Skip to main content
Version: 1.6

SQL Server backup files → Parquet

Datatypes conversion mapping when exporting a table from SQL Server backups (.bak, full and differential) using the ArpeMSBackup ADBC driver (--connectiontype backup_mssql) to Parquet format.

Platform availability

backup_mssql is available on Windows x64 and Linux x64 only. It is not supported on Linux ARM64.

Compared with adbc_mssql

Since FastBCP 1.6.5, the mapping and the values are those of adbc_mssql, so a backup and a live export of the same table give the same Parquet. The exception is json and vector(n, float16) (SQL Server 2025): backup_mssql refuses them.

Changed in FastBCP 1.6.5
  • time(7): microsecond unit, instead of nanosecond.
  • geography, geometry: WKB with GeoParquet metadata, instead of the SQL Server binary serialization.
  • vector(n): list of 32-bit floats, instead of the SQL Server binary serialization.
  • sql_variant: values are written as with adbc_mssql.
SourceSQL Server backup file (backup_mssql)
FormatParquet
FastBCP1.6.4.0; time, geography, geometry, uniqueidentifier, json and vector: 1.6.5.0
Test StatusPARTIAL (32/33 Nullable) — json refused. SQL Server 2022 and 2025 backups: uncompressed, striped, MS_XPRESS and ZSTD compression, --parallelmethod Physical

Parquet Schema: testdatatypes_nullable​

namephysical_typeconverted_typelogical_type
col_bigintINT64——
col_intINT32——
col_smallintINT32INT_16IntType(bitWidth=16, isSigned=1)
col_tinyintINT32UINT_8IntType(bitWidth=8, isSigned=0)
col_bitBOOLEAN——
col_decimalFIXED_LEN_BYTE_ARRAYDECIMALDecimalType(scale=2, precision=18)
col_numericFIXED_LEN_BYTE_ARRAYDECIMALDecimalType(scale=2, precision=18)
col_moneyFIXED_LEN_BYTE_ARRAYDECIMALDecimalType(scale=4, precision=19)
col_smallmoneyFIXED_LEN_BYTE_ARRAYDECIMALDecimalType(scale=4, precision=10)
col_floatDOUBLE——
col_realFLOAT——
col_dateINT32DATEDateType()
col_timeINT64—TimeType(isAdjustedToUTC=0, unit=MICROS)
col_datetimeINT64TIMESTAMP_MILLISTimestampType(isAdjustedToUTC=0, unit=MILLIS)
col_datetime2INT64TIMESTAMP_MICROSTimestampType(isAdjustedToUTC=0, unit=MICROS)
col_datetimeoffsetINT64TIMESTAMP_MICROSTimestampType(isAdjustedToUTC=1, unit=MICROS)
col_smalldatetimeINT64TIMESTAMP_MILLISTimestampType(isAdjustedToUTC=0, unit=MILLIS)
col_charBYTE_ARRAYUTF8StringType()
col_varcharBYTE_ARRAYUTF8StringType()
col_ncharBYTE_ARRAYUTF8StringType()
col_nvarcharBYTE_ARRAYUTF8StringType()
col_textBYTE_ARRAYUTF8StringType()
col_ntextBYTE_ARRAYUTF8StringType()
col_binaryBYTE_ARRAY——
col_varbinaryBYTE_ARRAY——
col_imageBYTE_ARRAY——
col_uniqueidentifierBYTE_ARRAYUTF8StringType()
col_xmlBYTE_ARRAYUTF8StringType()
col_geographyBYTE_ARRAY—— (WKB, GeoParquet geo metadata, spherical edges)
col_geometryBYTE_ARRAY—— (WKB, GeoParquet geo metadata)
col_sql_variantBYTE_ARRAYUTF8StringType()
col_json——Refused
col_vectorFLOAT (repeated)LISTListType()

col_json (json) and col_vector (vector(3)) are SQL Server 2025 types, tested with FastBCP 1.6.5.0.

  • vector(n) is exported as a list of n 32-bit floats ([1.0, -2.5, 3.25]).
  • geography and geometry keep their coordinate system: the GeoParquet metadata carries the CRS from the column's SRID.
  • uniqueidentifier is written in lowercase by default. Since FastBCP 1.6.5, FASTBCP_UID_STRING_CASING applies to backup_mssql too.

To get geography and geometry another way, give the connection as a connection string: adbc.arpemsbackup.geospatial=wkb writes plain WKB without the metadata, and varbinary writes the SQL Server binary serialization, as before 1.6.5.

--sourceconnectstring "adbc.arpemsbackup.path=D:\Backups\MSSQL;adbc.arpemsbackup.file=mydb.bak;adbc.arpemsbackup.geospatial=varbinary"
json and vector(n, float16) columns are refused

A table with such a column is refused with a message naming the column (column j: json (SQL Server 2025) is not supported). Export the other columns by listing them in --query, or export the table with adbc_mssql.

Copyright © 2026 Architecture & Performance.