Google Cloud Dataflow Template Pipelines

These Dataflow templates are an effort to solve simple, but large, in-Cloud data tasks, including data import/export/backup/restore and bulk API operations, without a development environment. The technology under the hood which makes these operations possible is the Google Cloud Dataflow service combined with a set of Apache Beam SDK templated pipelines.

Google is providing this collection of pre-implemented Dataflow templates as a reference and to provide easy customization for developers wanting to extend their functionality.

Note on Default Branch

As of November 18, 2021, our default branch is now named "main". This does not affect forks. If you would like your fork and its local clone to reflect these changes you can follow GitHub's branch renaming guide .

Building

Maven commands should be run on the parent POM. An example would be:

mvn clean package -pl v2/pubsub-binary-to-bigquery -am

Template Pipelines

• BigQuery to Bigtable
• BigQuery to Datastore
• BigQuery to TFRecords
• Bigtable to GCS Avro
• Bulk Compressor
• Bulk Decompressor
• Datastore Bulk Delete *
• Datastore to BigQuery
• Datastore to GCS Text *
• Datastore to Pub/Sub *
• Datastore Unique Schema Count
• DLP Text to BigQuery (Streaming)
• File Format Conversion
• GCS Avro to Bigtable
• GCS Avro to Spanner
• GCS Text to Spanner
• GCS Text to BigQuery *
• GCS Text to Datastore
• GCS Text to Pub/Sub (Batch)
• GCS Text to Pub/Sub (Streaming)
• Jdbc to BigQuery
• Pub/Sub to BigQuery *
• Pub/Sub to Datastore *
• Pub/Sub to GCS Avro
• Pub/Sub to GCS Text
• Pub/Sub to Pub/Sub
• Pub/Sub to Splunk *
• Spanner to GCS Avro
• Spanner to GCS Text
• Word Count

* Supports user-defined functions (UDFs).

For documentation on each template's usage and parameters, please see the official docs .

Getting Started

Requirements

• Java 11
• Maven 3

Building the Project

Build the entire project using the maven compile command.

mvn clean compile

Building/Testing from IntelliJ

IntelliJ, by default, will often skip necessary Maven goals, leading to build failures. You can fix these in the Maven view by going to Module_Name > Plugins > Plugin_Name where Module_Name and Plugin_Name are the names of the respective module and plugin with the rule. From there, right-click the rule and select "Execute Before Build".

The list of known rules that require this are:

• common > Plugins > protobuf > protobuf:compile
• common > Plugins > protobuf > protobuf:test-compile

Formatting Code

From either the root directory or v2/ directory, run:

mvn spotless:apply

This will format the code and add a license header. To verify that the code is formatted correctly, run:

mvn spotless:check

The directory to run the commands from is based on whether the changes are under v2/ or not.

Creating a Template File

Dataflow templates can be created using a Maven command which builds the project and stages the template file on Google Cloud Storage. Any parameters passed at template build time will not be able to be overwritten at execution time.

mvn compile exec:java \
-Dexec.mainClass=com.google.cloud.teleport.templates.<template-class> \
-Dexec.cleanupDaemonThreads=false \
-Dexec.args=" \
--project=<project-id> \
--stagingLocation=gs://<bucket-name>/staging \
--tempLocation=gs://<bucket-name>/temp \
--templateLocation=gs://<bucket-name>/templates/<template-name>.json \
--runner=DataflowRunner"

Executing a Template File

Once the template is staged on Google Cloud Storage, it can then be executed using the gcloud CLI tool. The runtime parameters required by the template can be passed in the parameters field via comma-separated list of paramName=Value.

gcloud dataflow jobs run <job-name> \
--gcs-location=<template-location> \
--zone=<zone> \
--parameters <parameters>

Metadata Annotations

A template requires more information than just a name and description. For example, in order to be used from the Dataflow UI, parameters need a longer help text to guide users, as well as proper types and validations to make sure parameters are being passed correctly.

We introduced annotations to have the source code as a single source of truth, along with a set of utilities / plugins to generate template-accompanying artifacts (such as command specs, parameter specs).

@Template Annotation

Every template must be annotated with @Template. Existing templates can be used for reference, but the structure is as follows:

@Template(
    name = "BigQuery_to_Elasticsearch",
    category = TemplateCategory.BATCH,
    displayName = "BigQuery to Elasticsearch",
    description = "A pipeline which sends BigQuery records into an Elasticsearch instance as JSON documents.",
    optionsClass = BigQueryToElasticsearchOptions.class,
    flexContainerName = "bigquery-to-elasticsearch")
public class BigQueryToElasticsearch {

@TemplateParameter Annotation

A set of @TemplateParameter.{Type} annotations were created to allow the definition of options for a template, and the proper rendering in the UI, and validations by the template launch service. Examples can be found in the repository, but the general structure is as follows:

@TemplateParameter.Text(
    order = 2,
    optional = false,
    regexes = {"[a-zA-Z0-9._-,]+"},
    description = "Kafka topic(s) to read the input from",
    helpText = "Kafka topic(s) to read the input from.",
    example = "topic1,topic2")
@Validation.Required
String getInputTopics();

@TemplateParameter.GcsReadFile(
    order = 1,
    description = "Cloud Storage Input File(s)",
    helpText = "Path of the file pattern glob to read from.",
    example = "gs://your-bucket/path/*.csv")
String getInputFilePattern();

@TemplateParameter.Boolean(
    order = 11,
    optional = true,
    description = "Whether to use column alias to map the rows.",
    helpText = "If enabled (set to true) the pipeline will consider column alias (\"AS\") instead of the column name to map the rows to BigQuery.")
@Default.Boolean(false)
Boolean getUseColumnAlias();

@TemplateParameter.Enum(
    order = 21,
    enumOptions = {"INDEX", "CREATE"},
    optional = true,
    description = "Build insert method",
    helpText = "Whether to use INDEX (index, allows upsert) or CREATE (create, errors on duplicate _id) with Elasticsearch bulk requests.")
@Default.Enum("CREATE")
BulkInsertMethodOptions getBulkInsertMethod();

Note: order is relevant for templates that can be used from the UI, and specify the relative order of parameters.

@TemplateIntegrationTest Annotation

This annotation should be used by classes that are used for integration tests of other templates. This is used to wire a specific IT class with a template, and allows environment preparation / proper template staging before tests are executed on Dataflow.

Template tests have to follow this general format (please note the @TemplateIntegrationTest annotation and the TemplateTestBase super-class):

@TemplateIntegrationTest(PubsubToText.class)
@RunWith(JUnit4.class)
public final class PubsubToTextIT extends TemplateTestBase {

Please refer to Templates Plugin (Beta) to use and validate such annotations.

Templates Plugin (Beta)

Templates plugin was created to make the workflow of creating, testing and releasing Templates much simpler.

Before using the plugin, please make sure that the gcloud CLI is installed and up-to-date, and that the client is properly authenticated using:

gcloud init
gcloud auth application-default login

After authenticated, install the plugin into your local repository:

mvn install -pl plugins/templates-maven-plugin -am

Validating annotations

To validate the templates in the current context, the profile -PtemplatesValidate should be activated on Maven commands. For example, use the following commands:

Validate in all modules:

mvn compile -PtemplatesValidate

Validate in a specific module (for example, v2/googlecloud-to-googlecloud):

mvn compile \
  -PtemplatesValidate \
  -pl v2/googlecloud-to-googlecloud -am

Generating Specs

The spec files can be generated by using -PtemplatesSpec. For example:

mvn package \
  -PtemplatesSpec \
  -DskipTests \
  -pl v2/googlecloud-to-googlecloud -am

Specs are generated in the folder target, with the name {template}-spec-generated-metadata.json. Please verify if all parameters are listed correctly, following the desired order.

Deploying/Staging Templates

To deploy a Template, it is necessary to "stage" them to Artifact Registry ( Flex) and Cloud Storage. Even though there are different steps that depend on the kind of template being developed. The plugin allows a template to be staged using the following single command:

mvn package -PtemplatesStage  \
  -DskipTests \
  -DprojectId="{projectId}" \
  -DbucketName="{bucketName}" \
  -DstagePrefix="2022-10-16-test" \
  -DtemplateName="Cloud_PubSub_to_GCS_Text_Flex" \
  -pl v2/googlecloud-to-googlecloud -am

Note: if -DtemplateName is not specified, all templates for the module will be staged.

Running a Template

A template can also be executed on Dataflow, directly from the command line. The command-line is similar to staging a template, but it is required to specify -Dparameters with the parameters that will be used when launching the template. For example:

mvn package -PtemplatesRun \
  -DskipTests \
  -DprojectId="{projectId}" \
  -DbucketName="{bucketName}" \
  -Dregion="us-central1" \
  -DtemplateName="Cloud_PubSub_to_GCS_Text_Flex" \
  -Dparameters="inputTopic=projects/{projectId}/topics/{topicName},windowDuration=15s,outputDirectory=gs://{outputDirectory}/out,outputFilenamePrefix=output-,outputFilenameSuffix=.txt" \
  -pl v2/googlecloud-to-googlecloud -am

Notes

• When running a template, -DtemplateName is mandatory, as -Dparameters= are different across templates.
• -PtemplatesRun is self-contained, i.e., it is not required to run Deploying/Staging Templates before. In case you want to run a previously staged template, the existing path can be provided as -DspecPath=gs://.../path
• -DjobName="{name}" may be informed if a specific name is desirable (optional).

Using UDFs

User-defined functions (UDFs) allow you to customize a template's functionality by providing a short JavaScript function without having to maintain the entire codebase. This is useful in situations which you'd like to rename fields, filter values, or even transform data formats before output to the destination. All UDFs are executed by providing the payload of the element as a string to the JavaScript function. You can then use JavaScript's in-built JSON parser or other system functions to transform the data prior to the pipeline's output. The return statement of a UDF specifies the payload to pass forward in the pipeline. This should always return a string value. If no value is returned or the function returns undefined, the incoming record will be filtered from the output.

UDF Function Specification

Template	UDF Input Type	Input Description	UDF Output Type	Output Description
Datastore Bulk Delete	String	A JSON string of the entity	String	A JSON string of the entity to delete; filter entities by returning undefined
Datastore to Pub/Sub	String	A JSON string of the entity	String	The payload to publish to Pub/Sub
Datastore to GCS Text	String	A JSON string of the entity	String	A single-line within the output file
GCS Text to BigQuery	String	A single-line within the input file	String	A JSON string which matches the destination table's schema
Pub/Sub to BigQuery	String	A string representation of the incoming payload	String	A JSON string which matches the destination table's schema
Pub/Sub to Datastore	String	A string representation of the incoming payload	String	A JSON string of the entity to write to Datastore
Pub/Sub to Splunk	String	A string representation of the incoming payload	String	The event data to be sent to Splunk HEC events endpoint. Must be a string or a stringified JSON object

UDF Examples

Adding fields

/**
 * A transform which adds a field to the incoming data.
 * @param {string} inJson
 * @return {string} outJson
 */
function transform(inJson) {
  var obj = JSON.parse(inJson);
  obj.dataFeed = "Real-time Transactions";
  obj.dataSource = "POS";
  return JSON.stringify(obj);
}

Filtering records

/**
 * A transform function which only accepts 42 as the answer to life.
 * @param {string} inJson
 * @return {string} outJson
 */
function transform(inJson) {
  var obj = JSON.parse(inJson);
  // only output objects which have an answer to life of 42.
  if (obj.hasOwnProperty('answerToLife') && obj.answerToLife === 42) {
    return JSON.stringify(obj);
  }
}