com.google.cloud.dataflow.sdk.transforms

Class PTransform<InputT extends PInput,OutputT extends POutput>

java.lang.Object

com.google.cloud.dataflow.sdk.transforms.PTransform<InputT,OutputT>

Type Parameters:

InputT - the type of the input to this PTransform

OutputT - the type of the output of this PTransform

All Implemented Interfaces:

HasDisplayData, Serializable

Direct Known Subclasses:

AvroIO.Read.Bound, AvroIO.Write.Bound, BigQueryIO.Read.Bound, BigQueryIO.Write.Bound, BigtableIO.Read, BigtableIO.Write, CoGroupByKey, Combine.Globally, Combine.GloballyAsSingletonView, Combine.GroupedValues, Combine.PerKey, Combine.PerKeyWithHotKeyFanout, Count.PerElement, CountingInput.BoundedCountingInput, CountingInput.UnboundedCountingInput, Create.Values, DataflowAssert.GroupThenAssert, DataflowAssert.GroupThenAssertForSingleton, DataflowAssert.OneSideInputAssert, DatastoreV1.DeleteEntity, DatastoreV1.DeleteKey, DatastoreV1.Read, DatastoreV1.Write, Filter, FlatMapElements, Flatten.FlattenIterables, Flatten.FlattenPCollectionList, ForwardingPTransform, GroupByKey, GroupByKey.GroupAlsoByWindow, GroupByKey.GroupByKeyOnly, GroupByKey.ReifyTimestampsAndWindows, GroupByKey.SortValuesByTimestamp, IntraBundleParallelization.Bound, Keys, KvSwap, MapElements, ParDo.Bound, ParDo.BoundMulti, Partition, PubsubIO.Read.Bound, PubsubIO.Write.Bound, PubsubUnboundedSink, PubsubUnboundedSource, Read.Bounded, Read.Unbounded, RemoveDuplicates, RemoveDuplicates.WithRepresentativeValues, Sample.SampleAny, TestStream, TextIO.Read.Bound, TextIO.Write.Bound, Values, View.AsIterable, View.AsList, View.AsMap, View.AsMultimap, View.AsSingleton, View.CreatePCollectionView, Window.Bound, Window.Remerge, WithKeys, WithTimestamps, Write.Bound

public abstract class PTransform<InputT extends PInput,OutputT extends POutput>
extends Object
implements Serializable, HasDisplayData

A PTransform<InputT, OutputT> is an operation that takes an InputT (some subtype of PInput) and produces an OutputT (some subtype of POutput).

Common PTransforms include root PTransforms like TextIO.Read, Create, processing and conversion operations like ParDo, GroupByKey, CoGroupByKey, Combine, and Count, and outputting PTransforms like TextIO.Write. Users also define their own application-specific composite PTransforms.

Each PTransform<InputT, OutputT> has a single InputT type and a single OutputT type. Many PTransforms conceptually transform one input value to one output value, and in this case InputT and Output are typically instances of PCollection. A root PTransform conceptually has no input; in this case, conventionally a PBegin object produced by calling Pipeline.begin() is used as the input. An outputting PTransform conceptually has no output; in this case, conventionally PDone is used as its output type. Some PTransforms conceptually have multiple inputs and/or outputs; in these cases special "bundling" classes like PCollectionList, PCollectionTuple are used to combine multiple values into a single bundle for passing into or returning from the PTransform.

A PTransform<InputT, OutputT> is invoked by calling apply() on its InputT, returning its OutputT. Calls can be chained to concisely create linear pipeline segments. For example:

 
 PCollection<T1> pc1 = ...;
 PCollection<T2> pc2 =
     pc1.apply(ParDo.of(new MyDoFn<T1,KV<K,V>>()))
        .apply(GroupByKey.<K, V>create())
        .apply(Combine.perKey(new MyKeyedCombineFn<K,V>()))
        .apply(ParDo.of(new MyDoFn2<KV<K,V>,T2>()));
  

PTransform operations have unique names, which are used by the system when explaining what's going on during optimization and execution. Each PTransform gets a system-provided default name, but it's a good practice to specify an explicit name, where possible, using the named() method offered by some PTransforms such as ParDo. For example:

 
 ...
 .apply(ParDo.named("Step1").of(new MyDoFn3()))
 ...
  

Each PCollection output produced by a PTransform, either directly or within a "bundling" class, automatically gets its own name derived from the name of its producing PTransform.

Each PCollection output produced by a PTransform also records a Coder that specifies how the elements of that PCollection are to be encoded as a byte string, if necessary. The PTransform may provide a default Coder for any of its outputs, for instance by deriving it from the PTransform input's Coder. If the PTransform does not specify the Coder for an output PCollection, the system will attempt to infer a Coder for it, based on what's known at run-time about the Java type of the output's elements. The enclosing Pipeline's CoderRegistry (accessible via Pipeline.getCoderRegistry()) defines the mapping from Java types to the default Coder to use, for a standard set of Java types; users can extend this mapping for additional types, via CoderRegistry.registerCoder(java.lang.Class<?>, java.lang.Class<?>). If this inference process fails, either because the Java type was not known at run-time (e.g., due to Java's "erasure" of generic types) or there was no default Coder registered, then the Coder should be specified manually by calling TypedPValue.setCoder(com.google.cloud.dataflow.sdk.coders.Coder<T>) on the output PCollection. The Coder of every output PCollection must be determined one way or another before that output is used as an input to another PTransform, or before the enclosing Pipeline is run.

A small number of PTransforms are implemented natively by the Google Cloud Dataflow SDK; such PTransforms simply return an output value as their apply implementation. The majority of PTransforms are implemented as composites of other PTransforms. Such a PTransform subclass typically just implements apply(InputT), computing its Output value from its InputT value. User programs are encouraged to use this mechanism to modularize their own code. Such composite abstractions get their own name, and navigating through the composition hierarchy of PTransforms is supported by the monitoring interface. Examples of composite PTransforms can be found in this directory and in examples. From the caller's point of view, there is no distinction between a PTransform implemented natively and one implemented in terms of other PTransforms; both kinds of PTransform are invoked in the same way, using apply().

Note on Serialization

PTransform doesn't actually support serialization, despite implementing Serializable.

PTransform is marked Serializable solely because it is common for an anonymous DoFn, instance to be created within an apply() method of a composite PTransform.

Each of those *Fns is Serializable, but unfortunately its instance state will contain a reference to the enclosing PTransform instance, and so attempt to serialize the PTransform instance, even though the *Fn instance never references anything about the enclosing PTransform.

To allow such anonymous *Fns to be written conveniently, PTransform is marked as Serializable, and includes dummy writeObject() and readObject() operations that do not save or restore any state.

See Also:

Applying Transformations, Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected String	name The base name of this PTransform, e.g., from ParDo.named(String), or from defaults, or null if not yet assigned.

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	PTransform()
protected	PTransform(String name)

Method Summary

Modifier and Type	Method and Description
OutputT	apply(InputT input) Applies this PTransform on the given InputT, and returns its Output.
protected Coder<?>	getDefaultOutputCoder() Returns the default Coder to use for the output of this single-output PTransform.
protected Coder<?>	getDefaultOutputCoder(InputT input) Returns the default Coder to use for the output of this single-output PTransform when applied to the given input.
<T> Coder<T>	getDefaultOutputCoder(InputT input, TypedPValue<T> output) Returns the default Coder to use for the given output of this single-output PTransform when applied to the given input.
protected String	getKindString() Returns the name to use by default for this PTransform (not including the names of any enclosing PTransforms).
String	getName() Returns the transform name.
void	populateDisplayData(DisplayData.Builder builder) Register display data for the given transform or component.
String	toString()
void	validate(InputT input) Called before invoking apply (which may be intercepted by the runner) to verify this transform is fully specified and applicable to the specified input.

Methods inherited from class java.lang.Object

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

Field Detail

name

protected final transient String name

The base name of this PTransform, e.g., from ParDo.named(String), or from defaults, or null if not yet assigned.

Constructor Detail

PTransform

protected PTransform()

PTransform

protected PTransform(String name)

Method Detail

apply

public OutputT apply(InputT input)

Applies this PTransform on the given InputT, and returns its Output.

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).

The default implementation throws an exception. A derived class must either implement apply, or else each runner must supply a custom implementation via PipelineRunner.apply(com.google.cloud.dataflow.sdk.transforms.PTransform<InputT, OutputT>, InputT).

validate

public void validate(InputT input)

Called before invoking apply (which may be intercepted by the runner) to verify this transform is fully specified and applicable to the specified input.

By default, does nothing.

getName

public String getName()

Returns the transform name.

This name is provided by the transform creator and is not required to be unique.

toString

public String toString()

Overrides:

toString in class Object

getKindString

protected String getKindString()

Returns the name to use by default for this PTransform (not including the names of any enclosing PTransforms).

By default, returns the base name of this PTransform's class.

The caller is responsible for ensuring that names of applied PTransforms are unique, e.g., by adding a uniquifying suffix when needed.

getDefaultOutputCoder

protected Coder<?> getDefaultOutputCoder()
                                  throws CannotProvideCoderException

Returns the default Coder to use for the output of this single-output PTransform.

By default, always throws

Throws:

CannotProvideCoderException - if no coder can be inferred

getDefaultOutputCoder

protected Coder<?> getDefaultOutputCoder(InputT input)
                                  throws CannotProvideCoderException

Returns the default Coder to use for the output of this single-output PTransform when applied to the given input.

Throws:

CannotProvideCoderException - if none can be inferred.

By default, always throws.

getDefaultOutputCoder

public <T> Coder<T> getDefaultOutputCoder(InputT input,
                                          TypedPValue<T> output)
                                   throws CannotProvideCoderException

Returns the default Coder to use for the given output of this single-output PTransform when applied to the given input.

Throws:

CannotProvideCoderException - if none can be inferred.

By default, always throws.

populateDisplayData

public void populateDisplayData(DisplayData.Builder builder)

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

Parameters:

builder - The builder to populate with display data.

See Also:

HasDisplayData