Edit

Share via

Delta Lake table format interoperability

  • Article

In Microsoft Fabric, the Delta Lake table format is the standard for analytics. Delta Lake is an open-source storage layer that brings ACID (Atomicity, Consistency, Isolation, Durability) transactions to big data and analytics workloads.

All Fabric experiences generate and consume Delta Lake tables, driving interoperability and a unified product experience. Delta Lake tables produced by one compute engine, such as Synapse Data warehouse or Synapse Spark, can be consumed by any other engine, such as Power BI. When you ingest data into Fabric, Fabric stores it as Delta tables by default. You can easily integrate external data containing Delta Lake tables by using OneLake shortcuts.

Delta Lake features and Fabric experiences

To achieve interoperability, all the Fabric experiences align on the Delta Lake features and Fabric capabilities. Some experiences can only write to Delta Lake tables, while others can read from it.

  • Writers: Data warehouses, eventstreams, and exported Power BI semantic models into OneLake
  • Readers: SQL analytics endpoint and Power BI direct lake semantic models
  • Writers and readers: Fabric Spark runtime, dataflows, data pipelines, and Kusto Query Language (KQL) databases

The following matrix shows key Delta Lake features and their support on each Fabric capability.

Fabric capability Name-based column mappings Deletion vectors V-order writing Table optimization and maintenance Write partitions Read partitions Liquid Clustering TIMESTAMP_NTZ Delta reader/writer version and default table features
Data warehouse Delta Lake export No Yes Yes Yes No Yes No No Reader: 3
Writer: 7
Deletion Vectors
SQL analytics endpoint Yes Yes N/A (not applicable) N/A (not applicable) N/A (not applicable) Yes Yes No N/A (not applicable)
Fabric Spark Runtime 1.3 Yes Yes Yes Yes Yes Yes Yes Yes Reader: 1
Writer: 2
Fabric Spark Runtime 1.2 Yes Yes Yes Yes Yes Yes Yes, read only Yes Reader: 1
Writer: 2
Fabric Spark Runtime 1.1 Yes No Yes Yes Yes Yes Yes, read only No Reader: 1
Writer: 2
Dataflows Yes Yes Yes No Yes Yes Yes, read only No Reader: 1
Writer: 2
Data pipelines No No Yes No Yes, overwrite only Yes Yes, read only No Reader: 1
Writer: 2
Power BI direct lake semantic models Yes Yes N/A (not applicable) N/A (not applicable) N/A (not applicable) Yes Yes No N/A (not applicable)
Export Power BI semantic models into OneLake Yes N/A (not applicable) Yes No Yes N/A (not applicable) No No Reader: 2
Writer: 5
KQL databases Yes Yes No No* Yes Yes No No Reader: 1
Writer: 1
Eventstreams No No No No Yes N/A (not applicable) No No Reader: 1
Writer: 2

* KQL databases provide certain table maintenance capabilities such as retention. Data is removed at the end of the retention period from OneLake. For more information, see One Logical copy.

Note

  • Fabric doesn't write name-based column mappings by default. The default Fabric experience generates tables that are compatible across the service. Delta lake, produced by third-party services, may have incompatible table features.
  • Some Fabric experiences do not have inherited table optimization and maintenance capabilities, such as bin-compaction, V-order, and clean up of old unreferenced files. To keep Delta Lake tables optimal for analytics, follow the techniques in Use table maintenance feature to manage delta tables in Fabric for tables ingested using those experiences.

Current limitations

Currently, Fabric doesn't support these Delta Lake features:

  • Delta Lake 3.x Uniform
  • Identity columns writing (proprietary Databricks feature)
  • Delta Live Tables (proprietary Databricks feature)
  • RLE (Run Length Encoding) enabled on the checkpoint file

Related content