Template Pub/Sub Proto to BigQuery

Template Pub/Sub proto to BigQuery adalah pipeline streaming yang menyerap data proto dari langganan Pub/Sub ke dalam tabel BigQuery. Setiap error yang terjadi saat menulis ke tabel BigQuery di-streaming ke topik Pub/Sub yang belum diproses.

Fungsi yang ditentukan pengguna (UDF) JavaScript dapat diberikan untuk mengubah data. Error saat menjalankan UDF dapat dikirim ke topik Pub/Sub terpisah atau topik yang sama yang belum diproses seperti error BigQuery.

Sebelum menjalankan pipeline Dataflow untuk skenario ini, pertimbangkan apakah langganan Pub/Sub BigQuery dengan UDF memenuhi persyaratan Anda.

Persyaratan pipeline

• Langganan Pub/Sub input harus ada.
• File skema untuk data Proto harus ada di Cloud Storage.
• Topik Pub/Sub output harus ada.
• Set data BigQuery output harus ada.
• Jika tabel BigQuery ada, tabel tersebut harus memiliki skema yang cocok dengan data proto, terlepas dari nilai createDisposition.

Parameter template

Parameter yang Diperlukan

• protoSchemaPath (Jalur Cloud Storage ke File Skema Proto): Jalur Cloud Storage ke file set deskriptor mandiri. Contoh: gs://MyBucket/schema.pb. schema.pb dapat dibuat dengan menambahkan --descriptor_set_out=schema.pb ke perintah protoc yang mengompilasi proto. Flag --include_imports dapat digunakan untuk menjamin bahwa file bersifat mandiri.
• fullMessageName (Nama Pesan Proto Lengkap): Nama pesan lengkap (contoh: package.name.MessageName). Jika pesan disarangkan di dalam pesan lain, sertakan semua pesan dengan pemisah '.' (contoh: package.name.OuterMessage.InnerMessage). 'package.name' harus berasal dari pernyataan package, bukan pernyataan java_package.
• inputSubscription (Langganan input Pub/Sub): Langganan Pub/Sub untuk membaca input, dalam format 'projects/your-project-id/subscriptions/your-subscription-name' (Contoh: projects/your-project-id/subscriptions/your-subscription-name).
• outputTableSpec (tabel output BigQuery): Lokasi tabel BigQuery untuk menulis output. Nama harus dalam format <project>:<dataset>.<table_name>. Skema tabel harus cocok dengan objek input.
• outputTopic (Topik Pub/Sub output): Nama topik tempat data harus dipublikasikan, dalam format 'projects/your-project-id/topics/your-topic-name' (Contoh: projects/your-project-id/topics/your-topic-name).

Parameter Opsional

• preserveProtoFieldNames (Pertahankan Nama Field Proto): Flag untuk mengontrol apakah nama field proto harus dipertahankan atau dikonversi ke lowerCamelCase. Jika tabel sudah ada, ini harus didasarkan pada apa yang cocok dengan skema tabel. Jika tidak, kolom ini akan menentukan nama kolom tabel yang dibuat. Benar untuk mempertahankan snake_case proto. False akan mengonversi kolom menjadi lowerCamelCase. (Default: false).
• bigQueryTableSchemaPath (BigQuery Table Schema Path): Jalur Cloud Storage ke file JSON skema BigQuery. Jika tidak disetel, skema akan disimpulkan dari skema Proto. (Contoh: gs://MyBucket/bq_schema.json).
• udfOutputTopic (Topik output Pub/Sub untuk kegagalan UDF): Topik output opsional untuk mengirim kegagalan UDF. Jika opsi ini tidak disetel, kegagalan akan ditulis ke topik yang sama dengan kegagalan BigQuery. (Contoh: projects/your-project-id/topics/your-topic-name).
• writeDisposition (Write Disposition yang akan digunakan untuk BigQuery): WriteDisposition BigQuery. Misalnya, WRITE_APPEND, WRITE_EMPTY, atau WRITE_TRUNCATE. Nilai defaultnya: WRITE_APPEND.
• createDisposition (Create Disposition yang akan digunakan untuk BigQuery): CreateDisposition BigQuery. Misalnya, CREATE_IF_NEEDED, CREATE_NEVER. Nilai defaultnya: CREATE_IF_NEEDED.
• javascriptTextTransformGcsPath (jalur Cloud Storage ke sumber UDF JavaScript): Pola jalur Cloud Storage untuk kode JavaScript yang berisi fungsi yang ditentukan pengguna (UDF). (Contoh: gs://your-bucket/your-function.js).
• javascriptTextTransformFunctionName (UDF Javascript Function Name): Nama fungsi yang akan dipanggil dari file JavaScript Anda. Hanya gunakan huruf, angka, dan garis bawah. (Contoh: 'transform' atau 'transform_udf1').
• javascriptTextTransformReloadIntervalMinutes (Interval pemuatan ulang otomatis UDF JavaScript (menit)): Tentukan interval yang dapat digunakan pekerja untuk memeriksa perubahan UDF JavaScript guna memuat ulang file. Nilai defaultnya adalah: 0.
• useStorageWriteApi (Gunakan BigQuery Storage Write API): Jika benar (true), pipeline akan menggunakan Storage Write API saat menulis data ke BigQuery (lihat https://cloud.google.com/blog/products/data-analytics/streaming-data-into-bigquery-using-storage-write-api). Nilai defaultnya adalah false. Saat menggunakan Storage Write API dalam mode exactly-once, Anda harus menetapkan parameter berikut: "Number of streams for BigQuery Storage Write API" dan "Triggering frequency in seconds for BigQuery Storage Write API". Jika Anda mengaktifkan mode at-least-once Dataflow atau menyetel parameter useStorageWriteApiAtLeastOnce ke benar (true), Anda tidak perlu menyetel jumlah streaming atau frekuensi pemicuan.
• useStorageWriteApiAtLeastOnce (Gunakan semantik setidaknya sekali di BigQuery Storage Write API): Parameter ini hanya berlaku jika "Gunakan BigQuery Storage Write API" diaktifkan. Jika diaktifkan, semantik setidaknya satu kali akan digunakan untuk Storage Write API, jika tidak, semantik tepat satu kali akan digunakan. Nilai defaultnya: salah.
• numStorageWriteApiStreams (Jumlah streaming untuk BigQuery Storage Write API): Jumlah streaming menentukan paralelisme transformasi Penulisan BigQueryIO dan kira-kira sesuai dengan jumlah streaming Storage Write API yang akan digunakan oleh pipeline. Lihat https://cloud.google.com/blog/products/data-analytics/streaming-data-into-bigquery-using-storage-write-api untuk mengetahui nilai yang direkomendasikan. Nilai defaultnya adalah: 0.
• storageWriteApiTriggeringFrequencySec (Frekuensi pemicuan dalam detik untuk BigQuery Storage Write API): Frekuensi pemicuan akan menentukan seberapa cepat data akan terlihat untuk dikueri di BigQuery. Lihat https://cloud.google.com/blog/products/data-analytics/streaming-data-into-bigquery-using-storage-write-api untuk mengetahui nilai yang direkomendasikan.

Fungsi yang ditentukan pengguna (UDF)

Secara opsional, Anda dapat memperluas template ini dengan menulis fungsi yang ditentukan pengguna (UDF). Template ini memanggil UDF untuk setiap elemen input. Payload elemen diserialisasi sebagai string JSON. Untuk mengetahui informasi selengkapnya, lihat Membuat fungsi yang ditentukan pengguna (UDF) untuk template Dataflow.

Spesifikasi fungsi

UDF memiliki spesifikasi berikut:

Input: kolom data pesan Pub/Sub, diserialisasi sebagai string JSON.

Output: string JSON yang cocok dengan skema tabel tujuan BigQuery.

Menjalankan template

Konsol

1. Buka halaman Dataflow Membuat tugas dari template.
Buka Membuat tugas dari template
2. Di kolom Nama tugas, masukkan nama tugas yang unik.
3. Opsional: Untuk Endpoint regional, pilih nilai dari menu drop-down. Region default-nya adalah us-central1.

   Untuk mengetahui daftar region tempat Anda dapat menjalankan tugas Dataflow, lihat Lokasi Dataflow.

4. Dari menu drop-down Template Dataflow, pilih the Pub/Sub Proto to BigQuery template.
5. Di kolom parameter yang disediakan, masukkan nilai parameter Anda.
6. Klik Jalankan tugas.

gcloud

Catatan: Untuk menggunakan Google Cloud CLI guna menjalankan template flex, Anda harus memiliki Google Cloud CLI versi 284.0.0 atau yang lebih baru.

Di shell atau terminal Anda, jalankan template:

gcloud dataflow flex-template run JOB_NAME \
    --region=REGION_NAME \
    --template-file-gcs-location=gs://dataflow-templates-REGION_NAME/VERSION/flex/PubSub_Proto_to_BigQuery_Flex \
    --parameters \
schemaPath=SCHEMA_PATH,\
fullMessageName=PROTO_MESSAGE_NAME,\
inputSubscription=SUBSCRIPTION_NAME,\
outputTableSpec=BIGQUERY_TABLE,\
outputTopic=UNPROCESSED_TOPIC
  

Ganti kode berikut:

• JOB_NAME: nama tugas unik pilihan Anda
• REGION_NAME: region tempat Anda ingin men-deploy tugas Dataflow—misalnya, us-central1
• VERSION: versi template yang ingin Anda gunakan

  Anda dapat menggunakan nilai berikut:

  • latest untuk menggunakan template versi terbaru, yang tersedia di folder induk tanpa tanggal di bucket—gs://dataflow-templates-REGION_NAME/latest/
  • nama versi, seperti 2023-09-12-00_RC00, untuk menggunakan template versi tertentu, yang dapat ditemukan bertingkat di masing-masing folder induk yang diberi tanggal dalam bucket—gs://dataflow-templates-REGION_NAME/
  Perhatian: Template versi terbaru dapat diupdate dengan perubahan yang dapat menyebabkan gangguan. Lingkungan produksi Anda harus menggunakan template yang disimpan di folder induk berisi tanggal terbaru untuk mencegah perubahan yang dapat menyebabkan gangguan ini memengaruhi alur kerja produksi Anda.
• SCHEMA_PATH: jalur Cloud Storage ke file skema Proto (misalnya, gs://MyBucket/file.pb)
• PROTO_MESSAGE_NAME: nama pesan Proto (misalnya, package.name.MessageName)
• SUBSCRIPTION_NAME: nama langganan input Pub/Sub
• BIGQUERY_TABLE: nama tabel output BigQuery
• UNPROCESSED_TOPIC: topik Pub/Sub yang akan digunakan untuk antrean yang belum diproses

API

Untuk menjalankan template menggunakan REST API, kirim permintaan POST HTTP. Untuk mengetahui informasi selengkapnya tentang API dan cakupan otorisasinya, lihat projects.templates.launch.

POST https://dataflow.googleapis.com/v1b3/projects/PROJECT_ID/locations/LOCATION/flexTemplates:launch
{
   "launch_parameter": {
      "jobName": "JOB_NAME",
      "containerSpecGcsPath": "gs://dataflow-templates-REGION_NAME/VERSION/flex/PubSub_Proto_to_BigQuery_Flex",
      "parameters": {
          "schemaPath": "SCHEMA_PATH",
          "fullMessageName": "PROTO_MESSAGE_NAME",
          "inputSubscription": "SUBSCRIPTION_NAME",
          "outputTableSpec": "BIGQUERY_TABLE",
          "outputTopic": "UNPROCESSED_TOPIC"
      }
   }
}
  

Ganti kode berikut:

• PROJECT_ID: Google Cloud Project ID tempat Anda ingin menjalankan tugas Dataflow
• JOB_NAME: nama tugas unik pilihan Anda
• LOCATION: region tempat Anda ingin men-deploy tugas Dataflow—misalnya, us-central1
• VERSION: versi template yang ingin Anda gunakan

  Anda dapat menggunakan nilai berikut:

  • latest untuk menggunakan template versi terbaru, yang tersedia di folder induk tanpa tanggal di bucket—gs://dataflow-templates-REGION_NAME/latest/
  • nama versi, seperti 2023-09-12-00_RC00, untuk menggunakan template versi tertentu, yang dapat ditemukan bertingkat di masing-masing folder induk yang diberi tanggal dalam bucket—gs://dataflow-templates-REGION_NAME/
  Perhatian: Template versi terbaru dapat diupdate dengan perubahan yang dapat menyebabkan gangguan. Lingkungan produksi Anda harus menggunakan template yang disimpan di folder induk berisi tanggal terbaru untuk mencegah perubahan yang dapat menyebabkan gangguan ini memengaruhi alur kerja produksi Anda.
• SCHEMA_PATH: jalur Cloud Storage ke file skema Proto (misalnya, gs://MyBucket/file.pb)
• PROTO_MESSAGE_NAME: nama pesan Proto (misalnya, package.name.MessageName)
• SUBSCRIPTION_NAME: nama langganan input Pub/Sub
• BIGQUERY_TABLE: nama tabel output BigQuery
• UNPROCESSED_TOPIC: topik Pub/Sub yang akan digunakan untuk antrean yang belum diproses

Kode sumber template

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.templates;

import static com.google.cloud.teleport.v2.transforms.PythonExternalTextTransformer.FailsafeRowPythonExternalUdf.stringMappingFunction;
import static java.nio.charset.StandardCharsets.UTF_8;

import com.google.cloud.teleport.metadata.MultiTemplate;
import com.google.cloud.teleport.metadata.Template;
import com.google.cloud.teleport.metadata.TemplateCategory;
import com.google.cloud.teleport.metadata.TemplateParameter;
import com.google.cloud.teleport.v2.coders.FailsafeElementCoder;
import com.google.cloud.teleport.v2.common.UncaughtExceptionLogger;
import com.google.cloud.teleport.v2.options.BigQueryCommonOptions.WriteOptions;
import com.google.cloud.teleport.v2.options.BigQueryStorageApiStreamingOptions;
import com.google.cloud.teleport.v2.options.PubsubCommonOptions.ReadSubscriptionOptions;
import com.google.cloud.teleport.v2.options.PubsubCommonOptions.WriteTopicOptions;
import com.google.cloud.teleport.v2.templates.PubsubProtoToBigQuery.PubSubProtoToBigQueryOptions;
import com.google.cloud.teleport.v2.transforms.BigQueryConverters;
import com.google.cloud.teleport.v2.transforms.ErrorConverters;
import com.google.cloud.teleport.v2.transforms.FailsafeElementTransforms.ConvertFailsafeElementToPubsubMessage;
import com.google.cloud.teleport.v2.transforms.JavascriptTextTransformer.FailsafeJavascriptUdf;
import com.google.cloud.teleport.v2.transforms.PythonExternalTextTransformer;
import com.google.cloud.teleport.v2.transforms.PythonExternalTextTransformer.RowToStringFailsafeElementFn;
import com.google.cloud.teleport.v2.utils.BigQueryIOUtils;
import com.google.cloud.teleport.v2.utils.GCSUtils;
import com.google.cloud.teleport.v2.utils.SchemaUtils;
import com.google.cloud.teleport.v2.values.FailsafeElement;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Strings;
import com.google.protobuf.Descriptors.Descriptor;
import com.google.protobuf.DynamicMessage;
import com.google.protobuf.InvalidProtocolBufferException;
import com.google.protobuf.util.JsonFormat;
import org.apache.beam.sdk.Pipeline;
import org.apache.beam.sdk.PipelineResult;
import org.apache.beam.sdk.coders.NullableCoder;
import org.apache.beam.sdk.coders.RowCoder;
import org.apache.beam.sdk.coders.StringUtf8Coder;
import org.apache.beam.sdk.io.gcp.bigquery.BigQueryIO.Write;
import org.apache.beam.sdk.io.gcp.bigquery.WriteResult;
import org.apache.beam.sdk.io.gcp.pubsub.PubsubIO;
import org.apache.beam.sdk.io.gcp.pubsub.PubsubIO.Read;
import org.apache.beam.sdk.io.gcp.pubsub.PubsubMessage;
import org.apache.beam.sdk.options.Default;
import org.apache.beam.sdk.options.PipelineOptionsFactory;
import org.apache.beam.sdk.options.Validation.Required;
import org.apache.beam.sdk.transforms.MapElements;
import org.apache.beam.sdk.transforms.PTransform;
import org.apache.beam.sdk.transforms.ParDo;
import org.apache.beam.sdk.values.PCollection;
import org.apache.beam.sdk.values.PCollectionTuple;
import org.apache.beam.sdk.values.TupleTag;
import org.apache.beam.sdk.values.TupleTagList;
import org.apache.beam.sdk.values.TypeDescriptor;
import org.apache.beam.sdk.values.TypeDescriptors;
import org.apache.commons.lang3.ArrayUtils;

/**
 * A template for writing <a href="https://developers.google.com/protocol-buffers">Protobuf</a>
 * records from Pub/Sub to BigQuery.
 *
 * <p>Persistent failures are written to a Pub/Sub unprocessed topic.
 *
 * <p>Check out <a
 * href="https://github.com/GoogleCloudPlatform/DataflowTemplates/blob/main/v2/pubsub-binary-to-bigquery/README_PubSub_Proto_to_BigQuery.md">README</a>
 * for instructions on how to use or modify this template.
 */
@MultiTemplate({
  @Template(
      name = "PubSub_Proto_to_BigQuery_Flex",
      category = TemplateCategory.STREAMING,
      displayName = "Pub/Sub Proto to BigQuery",
      description = {
        "The Pub/Sub proto to BigQuery template is a streaming pipeline that ingests proto data from a Pub/Sub subscription into a BigQuery table. "
            + "Any errors that occur while writing to the BigQuery table are streamed into a Pub/Sub unprocessed topic.\n",
        "A JavaScript user-defined function (UDF) can be provided to transform data. "
            + "Errors while executing the UDF can be sent to either a separate Pub/Sub topic or the same unprocessed topic as the BigQuery errors."
      },
      skipOptions = {
        "pythonExternalTextTransformGcsPath",
        "pythonExternalTextTransformFunctionName"
      },
      optionsClass = PubSubProtoToBigQueryOptions.class,
      flexContainerName = "pubsub-proto-to-bigquery",
      documentation =
          "https://cloud.google.com/dataflow/docs/guides/templates/provided/pubsub-proto-to-bigquery",
      contactInformation = "https://cloud.google.com/support",
      requirements = {
        "The input Pub/Sub subscription must exist.",
        "The schema file for the Proto records must exist on Cloud Storage.",
        "The output Pub/Sub topic must exist.",
        "The output BigQuery dataset must exist.",
        "If the BigQuery table exists, it must have a schema matching the proto data regardless of the <code>createDisposition</code> value."
      },
      streaming = true,
      supportsAtLeastOnce = true),
  @Template(
      name = "PubSub_Proto_to_BigQuery_Xlang",
      category = TemplateCategory.STREAMING,
      displayName = "Pub/Sub Proto to BigQuery with Python UDF",
      type = Template.TemplateType.XLANG,
      description = {
        "The Pub/Sub proto to BigQuery template is a streaming pipeline that ingests proto data from a Pub/Sub subscription into a BigQuery table. "
            + "Any errors that occur while writing to the BigQuery table are streamed into a Pub/Sub unprocessed topic.\n",
        "A Python user-defined function (UDF) can be provided to transform data. "
            + "Errors while executing the UDF can be sent to either a separate Pub/Sub topic or the same unprocessed topic as the BigQuery errors."
      },
      skipOptions = {
        "javascriptTextTransformGcsPath",
        "javascriptTextTransformFunctionName",
        "javascriptTextTransformReloadIntervalMinutes"
      },
      optionsClass = PubSubProtoToBigQueryOptions.class,
      flexContainerName = "pubsub-proto-to-bigquery-xlang",
      documentation =
          "https://cloud.google.com/dataflow/docs/guides/templates/provided/pubsub-proto-to-bigquery",
      contactInformation = "https://cloud.google.com/support",
      requirements = {
        "The input Pub/Sub subscription must exist.",
        "The schema file for the Proto records must exist on Cloud Storage.",
        "The output Pub/Sub topic must exist.",
        "The output BigQuery dataset must exist.",
        "If the BigQuery table exists, it must have a schema matching the proto data regardless of the <code>createDisposition</code> value."
      },
      streaming = true,
      supportsAtLeastOnce = true)
})
public final class PubsubProtoToBigQuery {
  private static final TupleTag<FailsafeElement<String, String>> UDF_SUCCESS_TAG = new TupleTag<>();
  private static final TupleTag<FailsafeElement<String, String>> UDF_FAILURE_TAG = new TupleTag<>();

  private static final FailsafeElementCoder<String, String> FAILSAFE_CODER =
      FailsafeElementCoder.of(StringUtf8Coder.of(), StringUtf8Coder.of());

  public static void main(String[] args) {
    UncaughtExceptionLogger.register();
    run(PipelineOptionsFactory.fromArgs(args).as(PubSubProtoToBigQueryOptions.class));
  }

  /** {@link org.apache.beam.sdk.options.PipelineOptions} for {@link PubsubProtoToBigQuery}. */
  public interface PubSubProtoToBigQueryOptions
      extends ReadSubscriptionOptions,
          WriteOptions,
          WriteTopicOptions,
          PythonExternalTextTransformer.PythonExternalTextTransformerOptions,
          BigQueryStorageApiStreamingOptions {

    @TemplateParameter.GcsReadFile(
        order = 1,
        description = "Cloud Storage Path to the Proto Schema File",
        helpText =
            "The Cloud Storage location of the self-contained proto schema file. For example,"
                + " `gs://path/to/my/file.pb`. You can generate this file with"
                + " the `--descriptor_set_out` flag of the protoc command."
                + " The `--include_imports` flag guarantees that the file is self-contained.")
    @Required
    String getProtoSchemaPath();

    void setProtoSchemaPath(String value);

    @TemplateParameter.Text(
        order = 2,
        regexes = {"^.+([a-zA-Z][a-zA-Z0-9_]+\\.?)+[a-zA-Z0-9_]$"},
        description = "Full Proto Message Name",
        helpText =
            "The full proto message name. For example, `package.name`."
                + " `MessageName`, where `package.name` is the value provided for the"
                + " `package` statement and not the `java_package` statement.")
    @Required
    String getFullMessageName();

    void setFullMessageName(String value);

    @TemplateParameter.Boolean(
        order = 3,
        optional = true,
        description = "Preserve Proto Field Names",
        helpText =
            "To preserve the original proto field name in JSON, set this property to `true`. "
                + "To use more standard JSON names, set to `false`."
                + " For example, `false` would change `field_name` to `fieldName`. Defaults to: `false`.")
    @Default.Boolean(false)
    Boolean getPreserveProtoFieldNames();

    void setPreserveProtoFieldNames(Boolean value);

    @TemplateParameter.GcsReadFile(
        order = 4,
        optional = true,
        description = "BigQuery Table Schema Path",
        helpText =
            "The Cloud Storage path to the BigQuery schema path. "
                + "If this value isn't provided, then the schema is inferred from the Proto schema.",
        example = "gs://MyBucket/bq_schema.json")
    String getBigQueryTableSchemaPath();

    void setBigQueryTableSchemaPath(String value);

    @TemplateParameter.PubsubTopic(
        order = 5,
        optional = true,
        description = "Pub/Sub output topic for UDF failures",
        helpText =
            "The Pub/Sub topic storing the UDF errors."
                + " If this value isn't provided, UDF errors are sent to the same topic as `outputTopic`.",
        example = "projects/your-project-id/topics/your-topic-name")
    String getUdfOutputTopic();

    void setUdfOutputTopic(String udfOutputTopic);

    // Hide the UseStorageWriteApiAtLeastOnce in the UI, because it will automatically be turned
    // on when pipeline is running on ALO mode and using the Storage Write API
    @TemplateParameter.Boolean(
        order = 6,
        optional = true,
        parentName = "useStorageWriteApi",
        parentTriggerValues = {"true"},
        description = "Use at at-least-once semantics in BigQuery Storage Write API",
        helpText =
            "When using the Storage Write API, specifies the write semantics."
                + " To use at-least-once semantics (https://beam.apache.org/documentation/io/built-in/google-bigquery/#at-least-once-semantics), set this parameter to true`. To use exactly-once semantics, set the parameter to `false`."
                + " This parameter applies only when `useStorageWriteApi` is `true`. The default value is `false`.",
        hiddenUi = true)
    @Default.Boolean(false)
    @Override
    Boolean getUseStorageWriteApiAtLeastOnce();

    void setUseStorageWriteApiAtLeastOnce(Boolean value);
  }

  /** Runs the pipeline and returns the results. */
  private static PipelineResult run(PubSubProtoToBigQueryOptions options) {
    BigQueryIOUtils.validateBQStorageApiOptionsStreaming(options);

    Pipeline pipeline = Pipeline.create(options);

    Descriptor descriptor = getDescriptor(options);
    PCollection<String> maybeForUdf =
        pipeline
            .apply("Read From Pubsub", readPubsubMessages(options, descriptor))
            .apply("Dynamic Message to TableRow", new ConvertDynamicProtoMessageToJson(options));

    WriteResult writeResult =
        runUdf(maybeForUdf, options)
            .apply("Write to BigQuery", writeToBigQuery(options, descriptor));
    BigQueryIOUtils.writeResultToBigQueryInsertErrors(writeResult, options)
        .apply(
            "Create Error Payload",
            ErrorConverters.BigQueryInsertErrorToPubsubMessage.<String>newBuilder()
                .setPayloadCoder(StringUtf8Coder.of())
                .setTranslateFunction(BigQueryConverters::tableRowToJson)
                .build())
        .apply("Write Failed BQ Records", PubsubIO.writeMessages().to(options.getOutputTopic()));

    return pipeline.run();
  }

  /** Gets the {@link Descriptor} for the message type in the Pub/Sub topic. */
  @VisibleForTesting
  static Descriptor getDescriptor(PubSubProtoToBigQueryOptions options) {
    String schemaPath = options.getProtoSchemaPath();
    String messageName = options.getFullMessageName();
    Descriptor descriptor = SchemaUtils.getProtoDomain(schemaPath).getDescriptor(messageName);

    if (descriptor == null) {
      throw new IllegalArgumentException(
          messageName + " is not a recognized message in " + schemaPath);
    }

    return descriptor;
  }

  /** Returns the {@link PTransform} for reading Pub/Sub messages. */
  private static Read<DynamicMessage> readPubsubMessages(
      PubSubProtoToBigQueryOptions options, Descriptor descriptor) {
    return PubsubIO.readProtoDynamicMessages(descriptor)
        .fromSubscription(options.getInputSubscription())
        .withDeadLetterTopic(options.getOutputTopic());
  }

  /**
   * Writes messages to BigQuery, creating the table if necessary and allowed in {@code options}.
   *
   * <p>The BigQuery schema will be inferred from {@code descriptor} unless a JSON schema path is
   * specified in {@code options}.
   */
  @VisibleForTesting
  static Write<String> writeToBigQuery(
      PubSubProtoToBigQueryOptions options, Descriptor descriptor) {
    Write<String> write =
        BigQueryConverters.<String>createWriteTransform(options)
            .withFormatFunction(BigQueryConverters::convertJsonToTableRow);

    String schemaPath = options.getBigQueryTableSchemaPath();
    if (Strings.isNullOrEmpty(schemaPath)) {
      return write.withSchema(
          SchemaUtils.createBigQuerySchema(descriptor, options.getPreserveProtoFieldNames()));
    } else {
      return write.withJsonSchema(GCSUtils.getGcsFileAsString(schemaPath));
    }
  }

  /** {@link PTransform} that handles converting {@link PubsubMessage} values to JSON. */
  private static class ConvertDynamicProtoMessageToJson
      extends PTransform<PCollection<DynamicMessage>, PCollection<String>> {
    private final boolean preserveProtoName;

    private ConvertDynamicProtoMessageToJson(PubSubProtoToBigQueryOptions options) {
      this.preserveProtoName = options.getPreserveProtoFieldNames();
    }

    @Override
    public PCollection<String> expand(PCollection<DynamicMessage> input) {
      return input.apply(
          "Map to JSON",
          MapElements.into(TypeDescriptors.strings())
              .via(
                  message -> {
                    try {
                      JsonFormat.Printer printer = JsonFormat.printer();
                      return preserveProtoName
                          ? printer.preservingProtoFieldNames().print(message)
                          : printer.print(message);
                    } catch (InvalidProtocolBufferException e) {
                      throw new RuntimeException(e);
                    }
                  }));
    }
  }

  /**
   * Handles running the UDF.
   *
   * <p>If {@code options} is configured so as not to run the UDF, then the UDF will not be called.
   *
   * <p>This may add a branch to the pipeline for outputting failed UDF records to an unprocessed
   * topic.
   *
   * @param jsonCollection {@link PCollection} of JSON strings for use as input to the UDF
   * @param options the options containing info on running the UDF
   * @return the {@link PCollection} of UDF output as JSON or {@code jsonCollection} if UDF not
   *     called
   */
  @VisibleForTesting
  static PCollection<String> runUdf(
      PCollection<String> jsonCollection, PubSubProtoToBigQueryOptions options) {

    boolean useJavascriptUdf = !Strings.isNullOrEmpty(options.getJavascriptTextTransformGcsPath());
    boolean usePythonUdf = !Strings.isNullOrEmpty(options.getPythonExternalTextTransformGcsPath());

    // In order to avoid generating a graph that makes it look like a UDF was called when none was
    // intended, simply return the input as "success" output.
    if (!useJavascriptUdf && !usePythonUdf) {
      return jsonCollection;
    }

    // For testing purposes, we need to do this check before creating the PTransform rather than
    // in `expand`. Otherwise, we get a NullPointerException due to the PTransform not returning
    // a value.
    if (useJavascriptUdf
        && Strings.isNullOrEmpty(options.getJavascriptTextTransformFunctionName())) {
      throw new IllegalArgumentException(
          "JavaScript function name cannot be null or empty if file is set");
    }
    if (usePythonUdf
        && Strings.isNullOrEmpty(options.getPythonExternalTextTransformFunctionName())) {
      throw new IllegalArgumentException(
          "Python function name cannot be null or empty if file is set");
    }
    if (usePythonUdf && useJavascriptUdf) {
      throw new IllegalArgumentException(
          "Either javascript or Python gcs path must be provided, but not both.");
    }

    PCollectionTuple maybeSuccess;
    if (usePythonUdf) {
      maybeSuccess = jsonCollection.apply("Run UDF", new RunPythonUdf(options));
    } else {
      maybeSuccess = jsonCollection.apply("Run UDF", new RunUdf(options));
    }

    maybeSuccess
        .get(UDF_FAILURE_TAG)
        .setCoder(FAILSAFE_CODER)
        .apply(
            "Get UDF Failures",
            ConvertFailsafeElementToPubsubMessage.<String, String>builder()
                .setOriginalPayloadSerializeFn(s -> ArrayUtils.toObject(s.getBytes(UTF_8)))
                .setErrorMessageAttributeKey("udfErrorMessage")
                .build())
        .apply("Write Failed UDF", writeUdfFailures(options));

    return maybeSuccess
        .get(UDF_SUCCESS_TAG)
        .setCoder(FAILSAFE_CODER)
        .apply(
            "Get UDF Output",
            MapElements.into(TypeDescriptors.strings()).via(FailsafeElement::getPayload))
        .setCoder(NullableCoder.of(StringUtf8Coder.of()));
  }

  /** {@link PTransform} that calls a UDF and returns both success and failure output. */
  private static class RunUdf extends PTransform<PCollection<String>, PCollectionTuple> {
    private final PubSubProtoToBigQueryOptions options;

    RunUdf(PubSubProtoToBigQueryOptions options) {
      this.options = options;
    }

    @Override
    public PCollectionTuple expand(PCollection<String> input) {
      return input
          .apply("Prepare Failsafe UDF", makeFailsafe())
          .setCoder(FAILSAFE_CODER)
          .apply(
              "Call UDF",
              FailsafeJavascriptUdf.<String>newBuilder()
                  .setFileSystemPath(options.getJavascriptTextTransformGcsPath())
                  .setFunctionName(options.getJavascriptTextTransformFunctionName())
                  .setReloadIntervalMinutes(
                      options.getJavascriptTextTransformReloadIntervalMinutes())
                  .setSuccessTag(UDF_SUCCESS_TAG)
                  .setFailureTag(UDF_FAILURE_TAG)
                  .build());
    }

    private static MapElements<String, FailsafeElement<String, String>> makeFailsafe() {
      return MapElements.into(new TypeDescriptor<FailsafeElement<String, String>>() {})
          .via((String json) -> FailsafeElement.of(json, json));
    }
  }

  /** {@link PTransform} that calls a python UDF and returns both success and failure output. */
  private static class RunPythonUdf extends PTransform<PCollection<String>, PCollectionTuple> {
    private final PubSubProtoToBigQueryOptions options;

    RunPythonUdf(PubSubProtoToBigQueryOptions options) {
      this.options = options;
    }

    @Override
    public PCollectionTuple expand(PCollection<String> input) {
      return input
          .apply("Prepare Failsafe row", stringMappingFunction())
          .setCoder(
              RowCoder.of(PythonExternalTextTransformer.FailsafeRowPythonExternalUdf.ROW_SCHEMA))
          .apply(
              "InvokeUDF",
              PythonExternalTextTransformer.FailsafePythonExternalUdf.newBuilder()
                  .setFileSystemPath(options.getPythonExternalTextTransformGcsPath())
                  .setFunctionName(options.getPythonExternalTextTransformFunctionName())
                  .build())
          .apply(
              "MapRowsToFailsafeElements",
              ParDo.of(new RowToStringFailsafeElementFn(UDF_SUCCESS_TAG, UDF_FAILURE_TAG))
                  .withOutputTags(UDF_SUCCESS_TAG, TupleTagList.of(UDF_FAILURE_TAG)));
    }
  }

  /**
   * Returns a {@link PubsubIO.Write} configured to write UDF failures to the appropriate output
   * topic.
   */
  private static PubsubIO.Write<PubsubMessage> writeUdfFailures(
      PubSubProtoToBigQueryOptions options) {
    PubsubIO.Write<PubsubMessage> write = PubsubIO.writeMessages();
    return Strings.isNullOrEmpty(options.getUdfOutputTopic())
        ? write.to(options.getOutputTopic())
        : write.to(options.getUdfOutputTopic());
  }
}

Langkah berikutnya

• Pelajari template Dataflow.
• Lihat daftar template yang disediakan Google.