Set Dataflow pipeline options

This page explains how to set pipeline options for your Dataflow jobs. These pipeline options configure how and where your pipeline runs and which resources it uses.

Pipeline execution is separate from your Apache Beam program's execution. The Apache Beam program that you've written constructs a pipeline for deferred execution. This means that the program generates a series of steps that any supported Apache Beam runner can execute. Compatible runners include the Dataflow runner on Google Cloud and the direct runner that executes the pipeline directly in a local environment.

You can pass parameters into a Dataflow job at runtime. For additional information about setting pipeline options at runtime, see Configuring pipeline options.

Note: If you're creating a classic template that accepts pipeline options, you must use the ValueProvider interface in your pipeline code. For more information, see Use runtime parameters in your pipeline code.

Use pipeline options with Apache Beam SDKs

You can use the following SDKs to set pipeline options for Dataflow jobs:

• Apache Beam SDK for Python
• Apache Beam SDK for Java
• Apache Beam SDK for Go

To use the SDKs, you set the pipeline runner and other execution parameters by using the Apache Beam SDK class PipelineOptions.

There are two methods for specifying pipeline options:

• Set pipeline options programmatically by supplying a list of pipeline options.
• Set pipeline options directly on the command line when you run your pipeline code.

Set pipeline options programmatically

You can set pipeline options programmatically by creating and modifying a PipelineOptions object.

Java

Construct a PipelineOptions object using the method PipelineOptionsFactory.fromArgs.

For an example, see the Launch on Dataflow sample section on this page.

Python

Create a PipelineOptions object.

For an example, see the Launch on Dataflow sample section on this page.

Go

Setting pipeline options programmatically using PipelineOptions is not supported in the Apache Beam SDK for Go. Use Go command-line arguments.

For an example, see the Launch on Dataflow sample section on this page.

Set pipeline options on the command line

You can set pipeline options by using command-line arguments.

Java

The following example syntax is from the WordCount pipeline in the Java quickstart.

mvn -Pdataflow-runner compile exec:java \
  -Dexec.mainClass=org.apache.beam.examples.WordCount \
  -Dexec.args="--project=PROJECT_ID \
  --gcpTempLocation=gs://BUCKET_NAME/temp/ \
  --output=gs://BUCKET_NAME/output \
  --runner=DataflowRunner \
  --region=REGION"

Replace the following:

• PROJECT_ID: your Google Cloud project ID
• BUCKET_NAME: the name of your Cloud Storage bucket
• REGION: a Dataflow region, us-central1

Python

The following example syntax is from the WordCount pipeline in the Python quickstart.

python -m apache_beam.examples.wordcount \
  --region DATAFLOW_REGION \
  --input gs://dataflow-samples/shakespeare/kinglear.txt \
  --output gs://STORAGE_BUCKET/results/outputs \
  --runner DataflowRunner \
  --project PROJECT_ID \
  --temp_location gs://STORAGE_BUCKET/tmp/

Replace the following:

• DATAFLOW_REGION: the region where you want to deploy the Dataflow job—for example, europe-west1

  The --region flag overrides the default region that is set in the metadata server, your local client, or environment variables.

• STORAGE_BUCKET: the Cloud Storage bucket name

• PROJECT_ID: the Google Cloud project ID

Go

The following example syntax is from the WordCount pipeline in the Go quickstart.

go run wordcount.go --input gs://dataflow-samples/shakespeare/kinglear.txt \
   --output gs://BUCKET_NAME/results/outputs \
   --runner dataflow \
   --project PROJECT_ID \
   --region DATAFLOW_REGION \
   --staging_location gs://BUCKET_NAME/binaries/

Replace the following:

• BUCKET_NAME: the Cloud Storage bucket name

• PROJECT_ID: the Google Cloud project ID

• DATAFLOW_REGION: The region where you want to deploy the Dataflow job. For example, europe-west1. The --region flag overrides the default region that is set in the metadata server, your local client, or environment variables.

Set experimental pipeline options

In the Java, Python, and Go SDKs, the experiments pipeline option enables experimental or pre-GA Dataflow features.

Set programmatically

To set the experiments option programmatically, use the following syntax.

Java

In your PipelineOptions object, include the experiments option by using the following syntax. This example sets the boot disk size to 80 GB with the experiment flag.

options.setExperiments("streaming_boot_disk_size_gb=80")

For an example that shows how to create the PipelineOptions object, see the Launch on Dataflow sample section on this page.

Python

In your PipelineOptions object, include the experiments option by using the following syntax. This example sets the boot disk size to 80 GB with the experiment flag.

beam_options = PipelineOptions(
  beam_args,
  experiments=['streaming_boot_disk_size_gb=80'])

For an example that shows how to create the PipelineOptions object, see the Launch on Dataflow sample section on this page.

Go

Setting pipeline options programmatically using PipelineOptions is not supported in the Apache Beam SDK for Go. Use Go command-line arguments.

Set on the command line

To set the experiments option on the command line, use the following syntax.

Java

This example sets the boot disk size to 80 GB with the experiment flag.

--experiments=streaming_boot_disk_size_gb=80

Python

This example sets the boot disk size to 80 GB with the experiment flag.

--experiments=streaming_boot_disk_size_gb=80

Go

This example sets the boot disk size to 80 GB with the experiment flag.

--experiments=streaming_boot_disk_size_gb=80

Set in a template

To enable an experimental feature when running a Dataflow template, use the --additional-experiments flag.

Classic template

gcloud dataflow jobs run JOB_NAME --additional-experiments=EXPERIMENT[,...]

Flex template

gcloud dataflow flex-template run JOB_NAME --additional-experiments=EXPERIMENT[,...]

Access the pipeline options object

When you create your Pipeline object in your Apache Beam program, pass PipelineOptions. When the Dataflow service runs your pipeline, it sends a copy of the PipelineOptions to each worker.

Java

Access PipelineOptions inside any ParDo transform's DoFn instance by using the method ProcessContext.getPipelineOptions.

Python

This feature is not supported in the Apache Beam SDK for Python.

Go

Access pipeline options using beam.PipelineOptions.

Launch on Dataflow

Run your job on managed Google Cloud resources by using the Dataflow runner service. Running your pipeline with Dataflow creates a Dataflow job, which uses Compute Engine and Cloud Storage resources in your Google Cloud project. For information about Dataflow permissions, see Dataflow security and permissions.

Dataflow jobs use Cloud Storage to store temporary files during pipeline execution. To avoid being billed for unnecessary storage costs, turn off the soft delete feature on buckets that your Dataflow jobs use for temporary storage. For more information, see Remove a soft delete policy from a bucket.

Note: Typing Ctrl+C from the command line does not cancel your job. The Dataflow service is still running the job on Google Cloud. To cancel the job, use the Dataflow Monitoring Interface or the Dataflow command-line interface.

Set required options

To run your pipeline using Dataflow, set the following pipeline options:

Java

• project: the ID of your Google Cloud project.
• runner: the pipeline runner that executes your pipeline. For Google Cloud execution, this must be DataflowRunner.
• gcpTempLocation: a Cloud Storage path for Dataflow to stage most temporary files. If you want to specify a bucket, you must create the bucket ahead of time. If you don't set gcpTempLocation, you can set the pipeline option tempLocation and then gcpTempLocation is set to the value of tempLocation. If neither are specified, a default gcpTempLocation is created.

• stagingLocation: a Cloud Storage path for Dataflow to stage your binary files. If you're using the Apache Beam SDK 2.28 or higher, don't set this option. For the Apache Beam SDK 2.28 or lower, if you don't set this option, what you specified for the tempLocation is used for the staging location.

  A default gcpTempLocation is created if neither it nor tempLocation is specified. If tempLocation is specified and gcpTempLocation is not, tempLocation must be a Cloud Storage path, and gcpTempLocation defaults to it. If tempLocation is not specified and gcpTempLocation is, tempLocation is not populated.

Python

• project: your Google Cloud project ID.
• region: the region for your Dataflow job.
• runner: the pipeline runner that executes your pipeline. For Google Cloud execution, this must be DataflowRunner.
• temp_location: a Cloud Storage path for Dataflow to stage temporary job files created during the execution of the pipeline.

Go

• project: your Google Cloud project ID.
• region: the region for your Dataflow job.
• runner: the pipeline runner that executes your pipeline. For Google Cloud execution, this must be dataflow.
• staging_location: a Cloud Storage path for Dataflow to stage temporary job files created during the execution of the pipeline.

Set pipeline options programmatically

The following example code shows how to construct a pipeline by programmatically setting the runner and other required options to execute the pipeline using Dataflow.

Java

// Create and set your PipelineOptions.
DataflowPipelineOptions options = PipelineOptionsFactory.as(DataflowPipelineOptions.class);

// For cloud execution, set the Google Cloud project, staging location,
// and set DataflowRunner.
options.setProject("my-project-id");
options.setStagingLocation("gs://my-bucket/binaries");
options.setRunner(DataflowRunner.class);

// Create the Pipeline with the specified options.
Pipeline p = Pipeline.create(options);

Python

import argparse

import apache_beam as beam
from apache_beam.options.pipeline_options import PipelineOptions

parser = argparse.ArgumentParser()
# parser.add_argument('--my-arg', help='description')
args, beam_args = parser.parse_known_args()

# Create and set your PipelineOptions.
# For Cloud execution, specify DataflowRunner and set the Cloud Platform
# project, job name, temporary files location, and region.
# For more information about regions, check:
# https://cloud.google.com/dataflow/docs/concepts/regional-endpoints
beam_options = PipelineOptions(
    beam_args,
    runner='DataflowRunner',
    project='my-project-id',
    job_name='unique-job-name',
    temp_location='gs://my-bucket/temp',
    region='us-central1')
# Note: Repeatable options like dataflow_service_options or experiments must
# be specified as a list of string(s).
# e.g. dataflow_service_options=['enable_prime']

# Create the Pipeline with the specified options.
with beam.Pipeline(options=beam_options) as pipeline:
  pass  # build your pipeline here.

Go

The Apache Beam SDK for Go uses Go command-line arguments. Use flag.Set() to set flag values.

// Use the Go flag package to parse custom options.
flag.Parse()

// Set the required options programmatically.
// For Cloud execution, specify the Dataflow runner, Google Cloud
// project ID, region, and staging location.
// For more information about regions, see
// https://cloud.google.com/dataflow/docs/concepts/regional-endpoints
flag.Set("runner", "dataflow")
flag.Set("project", "my-project-id")
flag.Set("region", "us-central1")
flag.Set("staging_location", "gs://my-bucket/binaries")

beam.Init()

// Create the Pipeline.
p := beam.NewPipeline()
s := p.Root()

After you've constructed your pipeline, specify all the pipeline reads, transforms, and writes, and run the pipeline.

Use pipeline options from the command line

The following example shows how to use pipeline options that are specified on the command line. This example doesn't set the pipeline options programmatically.

Java

// Set your PipelineOptions to the specified command-line options
MyOptions options = PipelineOptionsFactory.fromArgs(args).withValidation();

// Create the Pipeline with the specified options.
Pipeline p = Pipeline.create(options);

Python

Use the Python argparse module to parse command-line options.

# Use Python argparse module to parse custom arguments
import argparse

import apache_beam as beam
from apache_beam.options.pipeline_options import PipelineOptions

# For more details on how to use argparse, take a look at:
#   https://docs.python.org/3/library/argparse.html
parser = argparse.ArgumentParser()
parser.add_argument(
    '--input-file',
    default='gs://dataflow-samples/shakespeare/kinglear.txt',
    help='The file path for the input text to process.')
parser.add_argument(
    '--output-path', required=True, help='The path prefix for output files.')
args, beam_args = parser.parse_known_args()

# Create the Pipeline with remaining arguments.
beam_options = PipelineOptions(beam_args)
with beam.Pipeline(options=beam_options) as pipeline:
  lines = (
      pipeline
      | 'Read files' >> beam.io.ReadFromText(args.input_file)
      | 'Write files' >> beam.io.WriteToText(args.output_path))

Go

Use the Go flag package to parse command-line options. You must parse the options before you call beam.Init(). In this example, output is a command-line option.

// Define configuration options
var (
  output = flag.String("output", "", "Output file (required).")
)

// Parse options before beam.Init()
flag.Parse()

beam.Init()

// Input validation must be done after beam.Init()
if *output == "" {
  log.Fatal("No output provided!")
}

p := beam.NewPipeline()

After you construct your pipeline, specify all the pipeline reads, transforms, and writes, and then run the pipeline.

Control execution modes

When an Apache Beam program runs a pipeline on a service such as Dataflow, the program can either run the pipeline asynchronously, or can block until pipeline completion. You can change this behavior by using the following guidance.

Java

When an Apache Beam Java program runs a pipeline on a service such as Dataflow, it is typically executed asynchronously. To run a pipeline and wait until the job completes, set DataflowRunner as the pipeline runner and explicitly call pipeline.run().waitUntilFinish().

When you use DataflowRunner and call waitUntilFinish() on the PipelineResult object returned from pipeline.run(), the pipeline executes on Google Cloud but the local code waits for the cloud job to finish and return the final DataflowPipelineJob object. While the job runs, the Dataflow service prints job status updates and console messages while it waits.

Python

When an Apache Beam Python program runs a pipeline on a service such as Dataflow, it is typically executed asynchronously. To block until pipeline completion, use the wait_until_finish() method of the PipelineResult object, returned from the run() method of the runner.

Go

When an Apache Beam Go program runs a pipeline on Dataflow, it is synchronous by default and blocks until pipeline completion. If you don't want to block, there are two options:

1. Start the job in a Go routine.

   go func() {
     pr, err := beamx.Run(ctx, p)
     if err != nil {
       // Handle the error.
     }
     // Send beam.PipelineResult into a channel.
     results <- pr
   }()
   // Do other operations while the pipeline runs.
   

2. Use the --async command-line flag, which is in the jobopts package.

To view execution details, monitor progress, and verify job completion status, use the Dataflow monitoring interface or the Dataflow command line interface.

Use streaming sources

Java

If your pipeline reads from an unbounded data source, such as Pub/Sub, the pipeline automatically executes in streaming mode.

Python

If your pipeline uses an unbounded data source, such as Pub/Sub, you must set the streaming option to true.

Go

If your pipeline reads from an unbounded data source, such as Pub/Sub, the pipeline automatically executes in streaming mode.

Streaming jobs use a Compute Engine machine type of n1-standard-2 or higher by default.

Note: If your pipeline uses unbounded data sources and sinks, you must pick a Windowing strategy for your unbounded PCollections before you use any aggregation such as a GroupByKey.

Launch locally

Instead of running your pipeline on managed cloud resources, you can choose to execute your pipeline locally. Local execution has certain advantages for testing, debugging, or running your pipeline over small data sets. For example, local execution removes the dependency on the remote Dataflow service and associated Google Cloud project.

When you use local execution, you must run your pipeline with datasets small enough to fit in local memory. You can create a small in-memory dataset using a Create transform, or you can use a Read transform to work with small local or remote files. Local execution typically provides a faster and easier way to perform testing and debugging with fewer external dependencies, but is limited by the memory that is available in your local environment.

The following example code shows how to construct a pipeline that executes in your local environment.

Java

// Create and set our Pipeline Options.
PipelineOptions options = PipelineOptionsFactory.create();

// Create the Pipeline with the specified options.
Pipeline p = Pipeline.create(options);

Note: For local mode, you don't need to set the runner because DirectRunner is already the default. However, you do need to either explicitly include DirectRunner as a dependency or add it to the classpath.

Python

import argparse

import apache_beam as beam
from apache_beam.options.pipeline_options import PipelineOptions

parser = argparse.ArgumentParser()
# parser.add_argument('--my-arg')
args, beam_args = parser.parse_known_args()

# Create and set your Pipeline Options.
beam_options = PipelineOptions(beam_args)
args = beam_options.view_as(MyOptions)

with beam.Pipeline(options=beam_options) as pipeline:
  lines = (
      pipeline
      | beam.io.ReadFromText(args.input)
      | beam.io.WriteToText(args.output))

Note: For local mode, you don't need to set the runner because DirectRunner is already the default.

Go

// Parse options before beam.Init()
flag.Parse()

beam.Init()

p := beam.NewPipeline()

Note: For local mode, you don't need to set the runner because direct is already the default.

After you've constructed your pipeline, run it.

Create custom pipeline options

You can add your own custom options in addition to the standard PipelineOptions. Apache Beam's command line can also parse custom options using command line arguments specified in the same format.

Java

To add your own options, define an interface with getter and setter methods for each option, as in the following example:

public interface MyOptions extends PipelineOptions {
  String getMyCustomOption();
  void setMyCustomOption(String myCustomOption);
}

Python

To add your own options, use the add_argument() method (which behaves exactly like Python's standard argparse module), as in the following example:

from apache_beam.options.pipeline_options import PipelineOptions

class MyOptions(PipelineOptions):
  @classmethod
  def _add_argparse_args(cls, parser):
    parser.add_argument('--input')
    parser.add_argument('--output')

Go

To add your own options, use the Go flag package as shown in the following example:

var (
  input  = flag.String("input", "", "")
  output = flag.String("output", "", "")
)

You can also specify a description, which appears when a user passes --help as a command-line argument, and a default value.

Java

You set the description and default value using annotations, as follows:

public interface MyOptions extends PipelineOptions {
  @Description("My custom command line argument.")
  @Default.String("DEFAULT")
  String getMyCustomOption();
  void setMyCustomOption(String myCustomOption);
}

We recommend that you register your interface with PipelineOptionsFactory and then pass the interface when creating the PipelineOptions object. When you register your interface with PipelineOptionsFactory, the --help can find your custom options interface and add it to the output of the --help command. PipelineOptionsFactory validates that your custom options are compatible with all other registered options.

The following example code shows how to register your custom options interface with PipelineOptionsFactory:

PipelineOptionsFactory.register(MyOptions.class);
MyOptions options = PipelineOptionsFactory.fromArgs(args)
                                          .withValidation()
                                          .as(MyOptions.class);

Now your pipeline can accept --myCustomOption=value as a command-line argument.

Python

You set the description and default value as follows:

from apache_beam.options.pipeline_options import PipelineOptions

class MyOptions(PipelineOptions):
  @classmethod
  def _add_argparse_args(cls, parser):
    parser.add_argument(
        '--input',
        default='gs://dataflow-samples/shakespeare/kinglear.txt',
        help='The file path for the input text to process.')
    parser.add_argument(
        '--output', required=True, help='The path prefix for output files.')

Go

You set the description and default value as follows:

var (
  input  = flag.String("input", "gs://MY_STORAGE_BUCKET/input", "Input for the pipeline")
  output = flag.String("output", "gs://MY_STORAGE_BUCKET/output", "Output for the pipeline")
)