@PublicEvolving public interface SupportsReadingMetadata
ScanTableSources that support reading metadata columns.
Metadata columns add additional columns to the table's schema. A table source is responsible for adding requested metadata columns at the end of produced rows. This includes potentially forwarding metadata columns from contained formats.
Examples in SQL look like:
// reads the column from corresponding metadata key `timestamp`
CREATE TABLE t1 (i INT, s STRING, timestamp TIMESTAMP(3) WITH LOCAL TIME ZONE METADATA, d DOUBLE)
// reads the column from metadata key `timestamp` and casts to INT
CREATE TABLE t2 (i INT, s STRING, myTimestamp INT METADATA FROM 'timestamp', d DOUBLE)
By default, if this interface is not implemented, the statements above would fail because the table source does not provide a metadata key called `timestamp`.
If this interface is implemented,
listReadableMetadata() lists all metadata keys and
their corresponding data types that the source exposes to the planner. The planner will use this
information for validation and insertion of explicit casts if necessary.
The planner will select required metadata columns (i.e. perform projection push down) and will
applyReadableMetadata(List, DataType) with a list of metadata keys. An
implementation must ensure that metadata columns are appended at the end of the physical row in
the order of the provided list after the apply method has been called, e.g. using
Note: The final output data type emitted by a source changes from the physically produced data
type to a data type with metadata columns.
applyReadableMetadata(List, DataType) will
pass the updated data type for convenience. If a source implements
SupportsProjectionPushDown, the projection must be applied to the physical data in the first
step. The passed updated data type will have considered information from
The metadata column's data type must match with
listReadableMetadata(). For the
examples above, this means that a table source for `t2` returns a TIMESTAMP and not INT. The
casting to INT will be performed by the planner in a subsequent operation:
// physical output
ROW < i INT, s STRING, d DOUBLE >
// final output (i.e. produced type) for t1
ROW < i INT, s STRING, d DOUBLE, timestamp TIMESTAMP(3) WITH LOCAL TIME ZONE >
// final output (i.e. produced type) for t2
ROW < i INT, s STRING, d DOUBLE, myTimestamp TIMESTAMP(3) WITH LOCAL TIME ZONE >
|Modifier and Type
|Method and Description
Provides a list of metadata keys that the produced
RowData must contain as appended
Returns the map of metadata keys and their corresponding data types that can be produced by this table source for reading.
Defines whether projections can be applied to metadata columns.
The returned map will be used by the planner for validation and insertion of explicit
LogicalTypeCasts.supportsExplicitCast(LogicalType, LogicalType)) if
The iteration order of the returned map determines the order of metadata keys in the list
applyReadableMetadata(List, DataType). Therefore, it might be beneficial
to return a
LinkedHashMap if a strict metadata column order is required.
If a source forwards metadata from one or more formats, we recommend the following column order for consistency:
KEY FORMAT METADATA COLUMNS + VALUE FORMAT METADATA COLUMNS + SOURCE METADATA COLUMNS
Metadata key names follow the same pattern as mentioned in
Factory. In case of
duplicate names in format and source keys, format keys shall have higher precedence.
RowData must contain as appended
Implementations of this method must be idempotent. The planner might call this method multiple times.
Note: Use the passed data type instead of
for describing the final output data type when creating
TypeInformation. If the
SupportsProjectionPushDown, the projection is already considered in
the given output data type, use the
producedDataType provided by this method instead
producedDataType provided by
metadataKeys - a subset of the keys returned by
by the iteration order of returned map
producedDataType - the final output type of the source, it is intended to be only
forwarded and the planner will decide on the field names to avoid collisions
default boolean supportsMetadataProjection()
This method is only called if the source does not implement
SupportsProjectionPushDown. By default, the planner will only apply metadata columns which
have actually been selected in the query regardless. By returning
false instead the
source can inform the planner to apply all metadata columns defined in the table's schema.
Copyright © 2014–2024 The Apache Software Foundation. All rights reserved.