Google Cloud Dataflow provides a simple, powerful programming model for building both batch and streaming parallel data processing pipelines.

The Cloud Dataflow SDK is used to access the Google Cloud Dataflow service, which is currently in Alpha and restricted to whitelisted users.

The SDK is publicly available and can be used for local execution by anyone. Note, however, that the SDK is also an Alpha release and may change significantly over time. The SDK is built to be extensible and support additional execution environments ("runners") beyond local execution and the Google Cloud Dataflow service. As the product matures, we look forward to working with you to improve Cloud Dataflow.

The key concepts in this programming model are:

• PCollection: represents a collection of data, which could be bounded or unbounded in size.
• PTransform: represents a computation that transforms input PCollections into output PCollections.
• Pipeline: manages a directed acyclic graph of PTransforms and PCollections, which is ready for execution.
• PipelineRunner: specifies where and how the pipeline should execute.

Currently there are three PipelineRunners:

1. The DirectPipelineRunner runs the pipeline on your local machine.
2. The DataflowPipelineRunner submits the pipeline to the Dataflow Service, where it runs using managed resources in the Google Cloud Platform.
3. The BlockingDataflowPipelineRunner submits the pipeline to the Dataflow Service via the DataflowPipelineRunner and then prints messages about the job status until execution is complete.

The Dataflow Service is currently in the Alpha phase of development and access is limited to whitelisted users.

This repository consists of two modules:

• Java SDK module provides a set of basic Java APIs to program against.
• Examples module provides a few samples to get started. We recommend starting with the WordCount example.

The following command will build both modules and install them in your local Maven repository:

mvn clean install

You can speed up the build and install process by using the following options:

1. To skip execution of the unit tests, run:

   mvn install -DskipTests

2. While iterating on a specific module, use the following command to compile and reinstall it. For example, to reinstall the 'examples' module, run:

   mvn install -pl examples

Be careful, however, as this command will use the most recently installed SDK from the local repository (or Maven Central) even if you have changed it locally.

3. To run Maven using multiple threads, run:

   mvn -T 4 install

After building and installing, the following command will execute the WordCount example using the DirectPipelineRunner on your local machine:

mvn exec:java -pl examples \
-Dexec.mainClass=com.google.cloud.dataflow.examples.WordCount \
-Dexec.args="--input=<INPUT FILE PATTERN> --output=<OUTPUT FILE>"

If you have been whitelisted for Alpha access to the Dataflow Service and followed the developer setup steps, you can use the BlockingDataflowPipelineRunner to run the same program in the Google Cloud Platform (GCP):

mvn exec:java -pl examples \
-Dexec.mainClass=com.google.cloud.dataflow.examples.WordCount \
-Dexec.args="--project=<YOUR GCP PROJECT NAME> --stagingLocation=<YOUR GCS LOCATION> --runner=BlockingDataflowPipelineRunner"

Google Cloud Storage (GCS) location should be entered in the form of gs://bucket/path/to/staging/directory. Google Cloud Platform (GCP) project refers to its name (not number), which has been whitelisted for Cloud Dataflow. Refer here for instructions to get started with Google Cloud Platform.

Other examples can be run similarly by replacing the WordCount class name with BigQueryTornadoes, DatastoreWordCount, TfIdf, TopWikipediaSessions, etc. and adjusting runtime options under the Dexec.args parameter, as specified in the example itself.

• Google Cloud Dataflow
• Dataflow Concepts and Programming Model
• Javadoc