read_files table-valued function

Applies to: Databricks SQL Databricks Runtime 13.1 and later

Reads files under a provided location and returns the data in tabular form.

Supports reading JSON, CSV, TEXT, BINARYFILE, PARQUET, AVRO, and ORC file formats. Can detect the file format automatically and infer a unified schema across all files.

Syntax

read_files(path [, option_key => option_value ] [...])

Arguments

This function requires named parameter invocation for the option keys.

• path: A STRING with the URI of the location of the data. Supports reading from Azure Data Lake Storage Gen2 ('abfss://'), S3 (s3://) and Google Cloud Storage ('gs://'). Can contain globs. See File discovery for more details.

• option_key: The name of the option to configure. You need to use backticks (`) for options that contain dots (.).

• option_value: A constant expression to set the option to. Accepts literals and scalar functions.

Returns

A table comprised of the data from files read under the given path.

File discovery

read_files can read an individual file or read files under a provided directory. read_files discovers all files under the provided directory recursively unless a glob is provided, which instructs read_files to recurse into a specific directory pattern.

Filtering directories or files using glob patterns

Glob patterns can be used for filtering directories and files when provided in the path.

Pattern	Description
?	Matches any single character
*	Matches zero or more characters
[abc]	Matches a single character from character set {a,b,c}.
[a-z]	Matches a single character from the character range {a…z}.
[^a]	Matches a single character that is not from character set or range {a}. Note that the ^ character must occur immediately to the right of the opening bracket.
{ab,cd}	Matches a string from the string set {ab, cd}.
{ab,c{de, fh}}	Matches a string from the string set {ab, cde, cfh}.

read_files uses Auto Loader’s strict globber when discovering files with globs. This is configured by the useStrictGlobber option. When the strict globber is disabled, trailing slashes (/) are dropped and a star pattern such as /*/ can expand into discovering multiple directories. See the examples below to see the difference in behavior.

Pattern	File path	Strict globber disabled	Strict globber enabled
/a/b	/a/b/c/file.txt	Yes	Yes
/a/b	/a/b_dir/c/file.txt	No	No
/a/b	/a/b.txt	No	No
/a/b/	/a/b.txt	No	No
/a/*/c/	/a/b/c/file.txt	Yes	Yes
/a/*/c/	/a/b/c/d/file.txt	Yes	Yes
/a/*/d/	/a/b/c/d/file.txt	Yes	No
/a/*/c/	/a/b/x/y/c/file.txt	Yes	No
/a/*/c	/a/b/c_file.txt	Yes	No
/a/*/c/	/a/b/c_file.txt	Yes	No
/a/*/c	/a/b/cookie/file.txt	Yes	No
/a/b*	/a/b.txt	Yes	Yes
/a/b*	/a/b/file.txt	Yes	Yes
/a/{0.txt,1.txt}	/a/0.txt	Yes	Yes
/a/*/{0.txt,1.txt}	/a/0.txt	No	No
/a/b/[cde-h]/i/	/a/b/c/i/file.txt	Yes	Yes

Schema inference

The schema of the files can be explicitly provided to read_files with the schema option. When the schema is not provided, read_files attempts to infer a unified schema across the discovered files, which requires reading all the files unless a LIMIT statement is used. Even when using a LIMIT query, a larger set of files than required might be read to return a more representative schema of the data. Databricks automatically adds a LIMIT statement for SELECT queries in notebooks and the SQL editor if a user hasn’t provided one.

The schemaHints option can be used to fix subsets of the inferred schema. See Override schema inference with schema hints for more details.

A rescuedDataColumn is provided by default to rescue any data that doesn’t match the schema. See What is the rescued data column? for more details. You can drop the rescuedDataColumn by setting the option schemaEvolutionMode => 'none'.

Partition schema inference

read_files can also infer partitioning columns if files are stored under Hive-style partitioned directories, that is /column_name=column_value/. If a schema is provided, the discovered partition columns leverage the types provided in the schema. If the partition columns are not part of the provided schema, then the inferred partition columns are ignored.

If a column exists in both the partition schema and in the data columns, the value that is read from the partition value is used instead of the data value. If you would like to ignore the values coming from the directory and use the data column, you can provide the list of partition columns in a comma-separated list with the partitionColumns option.

The partitionColumns option can also be used to instruct read_files on which discovered columns to include in the final inferred schema. Providing an empty string ignores all partition columns.

The schemaHints option can also be provided to override the inferred schema for a partition column.

The TEXT and BINARYFILE formats have a fixed schema, but read_files also attempts to infer partitioning for these formats when possible.

Usage in streaming tables

read_files can be used in streaming tables to ingest files into Delta Lake. read_files leverages Auto Loader when used in a streaming table query. You must use the STREAM keyword with read_files. See What is Auto Loader? for more details.

When used in a streaming query, read_files will use a sample of the data to infer the schema, and can evolve the schema on-the-fly as it processes more data. See Configure schema inference and evolution in Auto Loader for more details.

Options

Basic Options

Option
format Type: String The data file format in the source path. Auto-inferred if not provided. Allowed values include: avro: Avro file binaryFile: Binary file csv: CSV file json: JSON file orc: ORC file parquet: Read Parquet files using Databricks text: Text files Default value: None
inferColumnTypes Type: Boolean Whether to infer exact column types when leveraging schema inference. By default, columns are inferred when inferring JSON and CSV datasets. See schema inference for more details. Note that this is the opposite of the default of Auto Loader. Default value: true
partitionColumns Type: String A comma separated list of Hive style partition columns that you would like inferred from the directory structure of the files. Hive style partition columns are key value pairs combined by an equality sign such as <base-path>/a=x/b=1/c=y/file.format. In this example, the partition columns are a, b, and c. By default these columns will be automatically added to your schema if you are using schema inference and provide the <base-path> to load data from. If you provide a schema, Auto Loader expects these columns to be included in the schema. If you do not want these columns as part of your schema, you can specify "" to ignore these columns. In addition, you can use this option when you want columns to be inferred the file path in complex directory structures, like the example below: <base-path>/year=2022/week=1/file1.csv <base-path>/year=2022/month=2/day=3/file2.csv <base-path>/year=2022/month=2/day=4/file3.csv Specifying cloudFiles.partitionColumns as year,month,day will return year=2022 for file1.csv, but the month and day columns will be null. month and day will be parsed correctly for file2.csv and file3.csv. Default value: None
schemaHints Type: String Schema information that you provide to Auto Loader during schema inference. See schema hints for more details. Default value: None
useStrictGlobber Type: Boolean Whether to use a strict globber that matches the default globbing behavior of other file sources in Apache Spark. See Common data loading patterns for more details. Available in Databricks Runtime 12.0 and above. Note that this is the opposite of the default for Auto Loader. Default value: true

Generic options

The following options apply to all file formats.

Option
ignoreCorruptFiles Type: Boolean Whether to ignore corrupt files. If true, the Spark jobs will continue to run when encountering corrupted files and the contents that have been read will still be returned. Observable as numSkippedCorruptFiles in the operationMetrics column of the Delta Lake history. Available in Databricks Runtime 11.0 and above. Default value: false
ignoreMissingFiles Type: Boolean Whether to ignore missing files. If true, the Spark jobs will continue to run when encountering missing files and the contents that have been read will still be returned. Available in Databricks Runtime 11.0 and above. Default value: false (true for COPY INTO)
modifiedAfter Type: Timestamp String, for example, 2021-01-01 00:00:00.000000 UTC+0 An optional timestamp to ingest files that have a modification timestamp after the provided timestamp. Default value: None
modifiedBefore Type: Timestamp String, for example, 2021-01-01 00:00:00.000000 UTC+0 An optional timestamp to ingest files that have a modification timestamp before the provided timestamp. Default value: None
pathGlobFilter or fileNamePattern Type: String A potential glob pattern to provide for choosing files. Equivalent to PATTERN in COPY INTO. fileNamePattern can be used in read_files. Default value: None
recursiveFileLookup Type: Boolean Whether to load data recursively within the base directory and skip partition inference. Default value: false

JSON options

Option
allowBackslashEscapingAnyCharacter Type: Boolean Whether to allow backslashes to escape any character that succeeds it. If not enabled, only characters that are explicitly listed by the JSON specification can be escaped. Default value: false
allowComments Type: Boolean Whether to allow the use of Java, C, and C++ style comments ('/', '*', and '//' varieties) within parsed content or not. Default value: false
allowNonNumericNumbers Type: Boolean Whether to allow the set of not-a-number (NaN) tokens as legal floating number values. Default value: true
allowNumericLeadingZeros Type: Boolean Whether to allow integral numbers to start with additional (ignorable) zeroes (for example, 000001). Default value: false
allowSingleQuotes Type: Boolean Whether to allow use of single quotes (apostrophe, character '\') for quoting strings (names and String values). Default value: true
allowUnquotedControlChars Type: Boolean Whether to allow JSON strings to contain unescaped control characters (ASCII characters with value less than 32, including tab and line feed characters) or not. Default value: false
allowUnquotedFieldNames Type: Boolean Whether to allow use of unquoted field names (which are allowed by JavaScript, but not by the JSON specification). Default value: false
badRecordsPath Type: String The path to store files for recording the information about bad JSON records. Default value: None
columnNameOfCorruptRecord Type: String The column for storing records that are malformed and cannot be parsed. If the mode for parsing is set as DROPMALFORMED, this column will be empty. Default value: _corrupt_record
dateFormat Type: String The format for parsing date strings. Default value: yyyy-MM-dd
dropFieldIfAllNull Type: Boolean Whether to ignore columns of all null values or empty arrays and structs during schema inference. Default value: false
encoding or charset Type: String The name of the encoding of the JSON files. See java.nio.charset.Charset for list of options. You cannot use UTF-16 and UTF-32 when multiline is true. Default value: UTF-8
inferTimestamp Type: Boolean Whether to try and infer timestamp strings as a TimestampType. When set to true, schema inference may take noticeably longer. You must enable cloudFiles.inferColumnTypes to use with Auto Loader. Default value: false
lineSep Type: String A string between two consecutive JSON records. Default value: None, which covers \r, \r\n, and \n
locale Type: String A java.util.Locale identifier. Influences default date, timestamp, and decimal parsing within the JSON. Default value: US
mode Type: String Parser mode around handling malformed records. One of 'PERMISSIVE', 'DROPMALFORMED', or 'FAILFAST'. Default value: PERMISSIVE
multiLine Type: Boolean Whether the JSON records span multiple lines. Default value: false
prefersDecimal Type: Boolean Attempts to infer strings as DecimalType instead of float or double type when possible. You must also use schema inference, either by enabling inferSchema or using cloudFiles.inferColumnTypes with Auto Loader. Default value: false
primitivesAsString Type: Boolean Whether to infer primitive types like numbers and booleans as StringType. Default value: false
rescuedDataColumn Type: String Whether to collect all data that can’t be parsed due to a data type mismatch or schema mismatch (including column casing) to a separate column. This column is included by default when using Auto Loader. For more details, refer to What is the rescued data column?. Default value: None
timestampFormat Type: String The format for parsing timestamp strings. Default value: yyyy-MM-dd'T'HH:mm:ss[.SSS][XXX]
timeZone Type: String The java.time.ZoneId to use when parsing timestamps and dates. Default value: None

CSV options

Option
badRecordsPath Type: String The path to store files for recording the information about bad CSV records. Default value: None
charToEscapeQuoteEscaping Type: Char The character used to escape the character used for escaping quotes. For example, for the following record: [ " a\\", b ]: If the character to escape the '\' is undefined, the record won’t be parsed. The parser will read characters: [a],[\],["],[,],[ ],[b] and throw an error because it cannot find a closing quote. If the character to escape the '\' is defined as '\', the record will be read with 2 values: [a\] and [b]. Default value: '\0'
columnNameOfCorruptRecord Type: String A column for storing records that are malformed and cannot be parsed. If the mode for parsing is set as DROPMALFORMED, this column will be empty. Default value: _corrupt_record
comment Type: Char Defines the character that represents a line comment when found in the beginning of a line of text. Use '\0' to disable comment skipping. Default value: '\u0000'
dateFormat Type: String The format for parsing date strings. Default value: yyyy-MM-dd
emptyValue Type: String String representation of an empty value. Default value: ""
encoding or charset Type: String The name of the encoding of the CSV files. See java.nio.charset.Charset for the list of options. UTF-16 and UTF-32 cannot be used when multiline is true. Default value: UTF-8
enforceSchema Type: Boolean Whether to forcibly apply the specified or inferred schema to the CSV files. If the option is enabled, headers of CSV files are ignored. This option is ignored by default when using Auto Loader to rescue data and allow schema evolution. Default value: true
escape Type: Char The escape character to use when parsing the data. Default value: '\'
header Type: Boolean Whether the CSV files contain a header. Auto Loader assumes that files have headers when inferring the schema. Default value: false
ignoreLeadingWhiteSpace Type: Boolean Whether to ignore leading whitespaces for each parsed value. Default value: false
ignoreTrailingWhiteSpace Type: Boolean Whether to ignore trailing whitespaces for each parsed value. Default value: false
inferSchema Type: Boolean Whether to infer the data types of the parsed CSV records or to assume all columns are of StringType. Requires an additional pass over the data if set to true. For Auto Loader, use cloudFiles.inferColumnTypes instead. Default value: false
lineSep Type: String A string between two consecutive CSV records. Default value: None, which covers \r, \r\n, and \n
locale Type: String A java.util.Locale identifier. Influences default date, timestamp, and decimal parsing within the CSV. Default value: US
maxCharsPerColumn Type: Int Maximum number of characters expected from a value to parse. Can be used to avoid memory errors. Defaults to -1, which means unlimited. Default value: -1
maxColumns Type: Int The hard limit of how many columns a record can have. Default value: 20480
mergeSchema Type: Boolean Whether to infer the schema across multiple files and to merge the schema of each file. Enabled by default for Auto Loader when inferring the schema. Default value: false
mode Type: String Parser mode around handling malformed records. One of 'PERMISSIVE', 'DROPMALFORMED', and 'FAILFAST'. Default value: PERMISSIVE
multiLine Type: Boolean Whether the CSV records span multiple lines. Default value: false
nanValue Type: String The string representation of a non-a-number value when parsing FloatType and DoubleType columns. Default value: "NaN"
negativeInf Type: String The string representation of negative infinity when parsing FloatType or DoubleType columns. Default value: "-Inf"
nullValue Type: String String representation of a null value. Default value: ""
parserCaseSensitive (deprecated) Type: Boolean While reading files, whether to align columns declared in the header with the schema case sensitively. This is true by default for Auto Loader. Columns that differ by case will be rescued in the rescuedDataColumn if enabled. This option has been deprecated in favor of readerCaseSensitive. Default value: false
positiveInf Type: String The string representation of positive infinity when parsing FloatType or DoubleType columns. Default value: "Inf"
preferDate Type: Boolean Attempts to infer strings as dates instead of timestamp when possible. You must also use schema inference, either by enabling inferSchema or using cloudFiles.inferColumnTypes with Auto Loader. Default value: true
quote Type: Char The character used for escaping values where the field delimiter is part of the value. Default value: "
readerCaseSensitive Type: Boolean Specifies the case sensitivity behavior when rescuedDataColumn is enabled. If true, rescue the data columns whose names differ by case from the schema; otherwise, read the data in a case-insensitive manner. Default value: true
rescuedDataColumn Type: String Whether to collect all data that can’t be parsed due to: a data type mismatch, and schema mismatch (including column casing) to a separate column. This column is included by default when using Auto Loader. For more details refer to What is the rescued data column?. Default value: None
sep or delimiter Type: String The separator string between columns. Default value: ","
skipRows Type: Int The number of rows from the beginning of the CSV file that should be ignored (including commented and empty rows). If header is true, the header will be the first unskipped and uncommented row. Default value: 0
timestampFormat Type: String The format for parsing timestamp strings. Default value: yyyy-MM-dd'T'HH:mm:ss[.SSS][XXX]
timeZone Type: String The java.time.ZoneId to use when parsing timestamps and dates. Default value: None
unescapedQuoteHandling Type: String The strategy for handling unescaped quotes. Allowed options: STOP_AT_CLOSING_QUOTE: If unescaped quotes are found in the input, accumulate the quote character and proceed parsing the value as a quoted value, until a closing quote is found. BACK_TO_DELIMITER: If unescaped quotes are found in the input, consider the value as an unquoted value. This will make the parser accumulate all characters of the current parsed value until the delimiter defined by sep is found. If no delimiter is found in the value, the parser will continue accumulating characters from the input until a delimiter or line ending is found. STOP_AT_DELIMITER: If unescaped quotes are found in the input, consider the value as an unquoted value. This will make the parser accumulate all characters until the delimiter defined by sep, or a line ending is found in the input. SKIP_VALUE: If unescaped quotes are found in the input, the content parsed for the given value will be skipped (until the next delimiter is found) and the value set in nullValue will be produced instead. RAISE_ERROR: If unescaped quotes are found in the input, a TextParsingException will be thrown. Default value: STOP_AT_DELIMITER

PARQUET options

Option
datetimeRebaseMode Type: String Controls the rebasing of the DATE and TIMESTAMP values between Julian and Proleptic Gregorian calendars. Allowed values: EXCEPTION, LEGACY, and CORRECTED. Default value: LEGACY
int96RebaseMode Type: String Controls the rebasing of the INT96 timestamp values between Julian and Proleptic Gregorian calendars. Allowed values: EXCEPTION, LEGACY, and CORRECTED. Default value: LEGACY
mergeSchema Type: Boolean Whether to infer the schema across multiple files and to merge the schema of each file. Default value: false
readerCaseSensitive Type: Boolean Specifies the case sensitivity behavior when rescuedDataColumn is enabled. If true, rescue the data columns whose names differ by case from the schema; otherwise, read the data in a case-insensitive manner. Default value: true
rescuedDataColumn Type: String Whether to collect all data that can’t be parsed due to: a data type mismatch, and schema mismatch (including column casing) to a separate column. This column is included by default when using Auto Loader. For more details refer to What is the rescued data column?. Default value: None

AVRO options

Option
avroSchema Type: String Optional schema provided by a user in Avro format. When reading Avro, this option can be set to an evolved schema, which is compatible but different with the actual Avro schema. The deserialization schema will be consistent with the evolved schema. For example, if you set an evolved schema containing one additional column with a default value, the read result will contain the new column too. Default value: None
datetimeRebaseMode Type: String Controls the rebasing of the DATE and TIMESTAMP values between Julian and Proleptic Gregorian calendars. Allowed values: EXCEPTION, LEGACY, and CORRECTED. Default value: LEGACY
mergeSchema Type: Boolean Whether to infer the schema across multiple files and to merge the schema of each file. mergeSchema for Avro does not relax data types. Default value: false
readerCaseSensitive Type: Boolean Specifies the case sensitivity behavior when rescuedDataColumn is enabled. If true, rescue the data columns whose names differ by case from the schema; otherwise, read the data in a case-insensitive manner. Default value: true
rescuedDataColumn Type: String Whether to collect all data that can’t be parsed due to: a data type mismatch, and schema mismatch (including column casing) to a separate column. This column is included by default when using Auto Loader. For more details refer to What is the rescued data column?. Default value: None

BINARYFILE options

Binary files do not have any additional configuration options.

TEXT options

Option
encoding Type: String The name of the encoding of the TEXT files. See java.nio.charset.Charset for list of options. Default value: UTF-8
lineSep Type: String A string between two consecutive TEXT records. Default value: None, which covers \r, \r\n and \n
wholeText Type: Boolean Whether to read a file as a single record. Default value: false

ORC options

Option
mergeSchema Type: Boolean Whether to infer the schema across multiple files and to merge the schema of each file. Default value: false

Streaming options

These options apply when using read_files inside a streaming table or streaming query.

Option
allowOverwrites Type: Boolean Whether to re-process files that have been modified after discovery. The latest available version of the file will be processed during a refresh if it has been modified since the last successful refresh query start time. Default value: false
includeExistingFiles Type: Boolean Whether to include existing files in the stream processing input path or to only process new files arriving after initial setup. This option is evaluated only when you start a stream for the first time. Changing this option after restarting the stream has no effect. Default value: true
maxBytesPerTrigger Type: Byte String The maximum number of new bytes to be processed in every trigger. You can specify a byte string such as 10g to limit each microbatch to 10 GB of data. This is a soft maximum. If you have files that are 3 GB each, Databricks processes 12 GB in a microbatch. When used together with maxFilesPerTrigger, Databricks consumes up to the lower limit of maxFilesPerTrigger or maxBytesPerTrigger, whichever is reached first. Default value: None
maxFilesPerTrigger Type: Integer The maximum number of new files to be processed in every trigger. When used together with maxBytesPerTrigger, Databricks consumes up to the lower limit of maxFilesPerTrigger or maxBytesPerTrigger, whichever is reached first. Default value: 1000
schemaEvolutionMode Type: String The mode for evolving the schema as new columns are discovered in the data. By default, columns are inferred as strings when inferring JSON datasets. See schema evolution for more details. This option doesn’t apply to text and binaryFile files. Default value: "addNewColumns" when a schema is not provided. "none" otherwise.
schemaLocation Type: String The location to store the inferred schema and subsequent changes. See schema inference for more details. The schema location is not required when used in a streaming table query. Default value: None

Examples

-- Reads the files available in the given path. Auto-detects the format and schema of the data.
> SELECT * FROM read_files('abfss://container@storageAccount.dfs.core.windows.net/base/path');

-- Reads the headerless CSV files in the given path with the provided schema.
> SELECT * FROM read_files(
    's3://bucket/path',
    format => 'csv',
    schema => 'id int, ts timestamp, event string');

-- Infers the schema of CSV files with headers. Because the schema is not provided,
-- the CSV files are assumed to have headers.
> SELECT * FROM read_files(
    's3://bucket/path',
    format => 'csv')

-- Reads files that have a csv suffix.
> SELECT * FROM read_files('s3://bucket/path/*.csv')

-- Reads a single JSON file
> SELECT * FROM read_files(
    'abfss://container@storageAccount.dfs.core.windows.net/path/single.json')

-- Reads JSON files and overrides the data type of the column `id` to integer.
> SELECT * FROM read_files(
    's3://bucket/path',
    format => 'json',
    schemaHints => 'id int')

-- Reads files that have been uploaded or modified yesterday.
> SELECT * FROM read_files(
    'gs://my-bucket/avroData',
    modifiedAfter => date_sub(current_date(), 1),
    modifiedBefore => current_date())

-- Creates a Delta table and stores the source file path as part of the data
> CREATE TABLE my_avro_data
  AS SELECT *, _metadata.file_path
  FROM read_files('gs://my-bucket/avroData')

-- Creates a streaming table that processes files that appear only after the table's creation.
-- The table will most likely be empty (if there's no clock skew) after being first created,
-- and future refreshes will bring new data in.
> CREATE OR REFRESH STREAMING TABLE avro_data
  AS SELECT * FROM STREAM read_files('gs://my-bucket/avroData', includeExistingFiles => false);

Related articles

• CREATE STREAMING TABLE

• read_kafka table-valued function