org.apache.beam.sdk.io

Class WriteFiles<UserT,DestinationT,OutputT>

java.lang.Object

org.apache.beam.sdk.transforms.PTransform<PCollection<UserT>,WriteFilesResult<DestinationT>>

org.apache.beam.sdk.io.WriteFiles<UserT,DestinationT,OutputT>

All Implemented Interfaces:

java.io.Serializable, HasDisplayData

public abstract class WriteFiles<UserT,DestinationT,OutputT>
extends PTransform<PCollection<UserT>,WriteFilesResult<DestinationT>>

A PTransform that writes to a FileBasedSink. A write begins with a sequential global initialization of a sink, followed by a parallel write, and ends with a sequential finalization of the write. The output of a write is PDone.

By default, every bundle in the input PCollection will be processed by a FileBasedSink.WriteOperation, so the number of output will vary based on runner behavior, though at least 1 output will always be produced. The exact parallelism of the write stage can be controlled using withNumShards(int), typically used to control how many files are produced or to globally limit the number of workers connecting to an external service. However, this option can often hurt performance: it adds an additional GroupByKey to the pipeline.

Example usage with runner-determined sharding:

 p.apply(WriteFiles.to(new MySink(...)));

Example usage with a fixed number of shards:

 p.apply(WriteFiles.to(new MySink(...)).withNumShards(3));

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.Class<? extends WriteFiles>	CONCRETE_CLASS For internal use by runners.
static int	FILE_TRIGGERING_BYTE_COUNT
static Duration	FILE_TRIGGERING_RECORD_BUFFERING_DURATION
static int	FILE_TRIGGERING_RECORD_COUNT

Fields inherited from class org.apache.beam.sdk.transforms.PTransform

annotations, displayData, name, resourceHints

Constructor Summary

Constructors

Constructor and Description
WriteFiles()

Method Summary

Modifier and Type	Method and Description
WriteFilesResult<DestinationT>	expand(PCollection<UserT> input) Override this method to specify how this PTransform should be expanded on the given InputT.
java.util.Map<TupleTag<?>,PValue>	getAdditionalInputs() Returns all PValues that are consumed as inputs to this PTransform that are independent of the expansion of the InputT within PTransform.expand(PInput).
abstract ErrorHandler<BadRecord,?>	getBadRecordErrorHandler()
abstract BadRecordRouter	getBadRecordRouter()
abstract @Nullable PTransform<PCollection<UserT>,PCollectionView<java.lang.Integer>>	getComputeNumShards()
abstract @Nullable ValueProvider<java.lang.Integer>	getNumShardsProvider()
abstract @Nullable ShardingFunction<UserT,DestinationT>	getShardingFunction()
abstract FileBasedSink<UserT,DestinationT,OutputT>	getSink()
abstract boolean	getWindowedWrites()
abstract boolean	getWithAutoSharding()
void	populateDisplayData(DisplayData.Builder builder) Register display data for the given transform or component.
static <UserT,DestinationT,OutputT> WriteFiles<UserT,DestinationT,OutputT>	to(FileBasedSink<UserT,DestinationT,OutputT> sink) Creates a WriteFiles transform that writes to the given FileBasedSink, letting the runner control how many different shards are produced.
void	validate(PipelineOptions options) Called before running the Pipeline to verify this transform is fully and correctly specified.
WriteFiles<UserT,DestinationT,OutputT>	withAutoSharding()
WriteFiles<UserT,DestinationT,OutputT>	withBadRecordErrorHandler(ErrorHandler<BadRecord,?> errorHandler) See FileIO.Write.withBadRecordErrorHandler(ErrorHandler) for details on usage.
WriteFiles<UserT,DestinationT,OutputT>	withBatchMaxBufferingDuration(@Nullable Duration batchMaxBufferingDuration) Returns a new WriteFiles that will batch the input records using specified max buffering duration.
WriteFiles<UserT,DestinationT,OutputT>	withBatchSize(@Nullable java.lang.Integer batchSize) Returns a new WriteFiles that will batch the input records using specified batch size.
WriteFiles<UserT,DestinationT,OutputT>	withBatchSizeBytes(@Nullable java.lang.Integer batchSizeBytes) Returns a new WriteFiles that will batch the input records using specified batch size in bytes.
WriteFiles<UserT,DestinationT,OutputT>	withMaxNumWritersPerBundle(int maxNumWritersPerBundle) Set the maximum number of writers created in a bundle before spilling to shuffle.
WriteFiles<UserT,DestinationT,OutputT>	withNoSpilling() Returns a new WriteFiles that writes all data without spilling, simplifying the pipeline.
WriteFiles<UserT,DestinationT,OutputT>	withNumShards(int numShards) Returns a new WriteFiles that will write to the current FileBasedSink using the specified number of shards.
WriteFiles<UserT,DestinationT,OutputT>	withNumShards(ValueProvider<java.lang.Integer> numShardsProvider) Returns a new WriteFiles that will write to the current FileBasedSink using the ValueProvider specified number of shards.
WriteFiles<UserT,DestinationT,OutputT>	withRunnerDeterminedSharding() Returns a new WriteFiles that will write to the current FileBasedSink with runner-determined sharding.
WriteFiles<UserT,DestinationT,OutputT>	withSharding(PTransform<PCollection<UserT>,PCollectionView<java.lang.Integer>> sharding) Returns a new WriteFiles that will write to the current FileBasedSink using the specified PTransform to compute the number of shards.
WriteFiles<UserT,DestinationT,OutputT>	withShardingFunction(ShardingFunction<UserT,DestinationT> shardingFunction) Returns a new WriteFiles that will write to the current FileBasedSink using the specified sharding function to assign shard for inputs.
WriteFiles<UserT,DestinationT,OutputT>	withSideInputs(java.util.List<PCollectionView<?>> sideInputs)
WriteFiles<UserT,DestinationT,OutputT>	withSkipIfEmpty()
WriteFiles<UserT,DestinationT,OutputT>	withSkipIfEmpty(boolean skipIfEmpty) Set this sink to skip writing any files if the PCollection is empty.
WriteFiles<UserT,DestinationT,OutputT>	withWindowedWrites() Returns a new WriteFiles that writes preserves windowing on it's input.

Methods inherited from class org.apache.beam.sdk.transforms.PTransform

addAnnotation, compose, compose, getAnnotations, getDefaultOutputCoder, getDefaultOutputCoder, getDefaultOutputCoder, getKindString, getName, getResourceHints, setDisplayData, setResourceHints, toString, validate

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

CONCRETE_CLASS

@Internal
public static final java.lang.Class<? extends WriteFiles> CONCRETE_CLASS

For internal use by runners.

FILE_TRIGGERING_RECORD_COUNT

public static final int FILE_TRIGGERING_RECORD_COUNT

See Also:

Constant Field Values

FILE_TRIGGERING_BYTE_COUNT

public static final int FILE_TRIGGERING_BYTE_COUNT

See Also:

Constant Field Values

FILE_TRIGGERING_RECORD_BUFFERING_DURATION

public static final Duration FILE_TRIGGERING_RECORD_BUFFERING_DURATION

Constructor Detail

WriteFiles

public WriteFiles()

Method Detail

to

public static <UserT,DestinationT,OutputT> WriteFiles<UserT,DestinationT,OutputT> to(FileBasedSink<UserT,DestinationT,OutputT> sink)

Creates a WriteFiles transform that writes to the given FileBasedSink, letting the runner control how many different shards are produced.

getSink

public abstract FileBasedSink<UserT,DestinationT,OutputT> getSink()

getComputeNumShards

public abstract @Nullable PTransform<PCollection<UserT>,PCollectionView<java.lang.Integer>> getComputeNumShards()

getNumShardsProvider

public abstract @Nullable ValueProvider<java.lang.Integer> getNumShardsProvider()

getWindowedWrites

public abstract boolean getWindowedWrites()

getWithAutoSharding

public abstract boolean getWithAutoSharding()

getShardingFunction

public abstract @Nullable ShardingFunction<UserT,DestinationT> getShardingFunction()

getBadRecordErrorHandler

public abstract ErrorHandler<BadRecord,?> getBadRecordErrorHandler()

getBadRecordRouter

public abstract BadRecordRouter getBadRecordRouter()

getAdditionalInputs

public java.util.Map<TupleTag<?>,PValue> getAdditionalInputs()

Description copied from class: PTransform

Returns all PValues that are consumed as inputs to this PTransform that are independent of the expansion of the InputT within PTransform.expand(PInput).

For example, this can contain any side input consumed by this PTransform.

Overrides:

getAdditionalInputs in class PTransform<PCollection<UserT>,WriteFilesResult<DestinationT>>

withNumShards

public WriteFiles<UserT,DestinationT,OutputT> withNumShards(int numShards)

Returns a new WriteFiles that will write to the current FileBasedSink using the specified number of shards.

This option should be used sparingly as it can hurt performance. See WriteFiles for more information.

A value less than or equal to 0 will be equivalent to the default behavior of runner-determined sharding.

withNumShards

public WriteFiles<UserT,DestinationT,OutputT> withNumShards(ValueProvider<java.lang.Integer> numShardsProvider)

Returns a new WriteFiles that will write to the current FileBasedSink using the ValueProvider specified number of shards.

This option should be used sparingly as it can hurt performance. See WriteFiles for more information.

withMaxNumWritersPerBundle

public WriteFiles<UserT,DestinationT,OutputT> withMaxNumWritersPerBundle(int maxNumWritersPerBundle)

Set the maximum number of writers created in a bundle before spilling to shuffle.

withSkipIfEmpty

public WriteFiles<UserT,DestinationT,OutputT> withSkipIfEmpty(boolean skipIfEmpty)

Set this sink to skip writing any files if the PCollection is empty.

withBatchSize

public WriteFiles<UserT,DestinationT,OutputT> withBatchSize(@Nullable java.lang.Integer batchSize)

Returns a new WriteFiles that will batch the input records using specified batch size. The default value is FILE_TRIGGERING_RECORD_COUNT.

This option is used only for writing unbounded data with auto-sharding.

withBatchSizeBytes

public WriteFiles<UserT,DestinationT,OutputT> withBatchSizeBytes(@Nullable java.lang.Integer batchSizeBytes)

Returns a new WriteFiles that will batch the input records using specified batch size in bytes. The default value is FILE_TRIGGERING_BYTE_COUNT.

This option is used only for writing unbounded data with auto-sharding.

withBatchMaxBufferingDuration

public WriteFiles<UserT,DestinationT,OutputT> withBatchMaxBufferingDuration(@Nullable Duration batchMaxBufferingDuration)

Returns a new WriteFiles that will batch the input records using specified max buffering duration. The default value is FILE_TRIGGERING_RECORD_BUFFERING_DURATION.

This option is used only for writing unbounded data with auto-sharding.

withSideInputs

public WriteFiles<UserT,DestinationT,OutputT> withSideInputs(java.util.List<PCollectionView<?>> sideInputs)

withSharding

public WriteFiles<UserT,DestinationT,OutputT> withSharding(PTransform<PCollection<UserT>,PCollectionView<java.lang.Integer>> sharding)

Returns a new WriteFiles that will write to the current FileBasedSink using the specified PTransform to compute the number of shards.

This option should be used sparingly as it can hurt performance. See WriteFiles for more information.

withRunnerDeterminedSharding

public WriteFiles<UserT,DestinationT,OutputT> withRunnerDeterminedSharding()

Returns a new WriteFiles that will write to the current FileBasedSink with runner-determined sharding.

withAutoSharding

public WriteFiles<UserT,DestinationT,OutputT> withAutoSharding()

withShardingFunction

public WriteFiles<UserT,DestinationT,OutputT> withShardingFunction(ShardingFunction<UserT,DestinationT> shardingFunction)

Returns a new WriteFiles that will write to the current FileBasedSink using the specified sharding function to assign shard for inputs.

withWindowedWrites

public WriteFiles<UserT,DestinationT,OutputT> withWindowedWrites()

Returns a new WriteFiles that writes preserves windowing on it's input.

If this option is not specified, windowing and triggering are replaced by GlobalWindows and DefaultTrigger.

If there is no data for a window, no output shards will be generated for that window. If a window triggers multiple times, then more than a single output shard might be generated multiple times; it's up to the sink implementation to keep these output shards unique.

This option can only be used if withNumShards(int) is also set to a positive value.

withNoSpilling

public WriteFiles<UserT,DestinationT,OutputT> withNoSpilling()

Returns a new WriteFiles that writes all data without spilling, simplifying the pipeline. This option should not be used with withMaxNumWritersPerBundle(int) and it will eliminate this limit possibly causing many writers to be opened. Use with caution.

This option only applies to writes withRunnerDeterminedSharding().

withSkipIfEmpty

public WriteFiles<UserT,DestinationT,OutputT> withSkipIfEmpty()

withBadRecordErrorHandler

public WriteFiles<UserT,DestinationT,OutputT> withBadRecordErrorHandler(ErrorHandler<BadRecord,?> errorHandler)

See FileIO.Write.withBadRecordErrorHandler(ErrorHandler) for details on usage.

validate

public void validate(PipelineOptions options)

Description copied from class: PTransform

Called before running the Pipeline to verify this transform is fully and correctly specified.

By default, does nothing.

Overrides:

validate in class PTransform<PCollection<UserT>,WriteFilesResult<DestinationT>>

expand

public WriteFilesResult<DestinationT> expand(PCollection<UserT> input)

Description copied from class: PTransform

Override this method to specify how this PTransform should be expanded on the given InputT.

NOTE: This method should not be called directly. Instead apply the PTransform should be applied to the InputT using the apply method.

Composite transforms, which are defined in terms of other transforms, should return the output of one of the composed transforms. Non-composite transforms, which do not apply any transforms internally, should return a new unbound output and register evaluators (via backend-specific registration methods).

Specified by:

expand in class PTransform<PCollection<UserT>,WriteFilesResult<DestinationT>>

populateDisplayData

public void populateDisplayData(DisplayData.Builder builder)

Description copied from class: PTransform

Register display data for the given transform or component.

populateDisplayData(DisplayData.Builder) is invoked by Pipeline runners to collect display data via DisplayData.from(HasDisplayData). Implementations may call super.populateDisplayData(builder) in order to register display data in the current namespace, but should otherwise use subcomponent.populateDisplayData(builder) to use the namespace of the subcomponent.

By default, does not register any display data. Implementors may override this method to provide their own display data.

Specified by:

populateDisplayData in interface HasDisplayData

Overrides:

populateDisplayData in class PTransform<PCollection<UserT>,WriteFilesResult<DestinationT>>

Parameters:

builder - The builder to populate with display data.

See Also:

HasDisplayData