Atlas Data Federation is a distributed query engine for data stored in Atlas and external sources. A federated database instance exposes that data through virtual databases and collections mapped to the original stores. You can query and analyze combined data without first moving or duplicating it—for example, to examine live cluster data alongside historical data in cloud storage. section 16.0 section 16.1 section 16.2
Supported sources include Atlas clusters, Atlas Online Archive, Amazon S3, Azure Blob Storage, and HTTP or HTTPS endpoints. JSON storage configuration maps sources to virtual databases and collections. Applications can query through the MongoDB Query API; Atlas SQL provides a read-only SQL interface. Behind the instance, a control plane coordinates requests and results, while an elastic compute plane processes requests near the data sources. section 16.1 section 16.2 section 18.1
To deploy an instance in Atlas:
- Open Data Federation and select Create New Federated Database. Name the instance. section 16.3
- Choose one or more sources and the collections to expose. Authorize external access where needed—for S3, this may mean configuring AWS IAM roles. section 16.3
- Set any query or export options, review the configuration, and select Create. You can then run federated queries against the selected sources. section 16.3
Query live and archived data
Atlas Online Archive provides a read-only federated endpoint for both the live cluster and archived data, retaining the cluster’s database and collection names. Separate connection strings can target only the cluster or only the archive. This lets an application query historical records without placing those queries on the live collection. section 17.1 section 17.3
Plan for operational limits rather than treating a federated instance as an ordinary cluster. The stated limits include 60 simultaneous connections per region, 30 simultaneous queries per instance, a 16 MB document ceiling, and no index creation within the instance. Document order is not guaranteed without appropriate query operators, and cluster monitoring tools do not monitor federated instances. section 16.4
Costs depend on data processed and returned, including transfer during query execution. Partitioning can reduce scanned bytes: reading one relevant 1 GB segment of a partitioned 10 GB file is cheaper than processing the entire unpartitioned file. Set query limits where appropriate and account for transfer charges, including when using Atlas SQL. section 16.5 section 17.3 section 18.1
