# Azure Synapse Atlas Data Federation > Source-anchored research for Azure Synapse Atlas Data Federation. Run status: COMPLETED. Parent: [MongoDB Atlas on Azure](https://llms-explorer.com/tree/mongodb-atlas-on-azure/) · 6 facets · 18 facts · page: https://llms-explorer.com/tree/azure-synapse-atlas-data-federation/ ## Definitions - Atlas Data Federation is a distributed query engine that queries, transforms, and moves data across Atlas and external stores, including Azure Blob Storage; it can materialize aggregation results into cloud object storage. (high confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/adf-overview/overview.md) - Parquet output is not a lossless generic BSON interchange: Data Federation reads null and undefined values back as missing, represents polymorphic BSON fields as groups of typed columns, and stores MongoDB schema metadata to reconstruct BSON-specific types on its own reads. (low confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/supported-unsupported/pipeline/out.md) ## Structure and components - Architecture inference: export Atlas data as Parquet to Azure Blob Storage with Data Federation, then read those files using Synapse serverless SQL OPENROWSET with FORMAT='PARQUET'. The component documentation supports a storage-mediated integration; it does not demonstrate Synapse directly querying a Data Federation endpoint. (high confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/adf-overview/overview.md) - Atlas Data Federation writes arrays to Parquet using the LIST logical type and three-level list/element structure; this follows Apache Parquet's specification for new list writers. Nested BSON therefore requires deliberate downstream schema handling. (high confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/supported-unsupported/pipeline/out.md) ## How it works - Data Federation's Parquet writer maps a polymorphic BSON field into a group of typed child columns. Synapse OPENROWSET supports an explicit WITH schema. Design implication: validate the generated Parquet schema and SQL mappings when document fields have mixed types. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/supported-unsupported/pipeline/out.md) - Microsoft warns that Parquet string encoding and SQL collation mismatches can cause conversion errors. Its Synapse guide recommends UTF-8 collation and explains that Latin1_General_100_BIN2_UTF8 enables string-related file or column-segment pruning. (low confidence) — [source](https://learn.microsoft.com/en-us/azure/synapse-analytics/sql/query-parquet-files) - A Microsoft repository demonstrates collection change streams through Atlas triggers and functions, an Azure Function, and Synapse ADLS Gen2. The repository was archived on 2024-07-22; it is evidence of a custom historical integration, not proof of a currently maintained managed Data Federation CDC feature. (low confidence) (verified as of 2024-07-22) — [source](https://github.com/Azure/Azure_Synapse_RealTimeSync_Using_AtlasTrigger_and_AzureFunction) - OutageDeck reproduces a MongoDB status incident in which Atlas Data Federation could not read customer Azure Blob containers on 2026-03-19. Investigation began at 18:11 UTC and resolution was recorded at 19:19 UTC; the published updates give no technical root cause. (low confidence) — [source](https://outagedeck.com/incidents/mongodb-atlas-issue-querying-azure-blob-storage-containers-in-atlas-data-federation-2026-03-19) - Microsoft archived the Atlas Trigger/Azure Function Synapse example repository on 2024-07-22; its existence should not be interpreted as evidence of a maintained production integration in 2026. (low confidence) (verified as of 2026-10-24) — [source](https://github.com/Azure/Azure_Synapse_RealTimeSync_Using_AtlasTrigger_and_AzureFunction) - Synapse serverless SQL infers a multi-file Parquet schema from the first file it reads, which can omit expected columns. Microsoft recommends an explicit OPENROWSET WITH schema when inference is insufficient. (low confidence) — [source](https://learn.microsoft.com/en-us/azure/synapse-analytics/sql/query-parquet-files) ## Problems, failure modes and limitations - Azure Blob Storage queries require a federated database instance created in Azure. MongoDB recommends colocating federation with its source's cloud and region to limit transfer charges; the book independently documents the same placement restriction. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/atlas/data-federation/adf-overview/regions.md) ## Comparisons and alternatives - The archived Microsoft real-time sample sends Atlas change events through an Atlas Trigger and Function to an Azure Function, which writes to the Synapse workspace's ADLS Gen2 storage. This code path uses event forwarding rather than Data Federation. (low confidence) — [source](https://github.com/Azure/Azure_Synapse_RealTimeSync_Using_AtlasTrigger_and_AzureFunction) ## Facts and statements - Atlas Data Federation exposes virtual databases and collections over Atlas and external stores, including Azure Blob Storage, and supports querying and transforming these sources with MQL. (high confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/adf-overview/overview.md) - Synapse pipelines provide a separate direct-copy route: the MongoDB Atlas connector supports Copy Activity as both source and sink. Microsoft's sample describes these connector transfers as batch or microbatch operations. (high confidence) (verified as of 2026-03-25) — [source](https://learn.microsoft.com/en-us/azure/data-factory/connector-mongodb-atlas) - The direct-copy connector's linked service uses type MongoDbAtlas, a connectionString and database; the dataset uses MongoDbAtlasCollection and collectionName. Microsoft documents storing the connection string in Azure Key Vault. (low confidence) — [source](https://learn.microsoft.com/en-us/azure/data-factory/connector-mongodb-atlas) - MongoDB documents Data Federation $out as a way to make point-in-time copies in cloud object storage. Export scheduling and downstream consumption therefore need an explicit freshness plan; this export pattern alone does not establish continuous change-data capture. (high confidence) — [source](https://www.mongodb.com/docs/atlas/data-federation/data-lake-deprecation.md) - A Savonia University of Applied Sciences thesis dated 2024-12-16 independently evaluated Atlas Data Federation with Azure Blob Storage for Gebwell's measurement-data archiving needs and compared it with Online Archive. Its abstract reports a working archival proof of concept, but does not establish that the chosen implementation used Data Federation or validate Synapse interoperability. (low confidence) — [source](https://www.theseus.fi/bitstream/10024/875839/2/Kroger_Jani.pdf) - Synapse Parquet readers should use UTF-8 collation for string columns: Microsoft warns of conversion errors with mismatched collation, while Apache defines the Parquet STRING annotation as UTF-8. (high confidence) — [source](https://learn.microsoft.com/en-us/azure/synapse-analytics/sql/query-parquet-files)