Dataproc optional Flink component

You can activate additional components like Flink when you create a Dataproc cluster using the Optional components feature. This page shows you how to create a Dataproc cluster with the Apache Flink optional component activated (a Flink cluster), and then run Flink jobs on the cluster.

You can use your Flink cluster to:

  1. Run Flink jobs using the Dataproc Jobs resource from the Google Cloud console, Google Cloud CLI, or the Dataproc API.

  2. Run Flink jobs using the flink CLI running on the Flink cluster master node.

  3. Run Apache Beam jobs on Flink

  4. Run Flink on a Kerberized cluster

You can use the Google Cloud console, Google Cloud CLI, or the Dataproc API to create a Dataproc cluster that has the Flink component activated on the cluster.

Recommendation: Use a standard 1-master VM cluster with the Flink component. Dataproc High Availability mode clusters (with 3 master VMs) do not support Flink high-availability mode.

You can run Flink jobs using the Dataproc Jobs resource from the Google Cloud console, Google Cloud CLI, or Dataproc API.

Console

To submit a sample Flink wordcount job from the console:

  1. Open the Dataproc Submit a job page in the Google Cloud console in your browser.

  2. Fill in the fields on the Submit a job page:

    1. Select your Cluster name from the cluster list.
    2. Set Job type to Flink.
    3. Set Main class or jar to org.apache.flink.examples.java.wordcount.WordCount.
    4. Set Jar files to file:///usr/lib/flink/examples/batch/WordCount.jar.
      • file:/// denotes a file located on the cluster. Dataproc installed the WordCount.jar when it created the Flink cluster.
      • This field also accepts a Cloud Storage path (gs://BUCKET/JARFILE) or a Hadoop Distributed File System (HDFS) path (hdfs://PATH_TO_JAR).
  3. Click Submit.

    • Job driver output is displayed on the Job details page.
    • Flink jobs are listed on the Dataproc Jobs page in the Google Cloud console.
    • Click Stop or Delete from the Jobs or Job details page to stop or delete a job.

gcloud

To submit a Flink job to a Dataproc Flink cluster, run the gcloud CLI gcloud dataproc jobs submit command locally in a terminal window or in Cloud Shell.

gcloud dataproc jobs submit flink \
    --cluster=CLUSTER_NAME \
    --region=REGION \
    --class=MAIN_CLASS \
    --jar=JAR_FILE \
    -- JOB_ARGS

Notes:

  • CLUSTER_NAME: Specify the name of the Dataproc Flink cluster to submit the job to.
  • REGION: Specify a Compute Engine region where the cluster is located.
  • MAIN_CLASS: Specify the main class of your Flink application, such as:
    • org.apache.flink.examples.java.wordcount.WordCount
  • JAR_FILE: Specify the Flink application jar file. You can specify:
    • A jar file installed on the cluster, using the file:///` prefix:
      • file:///usr/lib/flink/examples/streaming/TopSpeedWindowing.jar
      • file:///usr/lib/flink/examples/batch/WordCount.jar
    • A jar file in Cloud Storage: gs://BUCKET/JARFILE
    • A jar file in HDFS: hdfs://PATH_TO_JAR
  • JOB_ARGS: Optionally, add job arguments after the double dash (--).

  • After submitting the job, job driver output is displayed in the local or Cloud Shell terminal.

    Program execution finished
    Job with JobID 829d48df4ebef2817f4000dfba126e0f has finished.
    Job Runtime: 13610 ms
    ...
    (after,1)
    (and,12)
    (arrows,1)
    (ay,1)
    (be,4)
    (bourn,1)
    (cast,1)
    (coil,1)
    (come,1)
    

REST

This section shows how to submit a Flink job to a Dataproc Flink cluster using the Dataproc jobs.submit API.

Before using any of the request data, make the following replacements:

  • PROJECT_ID: Google Cloud project ID
  • REGION: cluster region
  • CLUSTER_NAME: Specify the name of the Dataproc Flink cluster to submit the job to

HTTP method and URL:

POST https://dataproc.googleapis.com/v1/projects/PROJECT_ID/regions/REGION/jobs:submit

Request JSON body:

{
  "job": {
    "placement": {
      "clusterName": "CLUSTER_NAME"
    },
    "flinkJob": {
      "mainClass": "org.apache.flink.examples.java.wordcount.WordCount",
      "jarFileUris": [
        "file:///usr/lib/flink/examples/batch/WordCount.jar"
      ]
    }
  }
}

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

{
  "reference": {
    "projectId": "PROJECT_ID",
    "jobId": "JOB_ID"
  },
  "placement": {
    "clusterName": "CLUSTER_NAME",
    "clusterUuid": "CLUSTER_UUID"
  },
  "flinkJob": {
    "mainClass": "org.apache.flink.examples.java.wordcount.WordCount",
    "args": [
      "1000"
    ],
    "jarFileUris": [
      "file:///usr/lib/flink/examples/batch/WordCount.jar"
    ]
  },
  "status": {
    "state": "PENDING",
    "stateStartTime": "2020-10-07T20:16:21.759Z"
  },
  "jobUuid": "JOB_UUID"
}
  • Flink jobs are listed on the Dataproc Jobs page in the Google Cloud console.
  • You can click Stop or Delete from the Jobs or Job details page in the Google Cloud console to stop or delete a job.

Instead of running Flink jobs using the Dataproc Jobs resource, you can run Flink jobs on the master node of your Flink cluster using the flink CLI.

The following sections describe different ways you can run a flink CLI job on your Dataproc Flink cluster.

  1. SSH into the master node: Use the SSH utility to open a terminal window on the cluster master VM.

  2. Set the classpath: Initialize the Hadoop classpath from the SSH terminal window on the Flink cluster master VM:

    export HADOOP_CLASSPATH=$(hadoop classpath)
    
  3. Run Flink jobs: You can run Flink jobs in different deployment modes on YARN: application, per-job, and session mode.

    1. Application mode: Flink Application mode is supported by Dataproc image version 2.0 and later. This mode executes the job's main() method on the YARN Job Manager. The cluster shuts down after the job finishes.

      Job submission example:

      flink run-application \
          -t yarn-application \
          -Djobmanager.memory.process.size=1024m \
          -Dtaskmanager.memory.process.size=2048m \
          -Djobmanager.heap.mb=820 \
          -Dtaskmanager.heap.mb=1640 \
          -Dtaskmanager.numberOfTaskSlots=2 \
          -Dparallelism.default=4 \
          /usr/lib/flink/examples/batch/WordCount.jar
      

      List running jobs:

      ./bin/flink list -t yarn-application -Dyarn.application.id=application_XXXX_YY
      

      Cancel a running job:

      ./bin/flink cancel -t yarn-application -Dyarn.application.id=application_XXXX_YY <jobId>
      
    2. Per-job mode: This Flink mode executes the job's main() method on the client side.

      Job submission example:

      flink run \
          -m yarn-cluster \
          -p 4 \
          -ys 2 \
          -yjm 1024m \
          -ytm 2048m \
          /usr/lib/flink/examples/batch/WordCount.jar
      
    3. Session mode: Start a long-running Flink YARN session, then submit one or more jobs to the session.

      1. Start a session: You can start a Flink session in one of the following ways:

        1. Create a Flink cluster, adding the --metadata flink-start-yarn-session=true flag to the gcloud dataproc clusters create command (See Create a Dataproc Flink cluster). With this flag enabled, after the cluster is created, Dataproc runs /usr/bin/flink-yarn-daemon to start a Flink session on the cluster.

          The session's YARN application ID is saved in /tmp/.yarn-properties-${USER}. You can list the ID with the yarn application -list command.

        2. Run the Flink yarn-session.sh script, which is pre-installed on the cluster master VM, with custom settings:

          Example with custom settings:

          /usr/lib/flink/bin/yarn-session.sh \
              -s 1 \
              -jm 1024m \
              -tm 2048m \
              -nm flink-dataproc \
              --detached
          
        3. Run the Flink the /usr/bin/flink-yarn-daemon wrapper script with default settings:

          . /usr/bin/flink-yarn-daemon
          
      2. Submit a job to a session: Run the following command to submit a Flink job to the session.

        flink run -m <var>FLINK_MASTER_URL</var>/usr/lib/flink/examples/batch/WordCount.jar
        
        • FLINK_MASTER_URL: the URL, including host and port, of the Flink master VM where jobs are executed. Remove the http:// prefix from the URL. This URL is listed in the command output when you start a Flink session. You can run the following command to list this URL in the Tracking-URL field:
        yarn application -list -appId=<yarn-app-id> | sed 's#http://##'
           ```
        
      3. List jobs in a session: To list Flink jobs in a session, do one of the following:

        • Run flink list without arguments. The command looks for the the session's YARN application ID in /tmp/.yarn-properties-${USER}.

        • Obtain the YARN application ID of the session from /tmp/.yarn-properties-${USER} or the output of yarn application -list, and then run <code>flink list -yid YARN_APPLICATION_ID.

        • Run flink list -m FLINK_MASTER_URL.

      4. Stop a session: To stop the session, obtain the YARN application ID of the session from /tmp/.yarn-properties-${USER} or the output of yarn application -list, then run either of the following commands:

        echo "stop" | /usr/lib/flink/bin/yarn-session.sh -id YARN_APPLICATION_ID
        
        yarn application -kill YARN_APPLICATION_ID
        

You can run Apache Beam jobs on Dataproc using the FlinkRunner.

You can run Beam jobs on Flink in the following ways:

  1. Java Beam jobs
  2. Portable Beam jobs

Java Beam jobs

Package your Beam jobs into a JAR file. Supply the bundled JAR file with the dependencies needed to run the job.

The following example runs a Java Beam job from the Dataproc cluster's master node.

  1. Create a Dataproc cluster with the Flink component enabled.

    gcloud dataproc clusters create CLUSTER_NAME \
        --optional-components=FLINK \
        --image-version=DATAPROC_IMAGE_VERSION \
        --region=REGION \
        --enable-component-gateway \
        --scopes=https://www.googleapis.com/auth/cloud-platform
    
    • --optional-components: Flink.
    • --image-version: the cluster's image version, which determines the Flink version installed on the cluster (for example, see the Apache Flink component versions listed for the latest and previous four 2.0.x image release versions).
    • --region: a supported Dataproc region.
    • --enable-component-gateway: enable access to the Flink Job Manager UI.
    • --scopes: enable access to Google Cloud APIs by your cluster (see Scopes best practice). cloud-platform scope is enabled by default (you do not need to include this flag setting) when you create a cluster that uses Dataproc image version 2.1 or later.
  2. Use the SSH utility to open a terminal window on the Flink cluster master node.

  3. Start a Flink YARN session on the Dataproc cluster master node.

    . /usr/bin/flink-yarn-daemon
    

    Take note of the Flink version on your Dataproc cluster.

    flink --version
    
  4. On your local machine, generate the canonical Beam word count example in Java.

    Choose a Beam version that is compatible with the Flink version on your Dataproc cluster. See the Flink Version Compatibility table that lists Beam-Flink version compatibility.

    Open the generated POM file. Check the Beam Flink runner version specified by the tag <flink.artifact.name>. If the Beam Flink runner version in the Flink artifact name does not match the Flink version on your cluster, update the version number to match.

    mvn archetype:generate \
        -DarchetypeGroupId=org.apache.beam \
        -DarchetypeArtifactId=beam-sdks-java-maven-archetypes-examples \
        -DarchetypeVersion=BEAM_VERSION \
        -DgroupId=org.example \
        -DartifactId=word-count-beam \
        -Dversion="0.1" \
        -Dpackage=org.apache.beam.examples \
        -DinteractiveMode=false
    
  5. Package the word count example.

    mvn package -Pflink-runner
    
  6. Upload the packaged uber JAR file, word-count-beam-bundled-0.1.jar (~135 MB) to your Dataproc cluster's master node. You can use gcloud storage cp for faster file transfers to your Dataproc cluster from Cloud Storage.

    1. On your local terminal, create a Cloud Storage bucket, and upload the uber JAR.

      gcloud storage buckets create BUCKET_NAME
      
      gcloud storage cp target/word-count-beam-bundled-0.1.jar gs://BUCKET_NAME/
      
    2. On your Dataproc's master node, download the uber JAR.

      gcloud storage cp gs://BUCKET_NAME/word-count-beam-bundled-0.1.jar .
      
  7. Run the Java Beam job on the Dataproc cluster's master node.

    flink run -c org.apache.beam.examples.WordCount word-count-beam-bundled-0.1.jar \
        --runner=FlinkRunner \
        --output=gs://BUCKET_NAME/java-wordcount-out
    
  8. Check that the results were written to your Cloud Storage bucket.

    gcloud storage cat gs://BUCKET_NAME/java-wordcount-out-SHARD_ID
    
  9. Stop the Flink YARN session.

    yarn application -list
    
    yarn application -kill YARN_APPLICATION_ID
    

Portable Beam Jobs

To run Beam jobs written in Python, Go, and other supported languages, you can use the FlinkRunner and PortableRunner as described on the Beam's Flink Runner page (also see Portability Framework Roadmap).

The following example runs a portable Beam job in Python from the Dataproc cluster's master node.

  1. Create a Dataproc cluster with both the Flink and Docker components enabled.

    gcloud dataproc clusters create CLUSTER_NAME \
        --optional-components=FLINK,DOCKER \
        --image-version=DATAPROC_IMAGE_VERSION \
        --region=REGION \
        --enable-component-gateway \
        --scopes=https://www.googleapis.com/auth/cloud-platform
    

    Notes:

    • --optional-components: Flink and Docker.
    • --image-version: The cluster's image version, which determines the Flink version installed on the cluster (for example, see the Apache Flink component versions listed for the latest and previous four 2.0.x image release versions).
    • --region: An available Dataproc region.
    • --enable-component-gateway: Enable access to the Flink Job Manager UI.
    • --scopes: Enable access to Google Cloud APIs by your cluster (see Scopes best practice). cloud-platform scope is enabled by default (you do not need to include this flag setting) when you create a cluster that uses Dataproc image version 2.1 or later.
  2. Use the gcloud CLI locally or in Cloud Shell to create a Cloud Storage bucket. You will specify the BUCKET_NAME when you run a sample wordcount program.

    gcloud storage buckets create BUCKET_NAME
    
  3. In a terminal window on the cluster VM, start a Flink YARN session. Note the Flink master URL, the address of the Flink master where jobs are executed.. You will specify the FLINK_MASTER_URL when you run a sample wordcount program.

    . /usr/bin/flink-yarn-daemon
    

    Display and note the Flink version running the Dataproc cluster. You will specify the FLINK_VERSION when you run a sample wordcount program.

    flink --version
    
  4. Install Python libraries needed for the job on the cluster master node.

  5. Install a Beam version that is compatible with the Flink version on the cluster.

    python -m pip install apache-beam[gcp]==BEAM_VERSION
    
  6. Run the word count example on the cluster master node.

    python -m apache_beam.examples.wordcount \
        --runner=FlinkRunner \
        --flink_version=FLINK_VERSION \
        --flink_master=FLINK_MASTER_URL
        --flink_submit_uber_jar \
        --output=gs://BUCKET_NAME/python-wordcount-out
    

    Notes:

    • --runner: FlinkRunner.
    • --flink_version: FLINK_VERSION, noted earlier.
    • --flink_master: FLINK_MASTER_URL, noted earlier.
    • --flink_submit_uber_jar: Use the uber JAR to execute the Beam job.
    • --output: BUCKET_NAME, created earlier.
  7. Verify that results were written to your bucket.

    gcloud storage cat gs://BUCKET_NAME/python-wordcount-out-SHARD_ID
    
  8. Stop the Flink YARN session.

    1. Get the application ID.
    yarn application -list
    
    1. Insert the <var>YARN_APPLICATION_ID</var>, then stop the session.
    
    yarn application -kill 
    

The Dataproc Flink component supports Kerberized clusters. A valid Kerberos ticket is needed to submit and persist a Flink job or to start a Flink cluster. By default, a Kerberos ticket remains valid for seven days.

The Flink Job Manager web interface is available while a Flink job or Flink session cluster is running. To use the web interface:

  1. Create a Dataproc Flink cluster.
  2. After cluster creation, click the Component Gateway YARN ResourceManager link on the Web Interface tab on the Cluster details page in the Google Cloud console.
  3. On the YARN Resource Manager UI, identify the Flink cluster application entry. Depending on a job's completion status, an ApplicationMaster or History link will be listed.
  4. For a long-running streaming job, click the ApplicationManager link to open the Flink dashboard; for a completed job, click the History link to view job details.