Bigtable to Cloud Storage Parquet 模板

Bigtable to Cloud Storage Parquet 模板是一种流水线，可从 Bigtable 表读取数据并以 Parquet 格式将其写入 Cloud Storage 存储桶。您可以使用该模板将数据从 Bigtable 移动到 Cloud Storage。

流水线要求

• Bigtable 表必须已存在。
• 在运行此流水线之前，输出 Cloud Storage 存储桶必须已存在。

模板参数

必需参数

• bigtableProjectId：您要从中读取数据的 Cloud Bigtable 实例所属的 Google Cloud 项目的 ID。
• bigtableInstanceId：表所属的 Cloud Bigtable 实例的 ID。
• bigtableTableId：要导出的 Cloud Bigtable 表的 ID。
• outputDirectory：用于写入输出文件的路径和文件名前缀。必须以斜杠结尾。日期时间格式用于解析日期和时间格式化程序的目录路径。例如：gs://your-bucket/your-path。
• filenamePrefix：Parquet 文件名的前缀。例如“table1-”。默认值为：part。

可选参数

• numShards：写入时生成的输出分片数上限。分片数越多，写入 Cloud Storage 的吞吐量越高，但处理输出 Cloud Storage 文件时跨分片聚合数据的费用也可能更高。默认值由 Dataflow 决定。
• bigtableAppProfileId：用于导出的 Bigtable 应用配置文件的 ID。如果您没有指定应用配置文件，Bigtable 将使用实例的默认应用配置文件：https://cloud.google.com/bigtable/docs/app-profiles#default-app-profile。

运行模板

控制台

1. 转到 Dataflow 基于模板创建作业页面。
转到“基于模板创建作业”
2. 在作业名称字段中，输入唯一的作业名称。
3. 可选：对于区域性端点，从下拉菜单中选择一个值。默认区域为 us-central1。

   如需查看可以在其中运行 Dataflow 作业的区域列表，请参阅 Dataflow 位置。

4. 从 Dataflow 模板下拉菜单中，选择 the Cloud Bigtable to Parquet Files on Cloud Storage template。
5. 在提供的参数字段中，输入您的参数值。
6. 点击运行作业。

gcloud

注意：如需使用 Google Cloud CLI 运行 classic模板，您必须具有 Google Cloud CLI 138.0.0 或更高版本。

在 shell 或终端中，运行模板：

gcloud dataflow jobs run JOB_NAME \
    --gcs-location gs://dataflow-templates-REGION_NAME/VERSION/Cloud_Bigtable_to_GCS_Parquet \
    --region REGION_NAME \
    --parameters \
bigtableProjectId=BIGTABLE_PROJECT_ID,\
bigtableInstanceId=INSTANCE_ID,\
bigtableTableId=TABLE_ID,\
outputDirectory=OUTPUT_DIRECTORY,\
filenamePrefix=FILENAME_PREFIX,\
numShards=NUM_SHARDS

替换以下内容：

• JOB_NAME：您选择的唯一性作业名称
• VERSION：您要使用的模板的版本

  您可使用以下值：

  • latest，以使用模板的最新版本，该模板在存储桶的未标示日期的父文件夹 (gs://dataflow-templates-REGION_NAME/latest/) 中可用
  • 版本名称（如 2023-09-12-00_RC00），以使用模板的特定版本，该版本嵌套在存储桶的相应日期父文件夹 (gs://dataflow-templates-REGION_NAME/) 中
  注意：最新版模板可能会随着重大更改而更新。为了防止这些重大更改影响您的生产工作流程，生产环境应使用有最近标示日期的父文件夹中保存的模板。
• REGION_NAME：要在其中部署 Dataflow 作业的区域，例如 us-central1
• BIGTABLE_PROJECT_ID：您要从中读取数据的 Bigtable 实例的 Google Cloud 项目的 ID。
• INSTANCE_ID：表所属的 Bigtable 实例的 ID。
• TABLE_ID：需要导出的 Bigtable 表的 ID。
• OUTPUT_DIRECTORY：写入数据的 Cloud Storage 路径，例如 gs://mybucket/somefolder
• FILENAME_PREFIX：Parquet 文件名的前缀，例如 output-
• NUM_SHARDS：要输出的 Parquet 文件的数量，例如 1

API

如需使用 REST API 来运行模板，请发送 HTTP POST 请求。如需详细了解 API 及其授权范围，请参阅 projects.templates.launch。

POST https://dataflow.googleapis.com/v1b3/projects/PROJECT_ID/locations/LOCATION/templates:launch?gcsPath=gs://dataflow-templates-LOCATION/VERSION/Cloud_Bigtable_to_GCS_Parquet
{
   "jobName": "JOB_NAME",
   "parameters": {
       "bigtableProjectId": "BIGTABLE_PROJECT_ID",
       "bigtableInstanceId": "INSTANCE_ID",
       "bigtableTableId": "TABLE_ID",
       "outputDirectory": "OUTPUT_DIRECTORY",
       "filenamePrefix": "FILENAME_PREFIX",
       "numShards": "NUM_SHARDS"
   },
   "environment": { "zone": "us-central1-f" }
}

替换以下内容：

• PROJECT_ID：您要在其中运行 Dataflow 作业的 Google Cloud 项目的 ID
• JOB_NAME：您选择的唯一性作业名称
• VERSION：您要使用的模板的版本

  您可使用以下值：

  • latest，以使用模板的最新版本，该模板在存储桶的未标示日期的父文件夹 (gs://dataflow-templates-REGION_NAME/latest/) 中可用
  • 版本名称（如 2023-09-12-00_RC00），以使用模板的特定版本，该版本嵌套在存储桶的相应日期父文件夹 (gs://dataflow-templates-REGION_NAME/) 中
  注意：最新版模板可能会随着重大更改而更新。为了防止这些重大更改影响您的生产工作流程，生产环境应使用有最近标示日期的父文件夹中保存的模板。
• LOCATION：要在其中部署 Dataflow 作业的区域，例如 us-central1
• BIGTABLE_PROJECT_ID：您要从中读取数据的 Bigtable 实例的 Google Cloud 项目的 ID。
• INSTANCE_ID：表所属的 Bigtable 实例的 ID。
• TABLE_ID：需要导出的 Bigtable 表的 ID。
• OUTPUT_DIRECTORY：写入数据的 Cloud Storage 路径，例如 gs://mybucket/somefolder
• FILENAME_PREFIX：Parquet 文件名的前缀，例如 output-
• NUM_SHARDS：要输出的 Parquet 文件的数量，例如 1

模板源代码

Java

/*
 * Copyright (C) 2019 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.bigtable;

import static com.google.cloud.teleport.bigtable.BigtableToAvro.toByteArray;

import com.google.bigtable.v2.Cell;
import com.google.bigtable.v2.Column;
import com.google.bigtable.v2.Family;
import com.google.bigtable.v2.Row;
import com.google.cloud.teleport.bigtable.BigtableToParquet.Options;
import com.google.cloud.teleport.metadata.Template;
import com.google.cloud.teleport.metadata.TemplateCategory;
import com.google.cloud.teleport.metadata.TemplateParameter;
import java.nio.ByteBuffer;
import java.util.ArrayList;
import java.util.List;
import org.apache.avro.generic.GenericRecord;
import org.apache.avro.generic.GenericRecordBuilder;
import org.apache.beam.runners.dataflow.options.DataflowPipelineOptions;
import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.PipelineResult;
import org.apache.beam.sdk.extensions.avro.coders.AvroCoder;
import org.apache.beam.sdk.io.FileIO;
import org.apache.beam.sdk.io.gcp.bigtable.BigtableIO;
import org.apache.beam.sdk.io.parquet.ParquetIO;
import org.apache.beam.sdk.options.Default;
import org.apache.beam.sdk.options.PipelineOptions;
import org.apache.beam.sdk.options.PipelineOptionsFactory;
import org.apache.beam.sdk.options.ValueProvider;
import org.apache.beam.sdk.transforms.MapElements;
import org.apache.beam.sdk.transforms.SimpleFunction;
import org.apache.beam.sdk.values.PCollection;

/**
 * Dataflow pipeline that exports data from a Cloud Bigtable table to Parquet files in GCS.
 * Currently, filtering on Cloud Bigtable table is not supported.
 *
 * <p>Check out <a
 * href="https://github.com/GoogleCloudPlatform/DataflowTemplates/blob/main/v1/README_Cloud_Bigtable_to_GCS_Parquet.md">README</a>
 * for instructions on how to use or modify this template.
 */
@Template(
    name = "Cloud_Bigtable_to_GCS_Parquet",
    category = TemplateCategory.BATCH,
    displayName = "Cloud Bigtable to Parquet Files on Cloud Storage",
    description =
        "The Bigtable to Cloud Storage Parquet template is a pipeline that reads data from a Bigtable table and writes it to a Cloud Storage bucket in Parquet format. "
            + "You can use the template to move data from Bigtable to Cloud Storage.",
    optionsClass = Options.class,
    documentation =
        "https://cloud.google.com/dataflow/docs/guides/templates/provided/bigtable-to-parquet",
    contactInformation = "https://cloud.google.com/support",
    requirements = {
      "The Bigtable table must exist.",
      "The output Cloud Storage bucket must exist before running the pipeline."
    })
public class BigtableToParquet {

  /** Options for the export pipeline. */
  public interface Options extends PipelineOptions {

    @TemplateParameter.ProjectId(
        order = 1,
        groupName = "Source",
        description = "Project ID",
        helpText =
            "The ID of the Google Cloud project that contains the Cloud Bigtable instance that you want to read data from.")
    ValueProvider<String> getBigtableProjectId();

    @SuppressWarnings("unused")
    void setBigtableProjectId(ValueProvider<String> projectId);

    @TemplateParameter.Text(
        order = 2,
        groupName = "Source",
        regexes = {"[a-z][a-z0-9\\-]+[a-z0-9]"},
        description = "Instance ID",
        helpText = "The ID of the Cloud Bigtable instance that contains the table.")
    ValueProvider<String> getBigtableInstanceId();

    @SuppressWarnings("unused")
    void setBigtableInstanceId(ValueProvider<String> instanceId);

    @TemplateParameter.Text(
        order = 3,
        groupName = "Source",
        regexes = {"[_a-zA-Z0-9][-_.a-zA-Z0-9]*"},
        description = "Table ID",
        helpText = "The ID of the Cloud Bigtable table to export.")
    ValueProvider<String> getBigtableTableId();

    @SuppressWarnings("unused")
    void setBigtableTableId(ValueProvider<String> tableId);

    @TemplateParameter.GcsWriteFolder(
        order = 4,
        groupName = "Target",
        description = "Output file directory in Cloud Storage",
        helpText =
            "The path and filename prefix for writing output files. Must end with a slash. DateTime formatting is used to parse the directory path for date and time formatters. For example: gs://your-bucket/your-path.")
    ValueProvider<String> getOutputDirectory();

    @SuppressWarnings("unused")
    void setOutputDirectory(ValueProvider<String> outputDirectory);

    @TemplateParameter.Text(
        order = 5,
        groupName = "Target",
        description = "Parquet file prefix",
        helpText =
            "The prefix of the Parquet file name. For example, \"table1-\". Defaults to: part.")
    @Default.String("part")
    ValueProvider<String> getFilenamePrefix();

    @SuppressWarnings("unused")
    void setFilenamePrefix(ValueProvider<String> filenamePrefix);

    @TemplateParameter.Integer(
        order = 6,
        groupName = "Target",
        optional = true,
        description = "Maximum output shards",
        helpText =
            "The maximum number of output shards produced when writing. A higher number of shards means higher throughput for writing to Cloud Storage, but potentially higher data aggregation cost across shards when processing output Cloud Storage files. The default value is decided by Dataflow.")
    @Default.Integer(0)
    ValueProvider<Integer> getNumShards();

    @SuppressWarnings("unused")
    void setNumShards(ValueProvider<Integer> numShards);

    @TemplateParameter.Text(
        order = 7,
        groupName = "Source",
        optional = true,
        regexes = {"[_a-zA-Z0-9][-_.a-zA-Z0-9]*"},
        description = "Application profile ID",
        helpText =
            "The ID of the Bigtable application profile to use for the export. If you don't specify an app profile, Bigtable uses the instance's default app profile: https://cloud.google.com/bigtable/docs/app-profiles#default-app-profile.")
    @Default.String("default")
    ValueProvider<String> getBigtableAppProfileId();

    @SuppressWarnings("unused")
    void setBigtableAppProfileId(ValueProvider<String> appProfileId);
  }

  /**
   * Main entry point for pipeline execution.
   *
   * @param args Command line arguments to the pipeline.
   */
  public static void main(String[] args) {
    Options options = PipelineOptionsFactory.fromArgs(args).withValidation().as(Options.class);

    PipelineResult result = run(options);

    // Wait for pipeline to finish only if it is not constructing a template.
    if (options.as(DataflowPipelineOptions.class).getTemplateLocation() == null) {
      result.waitUntilFinish();
    }
  }

  /**
   * Runs a pipeline to export data from a Cloud Bigtable table to Parquet file(s) in GCS.
   *
   * @param options arguments to the pipeline
   */
  public static PipelineResult run(Options options) {
    Pipeline pipeline = Pipeline.create(PipelineUtils.tweakPipelineOptions(options));
    BigtableIO.Read read =
        BigtableIO.read()
            .withProjectId(options.getBigtableProjectId())
            .withInstanceId(options.getBigtableInstanceId())
            .withAppProfileId(options.getBigtableAppProfileId())
            .withTableId(options.getBigtableTableId());

    // Do not validate input fields if it is running as a template.
    if (options.as(DataflowPipelineOptions.class).getTemplateLocation() != null) {
      read = read.withoutValidation();
    }

    /**
     * Steps: 1) Read records from Bigtable. 2) Convert a Bigtable Row to a GenericRecord. 3) Write
     * GenericRecord(s) to GCS in parquet format.
     */
    FileIO.Write<Void, GenericRecord> write =
        FileIO.<GenericRecord>write()
            .via(ParquetIO.sink(BigtableRow.getClassSchema()))
            .to(options.getOutputDirectory())
            .withPrefix(options.getFilenamePrefix())
            .withSuffix(".parquet");
    ValueProvider<Integer> numShardsOpt = options.getNumShards();
    if (numShardsOpt.isAccessible()) {
      Integer numShards = numShardsOpt.get();
      if (numShards != null && numShards > 0) {
        write = write.withNumShards(options.getNumShards());
      }
    }
    pipeline
        .apply("Read from Bigtable", read)
        .apply("Transform to Parquet", MapElements.via(new BigtableToParquetFn()))
        .setCoder(AvroCoder.of(GenericRecord.class, BigtableRow.getClassSchema()))
        .apply("Write to Parquet in GCS", write);

    return pipeline.run();
  }

  /**
   * Translates a {@link PCollection} of Bigtable {@link Row} to a {@link PCollection} of {@link
   * GenericRecord}.
   */
  static class BigtableToParquetFn extends SimpleFunction<Row, GenericRecord> {
    @Override
    public GenericRecord apply(Row row) {
      ByteBuffer key = ByteBuffer.wrap(toByteArray(row.getKey()));
      List<BigtableCell> cells = new ArrayList<>();
      for (Family family : row.getFamiliesList()) {
        String familyName = family.getName();
        for (Column column : family.getColumnsList()) {
          ByteBuffer qualifier = ByteBuffer.wrap(toByteArray(column.getQualifier()));
          for (Cell cell : column.getCellsList()) {
            long timestamp = cell.getTimestampMicros();
            ByteBuffer value = ByteBuffer.wrap(toByteArray(cell.getValue()));
            cells.add(new BigtableCell(familyName, qualifier, timestamp, value));
          }
        }
      }
      return new GenericRecordBuilder(BigtableRow.getClassSchema())
          .set("key", key)
          .set("cells", cells)
          .build();
    }
  }
}

后续步骤

• 了解 Dataflow 模板。
• 参阅 Google 提供的模板列表。