Use custom containers with C++ libraries


In this tutorial, you create a pipeline that uses custom containers with C++ libraries to run a Dataflow HPC highly parallel workflow. Use this tutorial to learn how to use Dataflow and Apache Beam to run grid computing applications that require data to be distributed to functions running on many cores.

The tutorial demonstrates how to run the pipeline first by using the Direct Runner and then by using the Dataflow Runner. By running the pipeline locally, you can test the pipeline before deploying it.

This example uses Cython bindings and functions from the GMP library. Regardless of the library or binding tool that you use, you can apply the same principles to your pipeline.

The example code is available on GitHub.

Objectives

  • Create a pipeline that uses custom containers with C++ libraries.

  • Build a Docker container image using a Dockerfile.

  • Package the code and dependencies into a Docker container.

  • Run the pipeline locally to test it.

  • Run the pipeline in a distributed environment.

Costs

In this document, you use the following billable components of Google Cloud:

  • Artifact Registry
  • Cloud Build
  • Cloud Storage
  • Compute Engine
  • Dataflow

To generate a cost estimate based on your projected usage, use the pricing calculator. New Google Cloud users might be eligible for a free trial.

When you finish the tasks that are described in this document, you can avoid continued billing by deleting the resources that you created. For more information, see Clean up.

Before you begin

  1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.
  2. Install the Google Cloud CLI.
  3. To initialize the gcloud CLI, run the following command:

    gcloud init
  4. Create or select a Google Cloud project.

    • Create a Google Cloud project:

      gcloud projects create PROJECT_ID

      Replace PROJECT_ID with a name for the Google Cloud project you are creating.

    • Select the Google Cloud project that you created:

      gcloud config set project PROJECT_ID

      Replace PROJECT_ID with your Google Cloud project name.

  5. Make sure that billing is enabled for your Google Cloud project.

  6. Enable the Cloud Storage, Cloud Storage JSON, Compute Engine, Dataflow, Resource Manager, Artifact Registry, and Cloud Build APIs:

    gcloud services enable compute.googleapis.com dataflow.googleapis.com storage_component storage_api cloudresourcemanager.googleapis.com artifactregistry.googleapis.com cloudbuild.googleapis.com
  7. Create local authentication credentials for your user account:

    gcloud auth application-default login
  8. Grant roles to your user account. Run the following command once for each of the following IAM roles: roles/iam.serviceAccountUser

    gcloud projects add-iam-policy-binding PROJECT_ID --member="USER_IDENTIFIER" --role=ROLE
    • Replace PROJECT_ID with your project ID.
    • Replace USER_IDENTIFIER with the identifier for your user account. For example, user:myemail@example.com.

    • Replace ROLE with each individual role.
  9. Install the Google Cloud CLI.
  10. To initialize the gcloud CLI, run the following command:

    gcloud init
  11. Create or select a Google Cloud project.

    • Create a Google Cloud project:

      gcloud projects create PROJECT_ID

      Replace PROJECT_ID with a name for the Google Cloud project you are creating.

    • Select the Google Cloud project that you created:

      gcloud config set project PROJECT_ID

      Replace PROJECT_ID with your Google Cloud project name.

  12. Make sure that billing is enabled for your Google Cloud project.

  13. Enable the Cloud Storage, Cloud Storage JSON, Compute Engine, Dataflow, Resource Manager, Artifact Registry, and Cloud Build APIs:

    gcloud services enable compute.googleapis.com dataflow.googleapis.com storage_component storage_api cloudresourcemanager.googleapis.com artifactregistry.googleapis.com cloudbuild.googleapis.com
  14. Create local authentication credentials for your user account:

    gcloud auth application-default login
  15. Grant roles to your user account. Run the following command once for each of the following IAM roles: roles/iam.serviceAccountUser

    gcloud projects add-iam-policy-binding PROJECT_ID --member="USER_IDENTIFIER" --role=ROLE
    • Replace PROJECT_ID with your project ID.
    • Replace USER_IDENTIFIER with the identifier for your user account. For example, user:myemail@example.com.

    • Replace ROLE with each individual role.
  16. Create a user-managed worker service account for your new pipeline and grant the necessary roles to the service account.

    1. To create the service account, run the gcloud iam service-accounts create command:

      gcloud iam service-accounts create parallelpipeline \
          --description="Highly parallel pipeline worker service account" \
          --display-name="Highly parallel data pipeline access"
    2. Grant roles to the service account. Run the following command once for each of the following IAM roles:

      • roles/dataflow.admin
      • roles/dataflow.worker
      • roles/storage.objectAdmin
      • roles/artifactregistry.reader
      gcloud projects add-iam-policy-binding PROJECT_ID --member="serviceAccount:parallelpipeline@PROJECT_ID.iam.gserviceaccount.com" --role=SERVICE_ACCOUNT_ROLE

      Replace SERVICE_ACCOUNT_ROLE with each individual role.

    3. Grant your Google Account a role that lets you create access tokens for the service account:

      gcloud iam service-accounts add-iam-policy-binding parallelpipeline@PROJECT_ID.iam.gserviceaccount.com --member="user:EMAIL_ADDRESS" --role=roles/iam.serviceAccountTokenCreator

Download the code sample and change directories

Download the code sample and then change directories. The code samples in the GitHub repository provide all the code that you need to run this pipeline. When you are ready to build your own pipeline, you can use this sample code as a template.

Clone the beam-cpp-example repository.

  1. Use the git clone command to clone the GitHub repository:

    git clone https://github.com/GoogleCloudPlatform/dataflow-sample-applications.git
    
  2. Switch to the application directory:

    cd dataflow-sample-applications/beam-cpp-example
    

Pipeline code

You can customize the pipeline code from this tutorial. This pipeline completes the following tasks:

  • Dynamically produces all integers in an input range.
  • Runs the integers through a C++ function and filters bad values.
  • Writes the bad values to a side channel.
  • Counts the occurrence of each stopping time and normalizes the results.
  • Prints the output, formatting and writing the results to a text file.
  • Creates a PCollection with a single element.
  • Processes the single element with a map function and passes the frequency PCollection as a side input.
  • Processes the PCollection and produces a single output.

The starter file looks like the following:

#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements.  See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License.  You may obtain a copy of the License at
#
#    http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#


import argparse
import logging
import os
import sys


def run(argv):
  # Import here to avoid __main__ session pickling issues.
  import io
  import itertools
  import matplotlib.pyplot as plt
  import collatz

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

  class RangeSdf(beam.DoFn, beam.RestrictionProvider):
    """An SDF producing all the integers in the input range.

    This is preferable to beam.Create(range(...)) as it produces the integers
    dynamically rather than materializing them up front.  It is an SDF to do
    so with perfect dynamic sharding.
    """
    def initial_restriction(self, desired_range):
      start, stop = desired_range
      return restriction_trackers.OffsetRange(start, stop)

    def restriction_size(self, _, restriction):
      return restriction.size()

    def create_tracker(self, restriction):
      return restriction_trackers.OffsetRestrictionTracker(restriction)

    def process(self, _, active_range=beam.DoFn.RestrictionParam()):
      for i in itertools.count(active_range.current_restriction().start):
        if active_range.try_claim(i):
          yield i
        else:
          break

  class GenerateIntegers(beam.PTransform):
    def __init__(self, start, stop):
      self._start = start
      self._stop = stop

    def expand(self, p):
      return (
          p
          | beam.Create([(self._start, self._stop + 1)])
          | beam.ParDo(RangeSdf()))

  parser = argparse.ArgumentParser()
  parser.add_argument('--start', dest='start', type=int, default=1)
  parser.add_argument('--stop', dest='stop', type=int, default=10000)
  parser.add_argument('--output', default='./out.png')

  known_args, pipeline_args = parser.parse_known_args(argv)
  # Store this as a local to avoid capturing the full known_args.
  output_path = known_args.output

  with beam.Pipeline(options=PipelineOptions(pipeline_args)) as p:

    # Generate the integers from start to stop (inclusive).
    integers = p | GenerateIntegers(known_args.start, known_args.stop)

    # Run them through our C++ function, filtering bad records.
    # Requires apache beam 2.34 or later.
    stopping_times, bad_values = (
        integers
        | beam.Map(collatz.total_stopping_time).with_exception_handling(
            use_subprocess=True))

    # Write the bad values to a side channel.
    bad_values | 'WriteBadValues' >> beam.io.WriteToText(
        os.path.splitext(output_path)[0] + '-bad.txt')

    # Count the occurrence of each stopping time and normalize.
    total = known_args.stop - known_args.start + 1
    frequencies = (
        stopping_times
        | 'Aggregate' >> (beam.Map(lambda x: (x, 1)) | beam.CombinePerKey(sum))
        | 'Normalize' >> beam.MapTuple(lambda x, count: (x, count / total)))

    if known_args.stop <= 10:
      # Print out the results for debugging.
      frequencies | beam.Map(print)
    else:
      # Format and write them to a text file.
      (
          frequencies
          | 'Format' >> beam.MapTuple(lambda count, freq: f'{count}, {freq}')
          | beam.io.WriteToText(os.path.splitext(output_path)[0] + '.txt'))

    # Define some helper functions.
    def make_scatter_plot(xy):
      x, y = zip(*xy)
      plt.plot(x, y, '.')
      png_bytes = io.BytesIO()
      plt.savefig(png_bytes, format='png')
      png_bytes.seek(0)
      return png_bytes.read()

    def write_to_path(path, content):
      """Most Beam IOs write multiple elements to some kind of a container
      file (e.g. strings to lines of a text file, avro records to an avro file,
      etc.)  This function writes each element to its own file, given by path.
      """
      # Write to a temporary path and to a rename for fault tolerence.
      tmp_path = path + '.tmp'
      fs = beam.io.filesystems.FileSystems.get_filesystem(path)
      with fs.create(tmp_path) as fout:
        fout.write(content)
      fs.rename([tmp_path], [path])

    (
        p
        # Create a PCollection with a single element.
        | 'CreateSingleton' >> beam.Create([None])
        # Process the single element with a Map function, passing the frequency
        # PCollection as a side input.
        # This will cause the normally distributed frequency PCollection to be
        # colocated and processed as a single unit, producing a single output.
        | 'MakePlot' >> beam.Map(
            lambda _,
            data: make_scatter_plot(data),
            data=beam.pvalue.AsList(frequencies))
        # Pair this with the desired filename.
        |
        'PairWithFilename' >> beam.Map(lambda content: (output_path, content))
        # And actually write it out, using MapTuple to split the tuple into args.
        | 'WriteToOutput' >> beam.MapTuple(write_to_path))


if __name__ == '__main__':
  logging.getLogger().setLevel(logging.INFO)
  run(sys.argv)

Set up your development environment

  1. Use the Apache Beam SDK for Python.

  2. Install the GMP library:

    apt-get install libgmp3-dev
    
  3. To install the dependencies, use the requirements.txt file.

    pip install -r requirements.txt
    
  4. To build the Python bindings, run the following command.

    python setup.py build_ext --inplace
    

You can customize the requirements.txt file from this tutorial. The starter file includes the following dependencies:

#
#    Licensed to the Apache Software Foundation (ASF) under one or more
#    contributor license agreements.  See the NOTICE file distributed with
#    this work for additional information regarding copyright ownership.
#    The ASF licenses this file to You under the Apache License, Version 2.0
#    (the "License"); you may not use this file except in compliance with
#    the License.  You may obtain a copy of the License at
#
#       http://www.apache.org/licenses/LICENSE-2.0
#
#    Unless required by applicable law or agreed to in writing, software
#    distributed under the License is distributed on an "AS IS" BASIS,
#    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#    See the License for the specific language governing permissions and
#    limitations under the License.
#

apache-beam[gcp]==2.46.0
cython==0.29.24
pyparsing==2.4.2
matplotlib==3.4.3

Run the pipeline locally

Running the pipeline locally is useful for testing. By running the pipeline locally, you can confirm that the pipeline runs and behaves as expected before you deploy the pipeline to a distributed environment.

You can run the pipeline locally by using the following command. This command outputs an image named out.png.

python pipeline.py

Create the Google Cloud resources

This section explains how to create the following resources:

  • A Cloud Storage bucket to use as a temporary storage location and an output location.
  • A Docker container to package the pipeline code and dependencies.

Create a Cloud Storage bucket

Begin by creating a Cloud Storage bucket using Google Cloud CLI. This bucket is used as a temporary storage location by the Dataflow pipeline.

To create the bucket, use the gcloud storage buckets create command:

gcloud storage buckets create gs://BUCKET_NAME --location=LOCATION

Replace the following:

  • BUCKET_NAME: a name for your Cloud Storage bucket that meets the bucket naming requirements. Cloud Storage bucket names must be globally unique.
  • LOCATION: the location for the bucket.

Create and build a container image

You can customize the Dockerfile from this tutorial. The starter file looks like the following:

FROM apache/beam_python3.9_sdk:2.46.0

# Install a C++ library.
RUN apt-get update
RUN apt-get install -y libgmp3-dev

# Install Python dependencies.
COPY requirements.txt requirements.txt
RUN pip install -r requirements.txt

# Install the code and some python bindings.
COPY pipeline.py pipeline.py
COPY collatz.pyx collatz.pyx
COPY setup.py setup.py
RUN python setup.py install

This Dockerfile contains the FROM, COPY, and RUN commands, which you can read about in the Dockerfile reference.

  1. To upload artifacts, create an Artifact Registry repository. Each repository can contain artifacts for a single supported format.

    All repository content is encrypted using either Google-owned and Google-managed keys or customer-managed encryption keys. Artifact Registry uses Google-owned and Google-managed keys by default and no configuration is required for this option.

    You must have at least Artifact Registry Writer access to the repository.

    Run the following command to create a new repository. The command uses the --async flag and returns immediately, without waiting for the operation in progress to complete.

    gcloud artifacts repositories create REPOSITORY \
       --repository-format=docker \
       --location=LOCATION \
       --async
    

    Replace REPOSITORY with a name for your repository. For each repository location in a project, repository names must be unique.

  2. Create the Dockerfile.

    For packages to be part of the Apache Beam container, you must specify them as part of the requirements.txt file. Ensure that you don't specify apache-beam as part of the requirements.txt file. The Apache Beam container already has apache-beam.

  3. Before you can push or pull images, configure Docker to authenticate requests for Artifact Registry. To set up authentication to Docker repositories, run the following command:

    gcloud auth configure-docker LOCATION-docker.pkg.dev
    

    The command updates your Docker configuration. You can now connect with Artifact Registry in your Google Cloud project to push images.

  4. Build the Docker image using your Dockerfile with Cloud Build.

    Update path in the following command to match the Dockerfile that you created. This command builds the file and pushes it to your Artifact Registry repository.

    gcloud builds submit --tag LOCATION-docker.pkg.dev/PROJECT_ID/REPOSITORY/dataflow/cpp_beam_container:latest .
    

Package the code and dependencies in a Docker container

  1. To run this pipeline in a distributed environment, package the code and dependencies into a docker container.

    docker build . -t cpp_beam_container
    
  2. After you package the code and dependencies, you can run the pipeline locally to test it.

    python pipeline.py \
       --runner=PortableRunner \
       --job_endpoint=embed \
       --environment_type=DOCKER \
       --environment_config="docker.io/library/cpp_beam_container"
    

    This command writes the output inside the Docker image. To view the output, run the pipeline with the --output, and write the output to a Cloud Storage bucket. For example, run the following command.

    python pipeline.py \
       --runner=PortableRunner \
       --job_endpoint=embed \
       --environment_type=DOCKER \
       --environment_config="docker.io/library/cpp_beam_container" \
       --output=gs://BUCKET_NAME/out.png
    

Run the pipeline

You can now run the Apache Beam pipeline in Dataflow by referring to the file with the pipeline code and passing the parameters required by the pipeline.

In your shell or terminal, run the pipeline with the Dataflow Runner.

python pipeline.py \
    --runner=DataflowRunner \
    --project=PROJECT_ID \
    --region=REGION \
    --temp_location=gs://BUCKET_NAME/tmp \
    --sdk_container_image="LOCATION-docker.pkg.dev/PROJECT_ID/REPOSITORY/dataflow/cpp_beam_container:latest" \
    --experiment=use_runner_v2 \
    --output=gs://BUCKET_NAME/out.png

After you execute the command to run the pipeline, the Dataflow returns a Job ID with the job status Queued. It might take several minutes before the job status reaches Running and you can access the job graph.

View your results

View data written to your Cloud Storage bucket. Use the gcloud storage ls command to list the contents at the top level of your bucket:

gcloud storage ls gs://BUCKET_NAME

If successful, the command returns a message similar to:

gs://BUCKET_NAME/out.png

Clean up

To avoid incurring charges to your Google Cloud account for the resources used in this tutorial, either delete the project that contains the resources, or keep the project and delete the individual resources.

Delete the project

The easiest way to eliminate billing is to delete the Google Cloud project that you created for the tutorial.

  1. In the Google Cloud console, go to the Manage resources page.

    Go to Manage resources

  2. In the project list, select the project that you want to delete, and then click Delete.
  3. In the dialog, type the project ID, and then click Shut down to delete the project.

Delete the individual resources

If you want to reuse the project, then delete the resources that you created for the tutorial.

Clean up Google Cloud project resources

  1. Delete the Artifact Registry repository.

    gcloud artifacts repositories delete REPOSITORY \
       --location=LOCATION --async
    
  2. Delete the Cloud Storage bucket. This bucket alone does not incur any charges.

    gcloud storage rm gs://BUCKET_NAME --recursive
    

Revoke credentials

  1. Revoke the roles that you granted to the user-managed worker service account. Run the following command once for each of the following IAM roles:

    • roles/dataflow.admin
    • roles/dataflow.worker
    • roles/storage.objectAdmin
    • roles/artifactregistry.reader
    gcloud projects remove-iam-policy-binding PROJECT_ID \
      --member=serviceAccount:parallelpipeline@PROJECT_ID.iam.gserviceaccount.com \
      --role=SERVICE_ACCOUNT_ROLE
  2. Optional: Revoke the authentication credentials that you created, and delete the local credential file.

    gcloud auth application-default revoke
  3. Optional: Revoke credentials from the gcloud CLI.

    gcloud auth revoke

What's next