restore jobs. Choose the one best suited to your use case:
Create as new appends the selected databases or collections to the existing data on the target cluster. If a selected database or collection has the same name as one on the target cluster, Atlas renames them to avoid conflict.
Overwrite existing overwrites all selected databases or collections with matching names that already exist on the target cluster. Any writes you make to these databases or collections during the restore job will be lost.
Restoring selected databases or collections temporarily increases disk usage. Atlas blocks restore jobs from starting if there isn't enough disk space available. Consider scaling up cluster storage before starting a restore job.
If disk auto-scaling is enabled, Atlas scales up storage automatically as needed, then restarts in-progress collection restores.
If disk auto-scaling is disabled, the restore job fails and rolls back if disk usage reaches 95%.
If one or more databases or collections fail to restore during a collection- and database-level restore operation, the entire operation fails.
Limitations
Collection- and Database-Level Restore doesn't restore MongoDB Search indexes. You must manually rebuild MongoDB Search indexes after restore.
Atlas doesn't support restoring individual time-series collections.
Atlas doesn't support restoring individual queryable encryption collections.
Atlas doesn't support views for collection- and database-level restore jobs.
You can't run multiple collection- and database-level restore jobs concurrently. You must wait for a running collection- and database-level restore job to finish before initiating another.
Atlas doesn't support collection- and database-level restore on NVMe clusters.
During a collection- and database-level restore job, Atlas transfers data over the public internet even if you have private endpoints configured.
You can't perform collection- and database-level restore with fallback snapshots.
You can specify up to 100 databases and 100 collections per collection- and database-level restore request. Selecting a database restores all its collections without counting them against the 100 collection limit. This limit applies only to namespaces explicitly listed in the request body.
Required Access
To monitor a backup restore job until it completes, you must have Project Read Only access or higher to the specific project.
To start a restore job, you must have Project Backup Recovery Operator, Project Backup Manager, or Project Owner access to the project.
For cross-organization or cross-project restores, the required permissions apply to both the source and target projects. You must have the necessary permissions: Project Backup Manager or Project Owner in both projects to initiate or manage such restore operations.