Schema discovery#

Schema discovery identifies and registers tables and views that are newly added to a known schema location. For example, when a logging process creates a new log file every hour, rolling over from the previous hour’s log file, schema discovery locates the newly added files so that SEP can query them.

Note

Schema discovery is available as a public preview. Contact Starburst Support with questions or feedback.

Enable schema discovery#

Schema discovery must be explicitly enabled for a catalog by adding the following catalog configuration property:

hive.experimental.schema-discovery.enabled=true

Before enabling schema discovery, review the limitations and warnings section at the end of this topic.

Run schema discovery#

Run schema discovery to analyze a schema in an object storage location and return the structure of any discovered tables using a system table or a table function in that location.

To run schema discovery using a system table, specify the system table in the FROM clause, and the URI and schema in the WHERE clause, along with any options as shown in the following statement:

SELECT sql FROM hive.schema_discovery.discovery
WHERE uri = 'hdfs://hadoop-master:9000/user/hive/warehouse/customer_orders'
AND schema = 'starburst_demo'
AND options = 'sampleFilesPerTableModulo=1,excludePatterns=**/_SUCCESS'

To run schema discovery using a table function, you must call the schema_discovery.discovery function in the FROM clause, and specify the URI and schema along with any options as follows:

SELECT sql FROM TABLE(schema_discovery.discovery(uri => 's3://test-bucket/path/to/discoverable/table',
schema => 'demo_schema', options => 'sampleFilesPerTableModulo=1,excludePatterns=**/_SUCCESS'))

The following sections describe the available options.

Default options#

Schema discovery is run by default with the following options and their default values. Specify a different value as shown in the previous examples to override the default values.

Option	Default value
`dateFormat`	yyyy-MM-dd
`positiveInf`	Inf
`generatedHeadersFormat`	COL%d
`nanValue`	NaN
`locale`	US
`lineSeparator`
`nullValue`
`complexHadoop`	false
`sampleLinesModulo`	3
`maxSampleFilesPerTable`	5
`skipFileExtensionsCheck`	false
`quote`	“
`includePatterns`	**
`delimiter`	,
`escape`	\
`headers`	true
`ignoreTrailingWhiteSpace`	false
`discoveryMode`	NORMAL
`maxSampleLines`	10
`excludePatterns`	.*
`sampleFilesPerTableModulo`	3
`partitionProjection`	false
`negativeInf`	-Inf
`encoding`	UTF-8
`inferJsonStringDecimal`	false
`supportBuckets`	false
`ignoreLeadingWhiteSpace`	false
`maxSampleTables`	2147483647
`hiveSkipHiddenFiles`	false
`timestampFormat`	yyyy-MM-dd HH:mm:ss[.SSSSSSS]
`comment`	`#`

Standard options#

The following options are set at runtime, and apply to all schema discovery queries:

Option	Description
`encoding`	Charset to use when reading files.
`locale`	Locale for date parsing.
`dateFormat`	Date format pattern.
`timestampFormat`	Time format pattern.
`nanValue`	Value to use for not-a-number.
`positiveInf`	Value to use for positive infinity.
`negativeInf`	Value to use for negative infinity.
`includePatterns`	HDFS include GLOB, split by.
`excludePatterns`	HDFS exclude GLOB, split by.
`maxSampleLines`	Max lines to sample from each file.
`sampleLinesModulo`	Each SAMPLE_LINES_MODULO line is sampled. i.e. if 3, every 3rd line is sampled.
`sampleFilesPerTableModulo`	Each SAMPLE_FILES_PER_TABLE_MODULO file is sampled. i.e. if 3, every 3rd file is sampled.
`maxSampleTables`	Max Tables to discovery.
`maxSampleFilesPerTable`	Max files per table to sample.
`supportBuckets`	If “true” attempt to infer buckets.
`forceTableFormat`	Optional = force the table format [JSON, CSV, ORC, PARQUET, ICEBERG, DELTA_LAKE, ERROR].
`discoveryMode`	Discovery mode, NORMAL - default discovery mode, directories in top level tables [RECURSIVE_DIRECTORIES, NORMAL].
`partitionProjection`	If “true” attempt to infer partition projection.
`skipFileExtensionsCheck`	Whether to skip looking at file extension when discovering file format. Use in case of mismatched file format/extension.
`inferJsonStringDecimal`	Try to parse string json values into Trino’s DECIMAL type hiveSkipHiddenFiles Whether to skip files/directories starting with _ or.

CSV and text file Options#

The following options are set at runtime, and apply only to schema discovery queries on CSV and text files:

Option	Description
`headers`	If “true” treat first line as columns names.
`generatedHeadersFormat`	Pattern to use if column names are auto-generated.
`delimiter`	Text file delimiter.
`quote`	Text file quote.
`escape`	Text file escape.
`comment`	Text file comment.
`nullValue`	Text file null value.
`ignoreLeadingWhiteSpace`	If “true” ignore leading white space.
`ignoreTrailingWhiteSpace`	If “true” ignore trailing white space.
`complexHadoop`	If “true” try to discover arrays, structs and maps.
`lineSeparator`	Text line separator.

Supported formats#

Schema discovery identifies objects stored in the Iceberg, Delta Lake, and Hive table formats. Schema discovery does not identify Hudi tables.

Note

For Iceberg and Delta Lake tables, schema discovery registers tables using the register_table procedure. For Hive tables, schema discovery registers tables using the table metadata.

Schema discovery identifies tables and views that are stored in the following file formats:

JSON in Hive-JSON-SerDe format with one JSON object per line and no line termination characters separating the objects.
CSV
ORC
PARQUET

Schema discovery identifies tables and views that use the following compression codecs:

ZSTD
LZ4
SNAPPY
GZIP
DEFLATE
BZIP2
LZO
LZOP

Note

Schema discovery identifies compression codecs by file extension only. Ensure that your compression codec ends with the proper file extension.

Limitations and warnings#

Schema discovery is only supported for the Hive connector.
Schema Discovery only creates hive tables.
Only JSON, CSV, ORC, and PARQUET file formats are supported.
Only ZSTD, LZ4, SNAPPY, GZIP, DEFLATE, BZIP2, LZO, and LZOP compression codes are supported.

Warning

The public preview of this feature is not subject to role-based access limitations; it can discover tables and views for all schema locations is has access to. If you do not have access to a specific table, you can still use schema discovery to create a new table with the data from the restricted table if:

You have access to schema discovery.
You know the location of the restricted table.