Google Cloud to Neo4j template

The Google Cloud to Neo4j template lets you import a dataset into a Neo4j database through a Dataflow job, sourcing data from CSV files hosted in Cloud Storage buckets. It also lets you manipulate and transform the data at various steps of the import. You can use the template for both first-time imports and incremental imports.

Pipeline requirements

  • A running Neo4j instance
  • A Cloud Storage bucket
  • A dataset to import, in the form of CSV files
  • A job specification file to use

Create a job specification file

The job specification file consists of a JSON object with the following sections:

  • config: global flags affecting how the import is performed.
  • sources: data source definitions (relational).
  • targets: data target definitions (graph: nodes/relationships).
  • actions: pre/post-load actions.

For more information, see Create a job specification file in the Neo4j documentation.

Template parameters

Required parameters

  • jobSpecUri: The path to the job specification file, which contains the JSON description of data sources, Neo4j targets and actions.

Optional parameters

  • neo4jConnectionUri: The path to the Neo4j connection JSON file.
  • neo4jConnectionSecretId: The secret ID for the Neo4j connection metadata. You can use this value as an alternative to the neo4jConnectionUri.
  • optionsJson: A JSON object that is also called runtime tokens For example, {token1:value1,token2:value2}. Spec can refer to $token1 and $token2.. Defaults to empty.
  • readQuery: SQL query override. Defaults to empty.
  • inputFilePattern: The text file path override For example, gs://your-bucket/path/*.json. Defaults to empty.
  • disabledAlgorithms: Comma separated algorithms to disable. If this value is set to none, no algorithm is disabled. Use this parameter with caution, because the algorithms disabled by default might have vulnerabilities or performance issues. For example, SSLv3, RC4.
  • extraFilesToStage: Comma separated Cloud Storage paths or Secret Manager secrets for files to stage in the worker. These files are saved in the /extra_files directory in each worker. For example, gs://<BUCKET_NAME>/file.txt,projects/<PROJECT_ID>/secrets/<SECRET_ID>/versions/<VERSION_ID>.

Run the template

ConsolegcloudAPI
  1. Go to the Dataflow Create job from template page.
    2. Go to Create job from template
  2. In the Job name field, enter a unique job name.
  3. Optional: For Regional endpoint, select a value from the drop-down menu. The default region is us-central1.

    For a list of regions where you can run a Dataflow job, see Dataflow locations.

  4. From the Dataflow template drop-down menu, select the Google Cloud to Neo4j template.
  5. In the provided parameter fields, enter your parameter values.
  6. Click Run job.

In your shell or terminal, run the template:

gcloud dataflow flex-template run JOB_NAME \
    --template-file-gcs-location=gs://dataflow-templates-REGION_NAME/VERSION/flex/Google_Cloud_to_Neo4j \
    --project=PROJECT_ID \
    --region=REGION_NAME \
    --parameters \
       jobSpecUri=JOB_SPEC_URI,\
       neo4jConnectionUri=NEO4J_CONNECTION_URI,\

Replace the following:

  • JOB_NAME: a unique job name of your choice
  • VERSION: the version of the template that you want to use

    You can use the following values:

  • REGION_NAME: the region where you want to deploy your Dataflow job—for example, us-central1
  • JOB_SPEC_URI: the path to the job specification file
  • NEO4J_CONNECTION_URI: the path to the Neo4j connection metadata

To run the template using the REST API, send an HTTP POST request. For more information on the API and its authorization scopes, see projects.templates.launch.

POST https://dataflow.googleapis.com/v1b3/projects/PROJECT_ID/locations/LOCATION/flexTemplates:launch
{
   "launchParameter": {
     "jobName": "JOB_NAME",
     "parameters": {
       "jobSpecUri": "JOB_SPEC_URI",
       "neo4jConnectionUri": "NEO4J_CONNECTION_URI",
     },
     "containerSpecGcsPath": "gs://dataflow-templates-LOCATION/VERSION/flex/Google_Cloud_to_Neo4j",
     "environment": { "maxWorkers": "10" }
  }
}

Replace the following:

  • PROJECT_ID: the Google Cloud project ID where you want to run the Dataflow job
  • JOB_NAME: a unique job name of your choice
  • VERSION: the version of the template that you want to use

    You can use the following values:

  • LOCATION: the region where you want to deploy your Dataflow job—for example, us-central1
  • JOB_SPEC_URI: the path to the job specification file
  • NEO4J_CONNECTION_URI: the path to the Neo4j connection metadata
Java
/*
 * Copyright (C) 2021 Google LLC
 *
 * Licensed 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.
 */
package com.google.cloud.teleport.v2.neo4j.templates;

import static java.util.stream.Collectors.groupingBy;
import static java.util.stream.Collectors.mapping;
import static java.util.stream.Collectors.toList;

import com.google.cloud.teleport.metadata.Template;
import com.google.cloud.teleport.metadata.Template.AdditionalDocumentationBlock;
import com.google.cloud.teleport.metadata.TemplateCategory;
import com.google.cloud.teleport.v2.common.UncaughtExceptionLogger;
import com.google.cloud.teleport.v2.neo4j.actions.ActionDoFnFactory;
import com.google.cloud.teleport.v2.neo4j.actions.ActionPreloadFactory;
import com.google.cloud.teleport.v2.neo4j.actions.preload.PreloadAction;
import com.google.cloud.teleport.v2.neo4j.database.Neo4jConnection;
import com.google.cloud.teleport.v2.neo4j.model.InputValidator;
import com.google.cloud.teleport.v2.neo4j.model.Json;
import com.google.cloud.teleport.v2.neo4j.model.Json.ParsingResult;
import com.google.cloud.teleport.v2.neo4j.model.connection.ConnectionParams;
import com.google.cloud.teleport.v2.neo4j.model.enums.ArtifactType;
import com.google.cloud.teleport.v2.neo4j.model.helpers.JobSpecMapper;
import com.google.cloud.teleport.v2.neo4j.model.helpers.OptionsParamsMapper;
import com.google.cloud.teleport.v2.neo4j.model.helpers.TargetQuerySpec;
import com.google.cloud.teleport.v2.neo4j.model.helpers.TargetQuerySpec.TargetQuerySpecBuilder;
import com.google.cloud.teleport.v2.neo4j.model.helpers.TargetSequence;
import com.google.cloud.teleport.v2.neo4j.model.job.ActionContext;
import com.google.cloud.teleport.v2.neo4j.model.job.OptionsParams;
import com.google.cloud.teleport.v2.neo4j.options.Neo4jFlexTemplateOptions;
import com.google.cloud.teleport.v2.neo4j.providers.Provider;
import com.google.cloud.teleport.v2.neo4j.providers.ProviderFactory;
import com.google.cloud.teleport.v2.neo4j.transforms.Neo4jRowWriterTransform;
import com.google.cloud.teleport.v2.neo4j.utils.BeamBlock;
import com.google.cloud.teleport.v2.neo4j.utils.FileSystemUtils;
import com.google.cloud.teleport.v2.neo4j.utils.ModelUtils;
import com.google.cloud.teleport.v2.neo4j.utils.ProcessingCoder;
import com.google.cloud.teleport.v2.utils.SecretManagerUtils;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Set;
import java.util.stream.Collectors;
import java.util.stream.Stream;
import org.apache.beam.runners.dataflow.options.DataflowPipelineOptions;
import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.coders.VarIntCoder;
import org.apache.beam.sdk.io.FileSystems;
import org.apache.beam.sdk.options.PipelineOptionsFactory;
import org.apache.beam.sdk.schemas.Schema;
import org.apache.beam.sdk.transforms.Create;
import org.apache.beam.sdk.transforms.ParDo;
import org.apache.beam.sdk.transforms.Wait;
import org.apache.beam.sdk.values.PCollection;
import org.apache.beam.sdk.values.Row;
import org.apache.beam.sdk.values.TypeDescriptor;
import org.apache.commons.lang3.StringUtils;
import org.jetbrains.annotations.NotNull;
import org.neo4j.importer.v1.Configuration;
import org.neo4j.importer.v1.ImportSpecification;
import org.neo4j.importer.v1.actions.Action;
import org.neo4j.importer.v1.actions.ActionStage;
import org.neo4j.importer.v1.sources.Source;
import org.neo4j.importer.v1.targets.CustomQueryTarget;
import org.neo4j.importer.v1.targets.NodeTarget;
import org.neo4j.importer.v1.targets.RelationshipTarget;
import org.neo4j.importer.v1.targets.Target;
import org.neo4j.importer.v1.targets.TargetType;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

/**
 * Dataflow template which reads Google Cloud data (Text, BigQuery) and writes it to Neo4j.
 *
 * <p>In case of BigQuery, the source data can be either a table or a SQL query.
 *
 * <p>Check out <a
 * href="https://github.com/GoogleCloudPlatform/DataflowTemplates/blob/main/v2/googlecloud-to-neo4j/README_Google_Cloud_to_Neo4j.md">README</a>
 * for instructions on how to use or modify this template.
 */
@Template(
    name = "Google_Cloud_to_Neo4j",
    category = TemplateCategory.BATCH,
    displayName = "Google Cloud to Neo4j",
    description =
        "The Google Cloud to Neo4j template lets you import a dataset into a Neo4j database through a Dataflow job, "
            + "sourcing data from CSV files hosted in Google Cloud Storage buckets. It also lets you to manipulate and transform the data "
            + "at various steps of the import. You can use the template for both first-time imports and incremental imports.",
    optionsClass = Neo4jFlexTemplateOptions.class,
    flexContainerName = "googlecloud-to-neo4j",
    contactInformation = "https://support.neo4j.com/",
    documentation =
        "https://cloud.google.com/dataflow/docs/guides/templates/provided/google-cloud-to-neo4j",
    requirements = {
      "A running Neo4j instance",
      "A Google Cloud Storage bucket",
      "A dataset to import, in the form of CSV files",
      "A job specification file to use"
    },
    additionalDocumentation = {
      @AdditionalDocumentationBlock(
          name = "Create a job specification file",
          content = {
            "The job specification file consists of a JSON object with the following sections:\n"
                + "- `config` - global flags affecting how the import is performed.\n"
                + "- `sources` - data source definitions (relational).\n"
                + "- `targets` - data target definitions (graph: nodes/relationships/custom queries).\n"
                + "- `actions` - pre/post-load actions.\n"
                + "For more information, see <a href=\"https://neo4j.com/docs/dataflow-google-cloud/job-specification/\" class=\"external\">Create a job specification file</a> in the Neo4j documentation."
          })
    },
    preview = true)
public class GoogleCloudToNeo4j {

  private static final Logger LOG = LoggerFactory.getLogger(GoogleCloudToNeo4j.class);

  private final OptionsParams optionsParams;
  private final ConnectionParams neo4jConnection;
  private final ImportSpecification importSpecification;
  private final Configuration globalSettings;
  private final Pipeline pipeline;
  private final String templateVersion;
  private final TargetSequence targetSequence = new TargetSequence();

  /**
   * Main class for template. Initializes job using run-time on pipelineOptions.
   *
   * @param pipelineOptions framework supplied arguments
   */
  public GoogleCloudToNeo4j(Neo4jFlexTemplateOptions pipelineOptions) {

    ////////////////////////////
    // Job name gets a date on it when running within the container, but not with DirectRunner
    // final String jobName = pipelineOptions.getJobName() + "-" + System.currentTimeMillis();
    // pipelineOptions.setJobName(jobName);

    // Set pipeline options
    this.pipeline = Pipeline.create(pipelineOptions);
    FileSystems.setDefaultPipelineOptions(pipelineOptions);
    this.optionsParams = OptionsParamsMapper.fromPipelineOptions(pipelineOptions);

    // Validate pipeline
    processValidations(
        "Errors found validating pipeline options: ",
        InputValidator.validateNeo4jPipelineOptions(pipelineOptions));

    this.templateVersion = readTemplateVersion(pipelineOptions);

    String neo4jConnectionJson = readConnectionSettings(pipelineOptions);
    ParsingResult parsingResult = InputValidator.validateNeo4jConnection(neo4jConnectionJson);
    if (!parsingResult.isSuccessful()) {
      processValidations(
          "Errors found validating Neo4j connection: ",
          parsingResult.formatErrors("Could not validate connection JSON"));
    }
    this.neo4jConnection = Json.map(parsingResult, ConnectionParams.class);

    this.importSpecification = JobSpecMapper.parse(pipelineOptions.getJobSpecUri(), optionsParams);
    globalSettings = importSpecification.getConfiguration();

    ///////////////////////////////////

    // Source specific validations
    for (Source source : importSpecification.getSources()) {
      // get provider implementation for source
      Provider providerImpl = ProviderFactory.of(source, targetSequence);
      providerImpl.configure(optionsParams);
    }
  }

  private static String readTemplateVersion(Neo4jFlexTemplateOptions options) {
    Map<String, String> labels = options.as(DataflowPipelineOptions.class).getLabels();
    String defaultVersion = "UNKNOWN";
    if (labels == null) {
      return defaultVersion;
    }
    return labels.getOrDefault("goog-dataflow-provided-template-version", defaultVersion);
  }

  private static String readConnectionSettings(Neo4jFlexTemplateOptions options) {
    String secretId = options.getNeo4jConnectionSecretId();
    if (StringUtils.isNotEmpty(secretId)) {
      return SecretManagerUtils.getSecret(secretId);
    }
    String uri = options.getNeo4jConnectionUri();
    try {
      return FileSystemUtils.getPathContents(uri);
    } catch (Exception e) {
      throw new RuntimeException(
          String.format("Unable to read Neo4j configuration at URI %s: ", uri), e);
    }
  }

  /**
   * Runs a pipeline which reads data from various sources and writes it to Neo4j.
   *
   * @param args arguments to the pipeline
   */
  public static void main(String[] args) {
    UncaughtExceptionLogger.register();

    Neo4jFlexTemplateOptions options =
        PipelineOptionsFactory.fromArgs(args).withValidation().as(Neo4jFlexTemplateOptions.class);

    // Allow users to supply their own list of disabled algorithms if necessary
    if (StringUtils.isBlank(options.getDisabledAlgorithms())) {
      options.setDisabledAlgorithms(
          "SSLv3, RC4, DES, MD5withRSA, DH keySize < 1024, EC keySize < 224, 3DES_EDE_CBC, anon,"
              + " NULL");
    }

    LOG.info("Job: {}", options.getJobSpecUri());
    GoogleCloudToNeo4j template = new GoogleCloudToNeo4j(options);
    template.run();
  }

  /** Raises RuntimeExceptions for validation errors. */
  private void processValidations(String description, List<String> validationMessages) {
    StringBuilder sb = new StringBuilder();
    if (!validationMessages.isEmpty()) {
      for (String msg : validationMessages) {
        sb.append(msg);
        sb.append(System.lineSeparator());
      }
      throw new RuntimeException(description + " " + sb);
    }
  }

  public void run() {

    try (Neo4jConnection directConnect =
        new Neo4jConnection(this.neo4jConnection, this.templateVersion)) {
      boolean resetDb = globalSettings.get(Boolean.class, "reset_db").orElse(false);
      if (!resetDb) {
        directConnect.verifyConnectivity();
      } else {
        directConnect.resetDatabase();
      }
    }

    ////////////////////////////
    // If an action transformation has no upstream PCollection, it will use this default context
    PCollection<Row> defaultActionContext =
        pipeline.apply(
            "Default Context",
            Create.empty(TypeDescriptor.of(Row.class)).withCoder(ProcessingCoder.of()));

    var processingQueue = new BeamBlock(defaultActionContext);

    runPreloadActions(findActionsByStage(ActionStage.START).collect(toList()));

    Map<ActionStage, List<PCollection<?>>> preActionRows =
        findActionsByStages(
                Set.of(
                    ActionStage.PRE_NODES, ActionStage.PRE_RELATIONSHIPS, ActionStage.PRE_QUERIES))
            .map(action -> Map.entry(action.getStage(), runAction(action, defaultActionContext)))
            .collect(
                groupingBy(
                    Entry::getKey, mapping(Entry::getValue, Collectors.<PCollection<?>>toList())));
    var sourceRows = new ArrayList<PCollection<?>>(importSpecification.getSources().size());
    var targetRows = new HashMap<TargetType, List<PCollection<?>>>(targetCount());
    var allActiveTargets =
        importSpecification.getTargets().getAll().stream()
            .filter(Target::isActive)
            .collect(toList());
    var allActiveNodeTargets =
        importSpecification.getTargets().getNodes().stream()
            .filter(Target::isActive)
            .collect(toList());

    ////////////////////////////
    // Process sources
    for (var source : importSpecification.getSources()) {
      String sourceName = source.getName();
      var activeSourceTargets =
          allActiveTargets.stream()
              .filter(target -> target.getSource().equals(sourceName))
              .collect(toList());
      if (activeSourceTargets.isEmpty()) {
        return;
      }

      // get provider implementation for source
      Provider provider = ProviderFactory.of(source, targetSequence);
      provider.configure(optionsParams);
      PCollection<Row> sourceMetadata =
          pipeline.apply(
              String.format("Metadata for source %s", sourceName), provider.queryMetadata());
      sourceRows.add(sourceMetadata);
      Schema sourceBeamSchema = sourceMetadata.getSchema();
      processingQueue.addToQueue(ArtifactType.source, sourceName, defaultActionContext);

      ////////////////////////////
      // Optimization: if some of the current source's targets either
      // - do not alter the source query (i.e. define no transformations)
      // - or the source provider does not support SQL pushdown
      // then the source PCollection can be defined here and reused across all the relevant targets
      PCollection<Row> nullableSourceBeamRows = null;
      if (!provider.supportsSqlPushDown()
          || activeSourceTargets.stream()
              .anyMatch(target -> !ModelUtils.targetHasTransforms(target))) {
        nullableSourceBeamRows =
            pipeline
                .apply("Query " + sourceName, provider.querySourceBeamRows(sourceBeamSchema))
                .setRowSchema(sourceBeamSchema);
      }

      List<NodeTarget> nodeTargets = getTargetsByType(activeSourceTargets, TargetType.NODE);
      for (NodeTarget target : nodeTargets) {
        TargetQuerySpec targetQuerySpec =
            new TargetQuerySpecBuilder()
                .sourceBeamSchema(sourceBeamSchema)
                .nullableSourceRows(nullableSourceBeamRows)
                .target(target)
                .build();
        String nodeStepDescription =
            targetSequence.getSequenceNumber(target)
                + ": "
                + sourceName
                + "->"
                + target.getName()
                + " nodes";
        PCollection<Row> preInsertBeamRows =
            pipeline.apply(
                "Query " + nodeStepDescription, provider.queryTargetBeamRows(targetQuerySpec));

        List<PCollection<?>> dependencies =
            new ArrayList<>(preActionRows.getOrDefault(ActionStage.PRE_NODES, List.of()));
        dependencies.add(
            processingQueue.resolveOutputs(target.getDependencies(), nodeStepDescription));

        PCollection<Row> blockingReturn =
            preInsertBeamRows
                .apply(
                    "** Unblocking "
                        + nodeStepDescription
                        + "(after "
                        + String.join(", ", target.getDependencies())
                        + " and pre-nodes actions)",
                    Wait.on(dependencies))
                .setCoder(preInsertBeamRows.getCoder())
                .apply(
                    "Writing " + nodeStepDescription,
                    new Neo4jRowWriterTransform(
                        importSpecification,
                        neo4jConnection,
                        templateVersion,
                        targetSequence,
                        target))
                .setCoder(preInsertBeamRows.getCoder());

        targetRows
            .computeIfAbsent(TargetType.NODE, (type) -> new ArrayList<>(nodeTargets.size()))
            .add(blockingReturn);

        processingQueue.addToQueue(ArtifactType.node, target.getName(), blockingReturn);
      }

      ////////////////////////////
      // Write relationship targets
      List<RelationshipTarget> relationshipTargets =
          getTargetsByType(activeSourceTargets, TargetType.RELATIONSHIP);
      for (var target : relationshipTargets) {
        var targetQuerySpec =
            new TargetQuerySpecBuilder()
                .nullableSourceRows(nullableSourceBeamRows)
                .sourceBeamSchema(sourceBeamSchema)
                .target(target)
                .startNodeTarget(
                    findNodeTargetByName(allActiveNodeTargets, target.getStartNodeReference()))
                .endNodeTarget(
                    findNodeTargetByName(allActiveNodeTargets, target.getEndNodeReference()))
                .build();
        String relationshipStepDescription =
            targetSequence.getSequenceNumber(target)
                + ": "
                + sourceName
                + "->"
                + target.getName()
                + " edges";
        PCollection<Row> preInsertBeamRows;
        if (ModelUtils.targetHasTransforms(target)) {
          preInsertBeamRows =
              pipeline.apply(
                  "Query " + relationshipStepDescription,
                  provider.queryTargetBeamRows(targetQuerySpec));
        } else {
          preInsertBeamRows = nullableSourceBeamRows;
        }

        List<PCollection<?>> dependencies =
            new ArrayList<>(preActionRows.getOrDefault(ActionStage.PRE_RELATIONSHIPS, List.of()));
        Set<String> dependencyNames = new LinkedHashSet<>(target.getDependencies());
        dependencyNames.add(target.getStartNodeReference());
        dependencyNames.add(target.getEndNodeReference());
        dependencies.add(
            processingQueue.resolveOutputs(dependencyNames, relationshipStepDescription));

        PCollection<Row> blockingReturn =
            preInsertBeamRows
                .apply(
                    "** Unblocking "
                        + relationshipStepDescription
                        + "(after "
                        + String.join(", ", dependencyNames)
                        + " and pre-relationships actions)",
                    Wait.on(dependencies))
                .setCoder(preInsertBeamRows.getCoder())
                .apply(
                    "Writing " + relationshipStepDescription,
                    new Neo4jRowWriterTransform(
                        importSpecification,
                        neo4jConnection,
                        templateVersion,
                        targetSequence,
                        target))
                .setCoder(preInsertBeamRows.getCoder());

        targetRows
            .computeIfAbsent(
                TargetType.RELATIONSHIP, (type) -> new ArrayList<>(relationshipTargets.size()))
            .add(blockingReturn);
        // serialize relationships
        processingQueue.addToQueue(ArtifactType.edge, target.getName(), blockingReturn);
      }
      ////////////////////////////
      // Custom query targets
      List<CustomQueryTarget> customQueryTargets =
          getTargetsByType(activeSourceTargets, TargetType.QUERY);
      for (Target target : customQueryTargets) {
        String customQueryStepDescription =
            targetSequence.getSequenceNumber(target)
                + ": "
                + sourceName
                + "->"
                + target.getName()
                + " (custom query)";

        List<PCollection<?>> dependencies =
            new ArrayList<>(preActionRows.getOrDefault(ActionStage.PRE_QUERIES, List.of()));
        dependencies.add(
            processingQueue.resolveOutputs(target.getDependencies(), customQueryStepDescription));

        // note: nullableSourceBeamRows is guaranteed to be non-null here since custom query targets
        // cannot define source transformations
        PCollection<Row> blockingReturn =
            nullableSourceBeamRows
                .apply(
                    "** Unblocking "
                        + customQueryStepDescription
                        + "(after "
                        + String.join(", ", target.getDependencies())
                        + ")",
                    Wait.on(dependencies))
                .setCoder(nullableSourceBeamRows.getCoder())
                .apply(
                    "Writing " + customQueryStepDescription,
                    new Neo4jRowWriterTransform(
                        importSpecification,
                        neo4jConnection,
                        templateVersion,
                        targetSequence,
                        target))
                .setCoder(nullableSourceBeamRows.getCoder());

        targetRows
            .computeIfAbsent(TargetType.QUERY, (type) -> new ArrayList<>(customQueryTargets.size()))
            .add(blockingReturn);
        processingQueue.addToQueue(ArtifactType.custom_query, target.getName(), blockingReturn);
      }
    }

    // Process POST-* actions, gather outputs and run END actions
    List<PCollection<?>> endActionDependencies =
        findActionsByStage(ActionStage.POST_SOURCES)
            .map(action -> runAction(action, defaultActionContext, sourceRows))
            .collect(Collectors.toCollection(ArrayList::new));
    endActionDependencies.addAll(
        findActionsByStage(ActionStage.POST_NODES)
            .map(
                action ->
                    runAction(
                        action,
                        defaultActionContext,
                        targetRows.getOrDefault(TargetType.NODE, List.of())))
            .collect(toList()));
    endActionDependencies.addAll(
        findActionsByStage(ActionStage.POST_RELATIONSHIPS)
            .map(
                action ->
                    runAction(
                        action,
                        defaultActionContext,
                        targetRows.getOrDefault(TargetType.RELATIONSHIP, List.of())))
            .collect(toList()));
    endActionDependencies.addAll(
        findActionsByStage(ActionStage.POST_QUERIES)
            .map(
                action ->
                    runAction(
                        action,
                        defaultActionContext,
                        targetRows.getOrDefault(TargetType.QUERY, List.of())))
            .collect(toList()));
    findActionsByStage(ActionStage.END)
        .map(action -> runAction(action, defaultActionContext, endActionDependencies))
        .forEach(GoogleCloudToNeo4j::noOp);

    // For a Dataflow Flex Template, do NOT waitUntilFinish().
    pipeline.run();
  }

  private PCollection<Row> runAction(Action action, PCollection<Row> defaultActionContext) {
    return runAction(action, defaultActionContext, List.of());
  }

  private PCollection<Row> runAction(
      Action action, PCollection<Row> defaultActionContext, List<PCollection<?>> dependencies) {
    var actionName = action.getName();
    return pipeline
        .apply(String.format("** Setup %s", actionName), Create.of(1))
        .apply(
            String.format("** Wait on %s dependencies", action.getStage()), Wait.on(dependencies))
        .setCoder(VarIntCoder.of())
        .apply(
            String.format("Running action %s", actionName),
            ParDo.of(ActionDoFnFactory.of(newActionContext(action))))
        .setCoder(defaultActionContext.getCoder());
  }

  private Stream<Action> findActionsByStage(ActionStage stage) {
    return findActionsByStages(Set.of(stage));
  }

  private Stream<Action> findActionsByStages(Set<ActionStage> stages) {
    return importSpecification.getActions().stream()
        .filter(action -> stages.contains(action.getStage()));
  }

  private void runPreloadActions(List<Action> actions) {
    for (Action action : actions) {
      LOG.debug("Executing START action: {}", action.getName());
      // Get targeted execution context
      ActionContext context = new ActionContext(action, neo4jConnection, templateVersion);
      PreloadAction actionImpl = ActionPreloadFactory.of(action, context);
      List<String> msgs = actionImpl.execute();
      for (String msg : msgs) {
        LOG.info("START action {} output: {}", action.getName(), msg);
      }
    }
  }

  @NotNull
  private ActionContext newActionContext(Action action) {
    return new ActionContext(action, this.neo4jConnection, this.templateVersion);
  }

  private static NodeTarget findNodeTargetByName(List<NodeTarget> nodes, String reference) {
    return nodes.stream()
        .filter(target -> reference.equals(target.getName()))
        .findFirst()
        .orElseThrow(
            () -> new IllegalArgumentException("Could not find active node target: " + reference));
  }

  @SuppressWarnings("unchecked")
  private <T extends Target> List<T> getTargetsByType(
      List<Target> activeSourceTargets, TargetType targetType) {
    return activeSourceTargets.stream()
        .filter(target -> target.getTargetType() == targetType)
        .map(target -> (T) target)
        .collect(toList());
  }

  private static <T> void noOp(T item) {}

  private int targetCount() {
    var targets = this.importSpecification.getTargets();
    return targets.getNodes().size()
        + targets.getRelationships().size()
        + targets.getCustomQueries().size();
  }
}

What's next