Supported data sources and file types

  • Article

This article discusses currently supported data sources, file types, and scanning concepts in the Microsoft Purview Data Map.

Microsoft Purview Data Map available data sources

The table below shows all sources that have technical metadata available in Microsoft Purview. Select the data source to learn more. The table also lists additional supported capabilities for each data source, and you can select the feature for more information.

Category Supported data store Classification Live view Lineage Labeling Policies
Azure Multiple sources Yes Limited No Source Dependent Yes
Azure Blob Storage Yes Yes Limited* No Yes (Preview)
Azure Cosmos DB (API for NoSQL) Yes No No* No No
Azure Data Explorer Yes No No* No No
Azure Data Factory No No Yes No No
Azure Data Lake Storage Gen1 Yes No Limited* No No
Azure Data Lake Storage Gen2 Yes Yes Limited* No Yes (Preview)
Azure Database for MySQL Yes No No* No No
Azure Database for PostgreSQL Yes No No* No No
Azure Databricks Hive Metastore No No Yes No No
Azure Databricks Unity Catalog Yes No No No No
Azure Dedicated SQL pool (formerly SQL DW) Yes No No* No No
Azure Files Yes No Limited* No No
Azure Machine Learning No No Yes No No
Azure SQL Database Yes Yes Yes (Preview) No Yes
Azure SQL Managed Instance Yes No No* No Yes
Azure Synapse Analytics (Workspace) Yes No Yes - Synapse pipelines No No
Services and apps Airflow No No Yes No No
Fabric No Yes Yes No No
Power BI No Yes** Yes No No

* Besides the lineage on assets within the data source, lineage is also supported if dataset is used as a source/sink in Data Factory or Synapse pipeline.

** Power BI items in a Fabric tenant are available using live view.

Note

Currently, the Microsoft Purview Data Map can't scan an asset that has /, \, or # in its name. To scope your scan and avoid scanning assets that have those characters in the asset name, use the example in Register and scan an Azure SQL Database.

Important

If you plan on using a self-hosted integration runtime, scanning some data sources requires additional setup on the self-hosted integration runtime machine. For example, JDK, Visual C++ Redistributable, or specific driver. For your source, refer to each source article for prerequisite details. Any requirements will be listed in the Prerequisites section.

Scan regions

The following is a list of all the Azure data source (data center) regions where the Microsoft Purview Data Map scanner runs. If your Azure data source is in a region outside of this list, the scanner will run in the region of your Microsoft Purview instance.

Microsoft Purview Data Map scanner regions

  • China North 3

File types supported for scanning

The following file types are supported for scanning, for schema extraction, and classification where applicable:

  • Structured file formats supported by extension include scanning, schema extraction, and asset and column level classification: AVRO, ORC, PARQUET, CSV, JSON, PSV, SSV, TSV, TXT, XML, GZIP
  • Document file formats supported by extension include scanning and asset level classification: DOC, DOCM, DOCX, DOT, ODP, ODS, ODT, PDF, POT, PPS, PPSX, PPT, PPTM, PPTX, XLC, XLS, XLSB, XLSM, XLSX, XLT
  • The Microsoft Purview Data Map also supports custom file extensions and custom parsers.

Note

Known Limitations:

  • The Microsoft Purview Data Map scanner only supports schema extraction for the structured file types listed above.
  • For AVRO, ORC, and PARQUET file types, the scanner does not support schema extraction for files that contain complex data types (for example, MAP, LIST, STRUCT).
  • The scanner supports scanning snappy compressed PARQUET types for schema extraction and classification.
  • For GZIP file types, the GZIP must be mapped to a single csv file within. Gzip files are subject to System and Custom Classification rules. We currently don't support scanning a gzip file mapped to multiple files within, or any file type other than csv.
  • For delimited file types (CSV, PSV, SSV, TSV, TXT):
    • We do not support data type detection. The data type will be listed as "string" for all columns.
    • We only support comma(‘,’), semicolon(‘;’), vertical bar(‘|’) and tab(‘\t’) as delimiters.
    • Delimited files with less than three rows cannot be determined to be CSV files if they are using a custom delimiter. For example: files with ~ delimiter and less than three rows will not be able to be determined to be CSV files.
    • If a field contains double quotes, the double quotes can only appear at the beginning and end of the field and must be matched. Double quotes that appear in the middle of the field or appear at the beginning and end but are not matched will be recognized as bad data and there will be no schema get parsed from the file. Rows that have different number of columns than the header row will be judged as error rows. (numbers of error rows / numbers of rows sampled ) must be less than 0.1.
  • For Parquet files, if you are using a self-hosted integration runtime, you need to install the 64-bit JRE 11 (Java Runtime Environment) or OpenJDK on your IR machine. Check our Java Runtime Environment section at the bottom of the page for an installation guide.

Schema extraction

For data sources which support schema extraction during scan, the asset schema will not be directly truncated by the number of columns.

Nested data

Currently, nested data is only supported for JSON content.

For all system supported file types, if there's nested JSON content in a column, then the scanner parses the nested JSON data and surfaces it within the schema tab of the asset.

Nested data, or nested schema parsing, isn't supported in SQL. A column with nested data will be reported and classified as is, and subdata won't be parsed.

Sampling data for classification

In Microsoft Purview Data Map terminology,

  • L1 scan: Extracts basic information and meta data like file name, size and fully qualified name
  • L2 scan: Extracts schema for structured file types and database tables
  • L3 scan: Extracts schema where applicable and subjects the sampled file to system and custom classification rules

Learn more about customizing the scan levels.

For all structured file formats, the Microsoft Purview Data Map scanner samples files in the following way:

  • For structured file types, it samples the top 128 rows in each column or the first 1 MB, whichever is lower.
  • For document file formats, it samples the first 20 MB of each file.
    • If a document file is larger than 20 MB, then it isn't subject to a deep scan (subject to classification). In that case, Microsoft Purview captures only basic meta data like file name and fully qualified name.
  • For tabular data sources (SQL), it samples the top 128 rows.
  • For Azure Cosmos DB for NoSQL, up to 300 distinct properties from the first 10 documents in a container will be collected for the schema and for each property, values from up to 128 documents or the first 1 MB will be sampled.

Resource set file sampling

A folder or group of partition files is detected as a resource set in the Microsoft Purview Data Map if it matches with a system resource set policy or a customer defined resource set policy. If a resource set is detected, then the scanner will sample each folder that it contains. Learn more about resource sets here.

File sampling for resource sets by file types:

  • Delimited files (CSV, PSV, SSV, TSV) - 1 in 100 files are sampled (L3 scan) within a folder or group of partition files that are considered a 'Resource set'
  • Data Lake file types (Parquet, Avro, Orc) - 1 in 18446744073709551615 (long max) files are sampled (L3 scan) within a folder or group of partition files that are considered a 'Resource set'
  • Other structured file types (JSON, XML, TXT) - 1 in 100 files are sampled (L3 scan) within a folder or group of partition files that are considered a 'Resource set'
  • SQL objects and Azure Cosmos DB entities - Each file is L3 scanned.
  • Document file types - Each file is L3 scanned. Resource set patterns don't apply to these file types.

Next steps