CREATE EXTERNAL FILE FORMAT (Transact-SQL)

Applies to: SQL Server 2016 (13.x) and later Azure SQL Managed Instance Azure Synapse Analytics Analytics Platform System (PDW)

Creates an external file format object defining external data stored in Hadoop, Azure Blob Storage, Azure Data Lake Store or for the input and output streams associated with external streams. Creating an external file format is a prerequisite for creating an External Table. By creating an External File Format, you specify the actual layout of the data referenced by an external table. To create an External Table, see CREATE EXTERNAL TABLE (Transact-SQL).

The following file formats are supported:

Syntax

Transact-SQL syntax conventions

-- Create an external file format for DELIMITED (CSV/TSV) files.
CREATE EXTERNAL FILE FORMAT file_format_name
WITH (
        FORMAT_TYPE = DELIMITEDTEXT
    [ , FORMAT_OPTIONS ( <format_options> [ ,...n  ] ) ]
    [ , DATA_COMPRESSION = {
           'org.apache.hadoop.io.compress.GzipCodec'
        }
     ]);

<format_options> ::=
{
    FIELD_TERMINATOR = field_terminator
    | STRING_DELIMITER = string_delimiter
    | FIRST_ROW = integer -- Applies to: Azure Synapse Analytics and SQL Server 2022 and later versions
    | DATE_FORMAT = datetime_format
    | USE_TYPE_DEFAULT = { TRUE | FALSE }
    | ENCODING = {'UTF8' | 'UTF16'}
    | PARSER_VERSION = {'parser_version'}

}

Arguments

file_format_name

Specifies a name for the external file format.

FORMAT_TYPE

Specifies the format of the external data.

  • FORMAT_TYPE = PARQUET

    Specifies a Parquet format.

  • FORMAT_TYPE = ORC

    Specifies an Optimized Row Columnar (ORC) format. This option requires Hive version 0.11 or higher on the external Hadoop cluster. In Hadoop, the ORC file format offers better compression and performance than the RCFILE file format.

  • FORMAT_TYPE = RCFILE, SERDE_METHOD = SERDE_method

    Specifies a Record Columnar file format (RcFile). This option requires you to specify a Hive Serializer and Deserializer (SerDe) method. This requirement is the same if you use Hive/HiveQL in Hadoop to query RC files. Note, the SerDe method is case-sensitive.

    Examples of specifying RCFile with the two SerDe methods that PolyBase supports.

    • FORMAT_TYPE = RCFILE, SERDE_METHOD = 'org.apache.hadoop.hive.serde2.columnar.LazyBinaryColumnarSerDe'
    • FORMAT_TYPE = RCFILE, SERDE_METHOD = 'org.apache.hadoop.hive.serde2.columnar.ColumnarSerDe'
  • FORMAT_TYPE = DELIMITEDTEXT

    Specifies a text format with column delimiters, also called field terminators.

  • FORMAT_TYPE = JSON

    Specifies a JSON format. Applies to Azure SQL Edge only.

  • FORMAT_TYPE = DELTA

    Specifies a Delta Lake format. Applies to serverless SQL pools in Azure Synapse Analytics and SQL Server 2022 (16.x).

FORMAT_OPTIONS

Optional. Only for delimited text data types.

Only serverless SQL pools in Azure Synapse Analytics support PARSER_VERSION.

Serverless SQL pools do not support the DATE_FORMAT option.

DATA_COMPRESSION = data_compression_method

Specifies the data compression method for the external data. When DATA_COMPRESSION isn't specified, the default is uncompressed data.

To work properly, Gzip compressed files must have the ".gz" file extension.

The DELIMITEDTEXT format type supports this compression method:

  • DATA_COMPRESSION = org.apache.hadoop.io.compress.GzipCodec

Delimited text format options

The format options described in this section are optional and only apply to delimited text files.

FIELD_TERMINATOR = field_terminator

Applies only to delimited text files. The field terminator specifies one or more characters that mark the end of each field (column) in the text-delimited file. The default is the pipe character |. For guaranteed support, we recommend using one or more ASCII characters.

Examples:

  • FIELD_TERMINATOR = '|'
  • FIELD_TERMINATOR = ' '
  • FIELD_TERMINATOR = ꞌ\tꞌ
  • FIELD_TERMINATOR = '~|~'

STRING_DELIMITER

STRING_DELIMITER = *string_delimiter*

Specifies the field terminator for data of type string in the text-delimited file. The string delimiter is one or more characters in length and is enclosed with single quotes. The default is the empty string "". For guaranteed support, we recommend using one or more ASCII characters.

Examples:

  • STRING_DELIMITER = '"'

  • STRING_DELIMITER = '0x22' (Double quote hex)

  • STRING_DELIMITER = '*'

  • STRING_DELIMITER = ꞌ,ꞌ

  • STRING_DELIMITER = '0x7E0x7E' (Two tildes, for example, ~~)

FIRST_ROW = first_row_int

Applies to: Azure Synapse Analytics, SQL Server 2022 and later versions

Specifies the row number that is read first in all files during a PolyBase load. This parameter can take values 1 - 15. If the value is set to two, the first row in every file (header row) is skipped when the data is loaded. Rows are skipped based on the existence of row terminators (/r/n, /r, /n). When this option is used for export, rows are added to the data to make sure the file can be read with no data loss. If the value is set to >2, the first row exported is the Column names of the external table.

DATE_FORMAT = datetime_format

Specifies a custom format for all date and time data that might appear in a delimited text file. If the source file uses default datetime formats, this option isn't necessary. Only one custom datetime format is allowed per file. You can't specify more than one custom datetime formats per file. However, you can use more than one datetime formats if each one is the default format for its respective data type in the external table definition.

Important

PolyBase only uses the custom date format for importing the data. It doesn't use the custom format for writing data to an external file.

When DATE_FORMAT isn't specified or is the empty string, PolyBase uses the following default formats:

  • datetime: 'yyyy-MM-dd HH:mm:ss'

  • smalldatetime: 'yyyy-MM-dd HH:mm'

  • date: 'yyyy-MM-dd'

  • datetime2: 'yyyy-MM-dd HH:mm:ss'

  • datetimeoffset: 'yyyy-MM-dd HH:mm:ss'

  • time: 'HH:mm:ss'

Important

Specifying custom DATE_FORMAT will override all default type formats. This means that you will need to have the same date formats in all datetime, date, and time cells in your files. With the overridden DATE_FORMAT you cannot have date and time values in different format.

Example date formats are in the following table:

Notes about the table:

  • Year, month, and day can have several formats and orders. The table shows only the ymd format. Month can have one or two digits, or three characters. Day can have one or two digits. Year can have two or four digits.

  • Milliseconds (fffffff) aren't required.

  • am or pm (tt) isn't required. The default is AM.

Date Type Example Description
datetime DATE_FORMAT = yyyy-MM-dd HH:mm:ss.fff In addition to year, month and day, this date format includes 00-24 hours, 00-59 minutes, 00-59 seconds, and 3 digits for milliseconds.
datetime DATE_FORMAT = yyyy-MM-dd hh:mm:ss.ffftt In addition to year, month and day, this date format includes 00-12 hours, 00-59 minutes, 00-59 seconds, 3 digits for milliseconds, and AM, am, PM, or pm.
smalldatetime DATE_FORMAT = yyyy-MM-dd HH:mm In addition to year, month, and day, this date format includes 00-23 hours, 00-59 minutes.
smalldatetime DATE_FORMAT = yyyy-MM-dd hh:mmtt In addition to year, month, and day, this date format includes 00-11 hours, 00-59 minutes, no seconds, and AM, am, PM, or pm.
date DATE_FORMAT = yyyy-MM-dd Year, month, and day. No time element is included.
date DATE_FORMAT = yyyy-MMM-dd Year, month, and day. When month is specified with MMM, the input value is one or the strings, Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, or Dec.
datetime2 DATE_FORMAT = yyyy-MM-dd HH:mm:ss.fffffff In addition to year, month, and day, this date format includes 00-23 hours, 00-59 minutes, 00-59 seconds, and 7 digits for milliseconds.
datetime2 DATE_FORMAT = yyyy-MM-dd hh:mm:ss.ffffffftt In addition to year, month, and day, this date format includes 00-11 hours, 00-59 minutes, 00-59 seconds, 7 digits for milliseconds, and AM, am, PM, or pm.
datetimeoffset DATE_FORMAT = yyyy-MM-dd HH:mm:ss.fffffff zzz In addition to year, month, and day, this date format includes 00-23 hours, 00-59 minutes, 00-59 seconds, and 7 digits for milliseconds, and the timezone offset which you put in the input file as {+&#124;-}HH:ss. For example, since Los Angeles time without daylight savings is 8 hours behind UTC, a value of -08:00 in the input file specifies the timezone for Los Angeles.
datetimeoffset DATE_FORMAT = yyyy-MM-dd hh:mm:ss.ffffffftt zzz In addition to year, month, and day, this date format includes 00-11 hours, 00-59 minutes, 00-59 seconds, 7 digits for milliseconds, (AM, am, PM, or pm), and the timezone offset. See the description in the previous row.
time DATE_FORMAT = HH:mm:ss There is no date value, only 00-23 hours, 00-59 minutes, and 00-59 seconds.

Supported date and time formats

External file format can describe a large number of date and time formats:

datetime smalldatetime date datetime2 datetimeoffset
[M[M]]M-[d]d-[yy]yy HH:mm:ss[.fff] [M[M]]M-[d]d-[yy]yy HH:mm[:00] [M[M]]M-[d]d-[yy]yy [M[M]]M-[d]d-[yy]yy HH:mm:ss[.fffffff] [M[M]]M-[d]d-[yy]yy HH:mm:ss[.fffffff] zzz
[M[M]]M-[d]d-[yy]yy hh:mm:ss[.fff][tt] [M[M]]M-[d]d-[yy]yy hh:mm[:00][tt] [M[M]]M-[d]d-[yy]yy hh:mm:ss[.fffffff][tt] [M[M]]M-[d]d-[yy]yy hh:mm:ss[.fffffff][tt] zzz
[M[M]]M-[yy]yy-[d]d HH:mm:ss[.fff] [M[M]]M-[yy]yy-[d]d HH:mm[:00] [M[M]]M-[yy]yy-[d]d [M[M]]M-[yy]yy-[d]d HH:mm:ss[.fffffff] [M[M]]M-[yy]yy-[d]d HH:mm:ss[.fffffff] zzz
[M[M]]M-[yy]yy-[d]d hh:mm:ss[.fff][tt] [M[M]]M-[yy]yy-[d]d hh:mm[:00][tt] [M[M]]M-[yy]yy-[d]d hh:mm:ss[.fffffff][tt] [M[M]]M-[yy]yy-[d]d hh:mm:ss[.fffffff][tt] zzz
[yy]yy-[M[M]]M-[d]d HH:mm:ss[.fff] [yy]yy-[M[M]]M-[d]d HH:mm[:00] [yy]yy-[M[M]]M-[d]d [yy]yy-[M[M]]M-[d]d HH:mm:ss[.fffffff] [yy]yy-[M[M]]M-[d]d HH:mm:ss[.fffffff] zzz
[yy]yy-[M[M]]M-[d]d hh:mm:ss[.fff][tt] [yy]yy-[M[M]]M-[d]d hh:mm[:00][tt] [yy]yy-[M[M]]M-[d]d hh:mm:ss[.fffffff][tt] [yy]yy-[M[M]]M-[d]d hh:mm:ss[.fffffff][tt] zzz
[yy]yy-[d]d-[M[M]]M HH:mm:ss[.fff] [yy]yy-[d]d-[M[M]]M HH:mm[:00] [yy]yy-[d]d-[M[M]]M [yy]yy-[d]d-[M[M]]M HH:mm:ss[.fffffff] [yy]yy-[d]d-[M[M]]M HH:mm:ss[.fffffff] zzz
[yy]yy-[d]d-[M[M]]M hh:mm:ss[.fff][tt] [yy]yy-[d]d-[M[M]]M hh:mm[:00][tt] [yy]yy-[d]d-[M[M]]M hh:mm:ss[.fffffff][tt] [yy]yy-[d]d-[M[M]]M hh:mm:ss[.fffffff][tt] zzz
[d]d-[M[M]]M-[yy]yy HH:mm:ss[.fff] [d]d-[M[M]]M-[yy]yy HH:mm[:00] [d]d-[M[M]]M-[yy]yy [d]d-[M[M]]M-[yy]yy HH:mm:ss[.fffffff] [d]d-[M[M]]M-[yy]yy HH:mm:ss[.fffffff] zzz
[d]d-[M[M]]M-[yy]yy hh:mm:ss[.fff][tt] [d]d-[M[M]]M-[yy]yy hh:mm[:00][tt] [d]d-[M[M]]M-[yy]yy hh:mm:ss[.fffffff][tt] [d]d-[M[M]]M-[yy]yy hh:mm:ss[.fffffff][tt] zzz
[d]d-[yy]yy-[M[M]]M HH:mm:ss[.fff] [d]d-[yy]yy-[M[M]]M HH:mm[:00] [d]d-[yy]yy-[M[M]]M [d]d-[yy]yy-[M[M]]M HH:mm:ss[.fffffff] [d]d-[yy]yy-[M[M]]M HH:mm:ss[.fffffff] zzz
[d]d-[yy]yy-[M[M]]M hh:mm:ss[.fff][tt] [d]d-[yy]yy-[M[M]]M hh:mm[:00][tt] [d]d-[yy]yy-[M[M]]M hh:mm:ss[.fffffff][tt] [d]d-[yy]yy-[M[M]]M hh:mm:ss[.fffffff][tt] zzz

Details:

  • To separate month, day and year values, you must use one of the following separators: -, /, or .. For simplicity, the table uses only the - separator.

  • To specify the month as text, use three or more characters. Months with one or two characters are interpreted as a number.

  • To separate time values, use the : symbol.

  • Letters enclosed in square brackets are optional.

  • The letters tt designate [AM|PM|am|pm]. AM is the default. When tt is specified, the hour value (hh) must be in the range of 0 to 12.

  • The letters zzz designate the time zone offset for the system's current time zone in the format {+|-}HH:ss].

USE_TYPE_DEFAULT = { TRUE | FALSE }

Specifies how to handle missing values in delimited text files when PolyBase retrieves data from the text file. The default value is FALSE.

  • TRUE

    When retrieving data from the text file, store each missing value by using the default value for the data type of the corresponding column in the external table definition. For example, replace a missing value with:

    • 0 if the column is defined as a numeric column. Decimal columns aren't supported and will error.

    • Empty string "" if the column is a string column.

    • 1900-01-01 if the column is a date column.

    • In Azure Synapse Analytics, USE_TYPE_DEFAULT=true is not supported for FORMAT_TYPE = DELIMITEDTEXT, PARSER_VERSION = '2.0'.

  • FALSE

    Store all missing values as NULL. Any NULL values that are stored by using the word NULL in the delimited text file are imported as the string NULL.

ENCODING = {'UTF8' | 'UTF16'}

In Azure Synapse Analytics and Analytics Platform System (PDW) (APS CU7.4), PolyBase can read UTF8 and UTF16-LE encoded delimited text files.

In SQL Server, PolyBase doesn't support reading UTF16 encoded files.

Permissions

Requires ALTER ANY EXTERNAL FILE FORMAT permission.

Remarks

The external file format is database-scoped in SQL Server and Azure Synapse Analytics. It is server-scoped in Analytics Platform System (PDW).

The format options are all optional and only apply to delimited text files.

When the data is stored in one of the compressed formats, PolyBase first decompresses the data before returning the data records.

Limitations

The row delimiter in delimited-text files must be supported by Hadoop's LineRecordReader. That is, it must be either \r, \n, or \r\n. These delimiters aren't user-configurable.

The combinations of supported SerDe methods with RCFiles, and the supported data compression methods are listed previously in this article. Not all combinations are supported.

The maximum number of concurrent PolyBase queries is 32. When 32 concurrent queries are running, each query can read a maximum of 33,000 files from the external file location. The root folder and each subfolder also count as a file. If the degree of concurrency is less than 32, the external file location can contain more than 33,000 files.

Because of the limitation on number of files in the external table, we recommend storing less than 30,000 files in the root and subfolders of the external file location. Also, we recommend keeping the number of subfolders under the root directory to a small number. When too many files are referenced, a Java Virtual Machine out-of-memory exception might occur.

When exporting data to Hadoop or Azure Blob Storage via PolyBase, only the data is exported, not the column names(metadata) as defined in the CREATE EXTERNAL TABLE command.

Locking

Takes a shared lock on the EXTERNAL FILE FORMAT object.

Performance

Using compressed files always comes with the tradeoff between transferring less data between the external data source and SQL Server while increasing the CPU usage to compress and decompress the data.

Gzip compressed text files aren't splittable. To improve performance for Gzip compressed text files, we recommend generating multiple files that are all stored in the same directory within the external data source. This file structure allows PolyBase to read and decompress the data faster by using multiple reader and decompression processes. The ideal number of compressed files is the maximum number of data reader processes per compute node. In SQL Server and Analytics Platform System (PDW), the maximum number of data reader processes is 8 per node except Azure Synapse Analytics Gen2, which is 20 readers per node. In Azure Synapse Analytics, the maximum number of data reader processes per node varies by SLO. See Azure Synapse Analytics loading patterns and strategies for details.

Examples

A. Create a DELIMITEDTEXT external file format

This example creates an external file format named textdelimited1 for a text-delimited file. The options listed for FORMAT_OPTIONS specify that the fields in the file should be separated using a pipe character |. The text file is also compressed with the Gzip codec. If DATA_COMPRESSION isn't specified, the text file is uncompressed.

For a delimited text file, the data compression method can either be the default Codec, org.apache.hadoop.io.compress.DefaultCodec, or the Gzip Codec, org.apache.hadoop.io.compress.GzipCodec.

CREATE EXTERNAL FILE FORMAT textdelimited1
WITH (
    FORMAT_TYPE = DELIMITEDTEXT,
    FORMAT_OPTIONS (
        FIELD_TERMINATOR = '|',
        DATE_FORMAT = 'MM/dd/yyyy' ),
    DATA_COMPRESSION = 'org.apache.hadoop.io.compress.GzipCodec'
);

B. Create an RCFile external file format

This example creates an external file format for a RCFile that uses the serialization/deserialization method org.apache.hadoop.hive.serde2.columnar.LazyBinaryColumnarSerDe. It also specifies to use the Default Codec for the data compression method. If DATA_COMPRESSION isn't specified, the default is no compression.

CREATE EXTERNAL FILE FORMAT rcfile1
WITH (
    FORMAT_TYPE = RCFILE,
    SERDE_METHOD = 'org.apache.hadoop.hive.serde2.columnar.LazyBinaryColumnarSerDe',
    DATA_COMPRESSION = 'org.apache.hadoop.io.compress.DefaultCodec'
);

C. Create an ORC external file format

This example creates an external file format for an ORC file that compresses the data with the org.apache.io.compress.SnappyCodec data compression method. If DATA_COMPRESSION isn't specified, the default is no compression.

CREATE EXTERNAL FILE FORMAT orcfile1
WITH (
    FORMAT_TYPE = ORC,
    DATA_COMPRESSION = 'org.apache.hadoop.io.compress.SnappyCodec'
);

D. Create a PARQUET external file format

This example creates an external file format for a Parquet file that compresses the data with the org.apache.io.compress.SnappyCodec data compression method. If DATA_COMPRESSION isn't specified, the default is no compression.

CREATE EXTERNAL FILE FORMAT parquetfile1
WITH (
    FORMAT_TYPE = PARQUET,
    DATA_COMPRESSION = 'org.apache.hadoop.io.compress.SnappyCodec'
);

E. Create a delimited text file skipping header row

Applies to: Azure Synapse Analytics, and SQL Server 2022 (16.x) and later.

This example creates an external file format for CSV file with a single header row. For more information, see Virtualize CSV file with PolyBase.

CREATE EXTERNAL FILE FORMAT skipHeader_CSV
WITH (FORMAT_TYPE = DELIMITEDTEXT,
      FORMAT_OPTIONS(
          FIELD_TERMINATOR = ',',
          STRING_DELIMITER = '"',
          FIRST_ROW = 2,
          USE_TYPE_DEFAULT = True)
);

F. Create a JSON external file format

Applies to: Azure SQL Edge.

This example creates an external file format for a JSON file that compresses the data with the org.apache.io.compress.SnappyCodec data compression method. If DATA_COMPRESSION isn't specified, the default is no compression. This example applies to Azure SQL Edge and is currently not supported for other SQL products.

CREATE EXTERNAL FILE FORMAT jsonFileFormat
WITH (
    FORMAT_TYPE = JSON,
    DATA_COMPRESSION = 'org.apache.hadoop.io.compress.SnappyCodec'
);

G. Create a Delta table external file format

This example creates an external file format for Delta table type file format. This example applies to SQL Server 2022 (16.x). For more information, see Virtualize delta table file with PolyBase.

CREATE EXTERNAL FILE FORMAT DeltaFileFormat
WITH (
    FORMAT_TYPE = DELTA
);