Enable Data Jobs Monitoring for Spark on Google Cloud Dataproc

Data Jobs Monitoring gives visibility into the performance and reliability of Apache Spark applications on Google Cloud Dataproc.

Requirements

This guide is for Dataproc clusters on Compute Engine. If you are using Dataproc on GKE, refer to the Kubernetes Installation Guide instead.

Dataproc Release 2.0.x or later is required. All of Debian, Rocky Linux, and Ubuntu Dataproc standard images are supported.

Setup

Follow these steps to enable Data Jobs Monitoring for GCP Dataproc.

  1. Store your Datadog API key in GCP Secret Manager (recommended).
  2. Create and configure your Dataproc cluster.
  3. Specify service tagging per Spark application.
  1. Take note of your Datadog API key.
  2. In GCP Secret Manager, choose Create secret.
    • Under Name, enter a Secret name. You can use dd_api_key.
    • Under Secret value, paste your Datadog API key in the Secret value text box.
      A section of the secret creation page titled 'Secret details'. On the top, a name field containing 'dd_api_key'. On the bottom, a text box to paste your own API key.
    • Click Create Secret.
  3. Optionally, under Rotation, you can turn on automatic rotation.
  4. In GCP Secret Manager, open the secret you created. Take note of the Resource ID, which is in the format “projects/<PROJECT_NAME>/secrets/<SECRET_NAME>”.
  5. Make sure the service account used by your Dataproc cluster has permission to read the secret. By default, this is the Compute Engine default service account. To grant access, copy the associated service account Principal, and click Grant Access on the Permissions tab of the secret’s page. Assign the secretmanager.secretAccessor role, or any other one that has secretmanager.versions.access permission. See the IAM roles documentation for a full description of available roles.

Create and configure your Dataproc cluster

When you create a new Dataproc Cluster on Compute Engine in the Google Cloud Console, add an initialization action on the Customize cluster page:

  1. Save the following script to a GCS bucket that your Dataproc cluster can read. Take note of the path to this script.

    #!/bin/bash

# Set required parameter DD_SITE
export DD_SITE=

# Set required parameter DD_API_KEY with Datadog API key.
# The commands below assumes the API key is stored in GCP Secret Manager, with the secret name as dd_api_key and the project <PROJECT_NAME>.
# IMPORTANT: Modify if you choose to manage and retrieve your secret differently.
# Change the project name, which you can find on the secrets page. The resource ID is in the format "projects/<PROJECT_NAME>/secrets/<SECRET_NAME>".
PROJECT_NAME=<PROJECT_NAME>
gcloud config set project $PROJECT_NAME
SECRET_NAME=dd_api_key
export DD_API_KEY=$(gcloud secrets versions access latest --secret $SECRET_NAME)

# Download and run the latest init script
curl -L https://install.datadoghq.com/scripts/install-dataproc.sh > djm-install-script; bash djm-install-script || true

    The script above sets the required parameters, and downloads and runs the latest init script for Data Jobs Monitoring in Dataproc. If you want to pin your script to a specific version, you can replace the filename in the URL with install-dataproc-0.9.1.sh to use version 0.9.1, for example. The source code used to generate this script can be found on the Datadog Agent repository.

  2. On the Customize cluster page, locate the Initialization Actions section. Enter the path where you saved the script from the previous step.

When your cluster is created, this initialization action installs the Datadog Agent and downloads the Java tracer on each node of the cluster.

Specify service tagging per Spark application

Tagging enables you to better filter, aggregate, and compare your telemetry in Datadog. You can configure tags by passing -Ddd.service, -Ddd.env, -Ddd.version, and -Ddd.tags options to your Spark driver and executor extraJavaOptions properties.

In Datadog, each job’s name corresponds to the value you set for -Ddd.service.

spark-submit \
 --conf spark.driver.extraJavaOptions="-Ddd.service=<JOB_NAME> -Ddd.env=<ENV> -Ddd.version=<VERSION> -Ddd.tags=<KEY_1>:<VALUE_1>,<KEY_2:VALUE_2>" \
 --conf spark.executor.extraJavaOptions="-Ddd.service=<JOB_NAME> -Ddd.env=<ENV> -Ddd.version=<VERSION> -Ddd.tags=<KEY_1>:<VALUE_1>,<KEY_2:VALUE_2>" \
 application.jar

Validation

In Datadog, view the Data Jobs Monitoring page to see a list of all your data processing jobs.

Troubleshooting

If you don’t see any data in DJM after installing the product, follow those steps.

The init script installs the Datadog Agent. To make sure it is properly installed, ssh into the cluster and run the Agent status command:

sudo datadog-agent status

Advanced Configuration

Tag spans at runtime

You can set tags on Spark spans at runtime. These tags are applied only to spans that start after the tag is added.

// Add tag for all next Spark computations
sparkContext.setLocalProperty("spark.datadog.tags.key", "value")
spark.read.parquet(...)

To remove a runtime tag:

// Remove tag for all next Spark computations
sparkContext.setLocalProperty("spark.datadog.tags.key", null)

Further Reading

Additional helpful documentation, links, and articles:

Data Jobs MonitoringDOCUMENTATION more more