Constructing Your Pipeline

Your Dataflow program expresses a data processing pipeline, from start to finish. This section explains the mechanics of using the classes in the Dataflow SDKs to build a pipeline. To construct a pipeline using the classes in the Dataflow SDKs, your program will need to perform the following general steps:

  • Create a Pipeline object.
  • Use a Read or Create transform to create one or more PCollections for your pipeline data.
  • Apply transforms to each PCollection. Transforms can change, filter, group, analyze, or otherwise process the elements in a PCollection. Each transform creates a new output PCollection, to which you can apply additional transforms until processing is complete.
  • Write or otherwise output the final, transformed PCollections.
  • Run the pipeline.

See the Simple Example Pipeline below for a complete example that demonstrates each general step.

Creating Your Pipeline Object

A Dataflow program often starts by creating a Pipeline object.

In the Dataflow SDKs, each pipeline is represented by an explicit object of type Pipeline. Each Pipeline object is an independent entity that encapsulates both the data the pipeline operates over and the transforms that get applied to that data.


To create a pipeline, declare a Pipeline object, and pass it some configuration options. You pass the configuration options by creating an object of type PipelineOptions, which you can build by using the static method PipelineOptionsFactory.create().

// Start by defining the options for the pipeline.
PipelineOptions options = PipelineOptionsFactory.create();

// Then create the pipeline.
Pipeline p = Pipeline.create(options);

Configuring Pipeline Options

Use the pipeline options to configure different aspects of your pipeline. These can include:

  • Where your pipeline runs
  • Where your pipeline job stages files
  • Which Cloud Platform project your pipeline is associated with
  • How many Compute Engine instances your pipeline uses as workers

The pipeline options properties include information about your Cloud Platform project required by the Cloud Dataflow service, such as your project ID and Cloud Storage staging locations. The pipeline options also let you control how many workers the Dataflow service should assign to your pipeline job, and where to direct your pipeline job's status messages.

A key property in the pipeline options that determines where your pipeline executes (either on the Cloud Dataflow Service, or locally) is the pipeline runner. The pipeline runner property also specifies whether your pipeline's execution is to be asynchronous or blocking.


While you can set the properties of the PipelineOptions object directly within your pipeline program using setter methods (PipelineOptions.set[OptionName]), a best practice is to pass in the values using command line options. The Dataflow SDK for Java provides a PipelineOptionsFactory class that parses and validates command line options passed in to your pipeline. By using command line options to determine the PipelineRunner and other fields in PipelineOptions at runtime, you can use the same code to construct and run your pipeline both locally and in the cloud.

See Specifying Execution Parameters for more information about how to set pipeline options programatically for either cloud or local mode execution. The WordCount example pipeline also demonstrates how to set pipeline options at runtime by using command-line options.

Reading Data Into Your Pipeline

To create your pipeline's initial PCollection, you apply a root transform to your pipeline object. A root transform creates a PCollection from either an external data source or some local data you specify.


There are two kinds of root transforms in the Dataflow Java SDK: Read and Create. Read transforms read data from an external source, such as BigQuery or a text file in Google Cloud Storage. Create transforms create a PCollection from an in-memory java.util.Collection.

The following example code shows how to apply a TextIO.Read root transform to read data from a text file in Google Cloud Storage. The transform is applied to a Pipeline object p, and returns a pipeline data set in the form of a PCollection<String>:

PCollection<String> lines = p.apply(

Applying Transforms to Process Pipeline Data

To use transforms in your pipeline, you apply them to the PCollection that you want to transform.


To apply a transform, you call the apply method on each PCollection that you want to process, passing the desired transform object as an argument.

The Dataflow SDKs contain a number of different transforms that you can apply to your pipeline's PCollections. These include general-purpose core transforms, such as ParDo or Combine. There are also pre-written composite transforms included in the SDK, which combine one or more of the core transforms in a useful processing pattern, such as counting or combining elements in a collection. You can also define your own more complex composite transforms to fit your pipeline's exact use case.


In the Dataflow Java SDK, each transform is a subclass of the base class PTransform. When you call apply on a PCollection, you pass the PTransform you want to use as an argument.

The following code shows how to apply a transform to a PCollection of strings. The transform is a user-defined custom transform that reverses the contents of each string and outputs a new PCollection containing the reversed strings.

The input is a PCollection<String> called words; the code passes an instance of a PTransform object called ReverseWords to apply, and saves the return value as the PCollection<String> called reversedWords.

PCollection<String> words = ...;

PCollection<String> reversedWords = words.apply(new ReverseWords());

Writing or Outputting Your Final Pipeline Data

Once your pipeline has applied all of its transforms, you'll usually need to output the results. To output your pipeline's final PCollections, you apply a Write transform to that PCollection. Write transforms can output the elements of a PCollection to an external data sink, such as a file in Google Cloud Storage or a BigQuery table. You can use Write to output a PCollection at any time in your pipeline, although you'll typically write out data at the end of your pipeline.


The following example code shows how to apply a TextIO.Write transform to write a PCollection of String to a text file in Google Cloud Storage:

PCollection<String> filteredWords = ...;

Running Your Pipeline

Once you have constructed your pipeline, you use the run method to execute the pipeline. Pipelines are executed asynchronously: the program you create sends a specification for your pipeline to a pipeline runner, which then constructs and runs the actual series of pipeline operations. You can specify where your pipeline runs: either locally for testing and debugging purposes, or on the Cloud Dataflow managed service. See Specifying Execution Parameters for more information on pipeline runners, configuring pipeline options, and local vs. cloud execution.

In the Dataflow SDKs, you specify a PipelineRunner in your pipeline options when you create your Pipeline object. When you've finished constructing your pipeline, you invoke run on your pipeline object as follows: