使用 Storage Write API 流式传输数据

本文档介绍如何使用 BigQuery Storage Write API 将数据流式传输至 BigQuery。

在流式传输场景中，数据会连续到达，并且应该能够以最低延时进行读取。使用 BigQuery Storage Write API 处理流式工作负载时，请考虑您需要确保的事项：

• 如果您的应用只需要“至少一次”语义，请使用默认流。
• 如果您需要“正好一次”语义，请在已提交类型中创建一个或多个流，并使用流偏移量来确保“正好一次”写入。

在已提交类型下，只要服务器确认了写入请求，写入到流的数据即可用于查询。默认流也会使用已提交类型，但不提供“正好一次”保证。

使用默认流实现“至少一次”语义

如果您的应用可以接受在目标表中出现重复记录的可能性，那么我们建议您对流式传输场景使用默认流。

以下代码展示了如何将数据写入默认流：

Java

如需了解如何安装和使用 BigQuery 客户端库，请参阅 BigQuery 客户端库。如需了解详情，请参阅 BigQuery Java API 参考文档。

如需向 BigQuery 进行身份验证，请设置应用默认凭据。 如需了解详情，请参阅为客户端库设置身份验证。

import com.google.api.core.ApiFuture;
import com.google.api.core.ApiFutureCallback;
import com.google.api.core.ApiFutures;
import com.google.api.gax.core.FixedExecutorProvider;
import com.google.api.gax.retrying.RetrySettings;
import com.google.cloud.bigquery.BigQuery;
import com.google.cloud.bigquery.BigQueryOptions;
import com.google.cloud.bigquery.QueryJobConfiguration;
import com.google.cloud.bigquery.TableResult;
import com.google.cloud.bigquery.storage.v1.AppendRowsRequest;
import com.google.cloud.bigquery.storage.v1.AppendRowsResponse;
import com.google.cloud.bigquery.storage.v1.BigQueryWriteClient;
import com.google.cloud.bigquery.storage.v1.BigQueryWriteSettings;
import com.google.cloud.bigquery.storage.v1.Exceptions;
import com.google.cloud.bigquery.storage.v1.Exceptions.AppendSerializationError;
import com.google.cloud.bigquery.storage.v1.Exceptions.MaximumRequestCallbackWaitTimeExceededException;
import com.google.cloud.bigquery.storage.v1.Exceptions.StorageException;
import com.google.cloud.bigquery.storage.v1.Exceptions.StreamWriterClosedException;
import com.google.cloud.bigquery.storage.v1.JsonStreamWriter;
import com.google.cloud.bigquery.storage.v1.TableName;
import com.google.common.util.concurrent.MoreExecutors;
import com.google.protobuf.ByteString;
import com.google.protobuf.Descriptors.DescriptorValidationException;
import java.io.IOException;
import java.util.Map;
import java.util.concurrent.Executors;
import java.util.concurrent.Phaser;
import java.util.concurrent.atomic.AtomicInteger;
import javax.annotation.concurrent.GuardedBy;
import org.json.JSONArray;
import org.json.JSONObject;
import org.threeten.bp.Duration;

public class WriteToDefaultStream {

  public static void runWriteToDefaultStream()
      throws DescriptorValidationException, InterruptedException, IOException {
    // TODO(developer): Replace these variables before running the sample.
    String projectId = "MY_PROJECT_ID";
    String datasetName = "MY_DATASET_NAME";
    String tableName = "MY_TABLE_NAME";
    writeToDefaultStream(projectId, datasetName, tableName);
  }

  private static ByteString buildByteString() {
    byte[] bytes = new byte[] {1, 2, 3, 4, 5};
    return ByteString.copyFrom(bytes);
  }

  // Create a JSON object that is compatible with the table schema.
  private static JSONObject buildRecord(int i, int j) {
    JSONObject record = new JSONObject();
    StringBuilder sbSuffix = new StringBuilder();
    for (int k = 0; k < j; k++) {
      sbSuffix.append(k);
    }
    record.put("test_string", String.format("record %03d-%03d %s", i, j, sbSuffix.toString()));
    ByteString byteString = buildByteString();
    record.put("test_bytes", byteString);
    return record;
  }

  public static void writeToDefaultStream(String projectId, String datasetName, String tableName)
      throws DescriptorValidationException, InterruptedException, IOException {
    TableName parentTable = TableName.of(projectId, datasetName, tableName);

    DataWriter writer = new DataWriter();
    // One time initialization for the worker.
    writer.initialize(parentTable);

    // Write two batches of fake data to the stream, each with 10 JSON records.  Data may be
    // batched up to the maximum request size:
    // https://cloud.google.com/bigquery/quotas#write-api-limits
    for (int i = 0; i < 2; i++) {
      JSONArray jsonArr = new JSONArray();
      for (int j = 0; j < 10; j++) {
        JSONObject record = buildRecord(i, j);
        jsonArr.put(record);
      }

      writer.append(new AppendContext(jsonArr));
    }

    // Final cleanup for the stream during worker teardown.
    writer.cleanup();
    verifyExpectedRowCount(parentTable, 12);
    System.out.println("Appended records successfully.");
  }

  private static void verifyExpectedRowCount(TableName parentTable, int expectedRowCount)
      throws InterruptedException {
    String queryRowCount =
        "SELECT COUNT(*) FROM `"
            + parentTable.getProject()
            + "."
            + parentTable.getDataset()
            + "."
            + parentTable.getTable()
            + "`";
    QueryJobConfiguration queryConfig = QueryJobConfiguration.newBuilder(queryRowCount).build();
    BigQuery bigquery = BigQueryOptions.getDefaultInstance().getService();
    TableResult results = bigquery.query(queryConfig);
    int countRowsActual =
        Integer.parseInt(results.getValues().iterator().next().get("f0_").getStringValue());
    if (countRowsActual != expectedRowCount) {
      throw new RuntimeException(
          "Unexpected row count. Expected: " + expectedRowCount + ". Actual: " + countRowsActual);
    }
  }

  private static class AppendContext {

    JSONArray data;

    AppendContext(JSONArray data) {
      this.data = data;
    }
  }

  private static class DataWriter {

    private static final int MAX_RECREATE_COUNT = 3;

    private BigQueryWriteClient client;

    // Track the number of in-flight requests to wait for all responses before shutting down.
    private final Phaser inflightRequestCount = new Phaser(1);
    private final Object lock = new Object();
    private JsonStreamWriter streamWriter;

    @GuardedBy("lock")
    private RuntimeException error = null;

    private AtomicInteger recreateCount = new AtomicInteger(0);

    private JsonStreamWriter createStreamWriter(String tableName)
        throws DescriptorValidationException, IOException, InterruptedException {
      // Configure in-stream automatic retry settings.
      // Error codes that are immediately retried:
      // * ABORTED, UNAVAILABLE, CANCELLED, INTERNAL, DEADLINE_EXCEEDED
      // Error codes that are retried with exponential backoff:
      // * RESOURCE_EXHAUSTED
      RetrySettings retrySettings =
          RetrySettings.newBuilder()
              .setInitialRetryDelay(Duration.ofMillis(500))
              .setRetryDelayMultiplier(1.1)
              .setMaxAttempts(5)
              .setMaxRetryDelay(Duration.ofMinutes(1))
              .build();

      // Use the JSON stream writer to send records in JSON format. Specify the table name to write
      // to the default stream.
      // For more information about JsonStreamWriter, see:
      // https://googleapis.dev/java/google-cloud-bigquerystorage/latest/com/google/cloud/bigquery/storage/v1/JsonStreamWriter.html
      return JsonStreamWriter.newBuilder(tableName, client)
          .setExecutorProvider(FixedExecutorProvider.create(Executors.newScheduledThreadPool(100)))
          .setChannelProvider(
              BigQueryWriteSettings.defaultGrpcTransportProviderBuilder()
                  .setKeepAliveTime(org.threeten.bp.Duration.ofMinutes(1))
                  .setKeepAliveTimeout(org.threeten.bp.Duration.ofMinutes(1))
                  .setKeepAliveWithoutCalls(true)
                  .setChannelsPerCpu(2)
                  .build())
          .setEnableConnectionPool(true)
          // If value is missing in json and there is a default value configured on bigquery
          // column, apply the default value to the missing value field.
          .setDefaultMissingValueInterpretation(
              AppendRowsRequest.MissingValueInterpretation.DEFAULT_VALUE)
          .setRetrySettings(retrySettings)
          .build();
    }

    public void initialize(TableName parentTable)
        throws DescriptorValidationException, IOException, InterruptedException {
      // Initialize client without settings, internally within stream writer a new client will be
      // created with full settings.
      client = BigQueryWriteClient.create();

      streamWriter = createStreamWriter(parentTable.toString());
    }

    public void append(AppendContext appendContext)
        throws DescriptorValidationException, IOException, InterruptedException {
      synchronized (this.lock) {
        if (!streamWriter.isUserClosed()
            && streamWriter.isClosed()
            && recreateCount.getAndIncrement() < MAX_RECREATE_COUNT) {
          streamWriter = createStreamWriter(streamWriter.getStreamName());
          this.error = null;
        }
        // If earlier appends have failed, we need to reset before continuing.
        if (this.error != null) {
          throw this.error;
        }
      }
      // Append asynchronously for increased throughput.
      ApiFuture<AppendRowsResponse> future = streamWriter.append(appendContext.data);
      ApiFutures.addCallback(
          future, new AppendCompleteCallback(this, appendContext), MoreExecutors.directExecutor());

      // Increase the count of in-flight requests.
      inflightRequestCount.register();
    }

    public void cleanup() {
      // Wait for all in-flight requests to complete.
      inflightRequestCount.arriveAndAwaitAdvance();

      client.close();
      // Close the connection to the server.
      streamWriter.close();

      // Verify that no error occurred in the stream.
      synchronized (this.lock) {
        if (this.error != null) {
          throw this.error;
        }
      }
    }

    static class AppendCompleteCallback implements ApiFutureCallback<AppendRowsResponse> {

      private final DataWriter parent;
      private final AppendContext appendContext;

      public AppendCompleteCallback(DataWriter parent, AppendContext appendContext) {
        this.parent = parent;
        this.appendContext = appendContext;
      }

      public void onSuccess(AppendRowsResponse response) {
        System.out.format("Append success\n");
        this.parent.recreateCount.set(0);
        done();
      }

      public void onFailure(Throwable throwable) {
        if (throwable instanceof AppendSerializationError) {
          AppendSerializationError ase = (AppendSerializationError) throwable;
          Map<Integer, String> rowIndexToErrorMessage = ase.getRowIndexToErrorMessage();
          if (rowIndexToErrorMessage.size() > 0) {
            // Omit the faulty rows
            JSONArray dataNew = new JSONArray();
            for (int i = 0; i < appendContext.data.length(); i++) {
              if (!rowIndexToErrorMessage.containsKey(i)) {
                dataNew.put(appendContext.data.get(i));
              } else {
                // process faulty rows by placing them on a dead-letter-queue, for instance
              }
            }

            // Retry the remaining valid rows, but using a separate thread to
            // avoid potentially blocking while we are in a callback.
            if (dataNew.length() > 0) {
              try {
                this.parent.append(new AppendContext(dataNew));
              } catch (DescriptorValidationException e) {
                throw new RuntimeException(e);
              } catch (IOException e) {
                throw new RuntimeException(e);
              } catch (InterruptedException e) {
                throw new RuntimeException(e);
              }
            }
            // Mark the existing attempt as done since we got a response for it
            done();
            return;
          }
        }

        boolean resendRequest = false;
        if (throwable instanceof MaximumRequestCallbackWaitTimeExceededException) {
          resendRequest = true;
        } else if (throwable instanceof StreamWriterClosedException) {
          if (!parent.streamWriter.isUserClosed()) {
            resendRequest = true;
          }
        }
        if (resendRequest) {
          // Retry this request.
          try {
            this.parent.append(new AppendContext(appendContext.data));
          } catch (DescriptorValidationException e) {
            throw new RuntimeException(e);
          } catch (IOException e) {
            throw new RuntimeException(e);
          } catch (InterruptedException e) {
            throw new RuntimeException(e);
          }
          // Mark the existing attempt as done since we got a response for it
          done();
          return;
        }

        synchronized (this.parent.lock) {
          if (this.parent.error == null) {
            StorageException storageException = Exceptions.toStorageException(throwable);
            this.parent.error =
                (storageException != null) ? storageException : new RuntimeException(throwable);
          }
        }
        done();
      }

      private void done() {
        // Reduce the count of in-flight requests.
        this.parent.inflightRequestCount.arriveAndDeregister();
      }
    }
  }
}

使用多路复用

您只能在默认流的流写入器级层启用多路复用。如需在 Java 中启用多路复用，请在构建 StreamWriter 或 JsonStreamWriter 对象时调用 setEnableConnectionPool 方法：

// One possible way for constructing StreamWriter
StreamWriter.newBuilder(streamName)
              .setWriterSchema(protoSchema)
              .setEnableConnectionPool(true)
              .build();
// One possible way for constructing JsonStreamWriter
JsonStreamWriter.newBuilder(tableName, bigqueryClient)
              .setEnableConnectionPool(true)
              .build();

如需在 Go 中启用多路复用，请参阅连接共享（多路复用）。

使用已提交类型实现“正好一次”语义

如果您需要“正好一次”写入语义，请在已提交类型下创建写入流。在已提交类型下，一旦客户端从后端收到确认，记录便可进行查询。

已提交类型通过使用记录偏移量在流中提供“正好一次”传送。通过使用记录偏移量，应用会在每次调用 AppendRows 时指定下一个附加偏移量。仅当偏移值与下一个附加偏移量匹配时，才会执行写入操作。如需了解详情，请参阅管理流偏移量以实现“正好一次”语义。

如果您未提供偏移量，则记录将附加到流的当前末尾处。在这种情况下，如果附加请求返回错误，重试附加请求可能会导致记录在流中多次出现。

如需使用已提交类型，请执行以下步骤：

1. 调用 CreateWriteStream 以在已提交类型下创建一个或多个流。
2. 对于每个流，在循环中调用 AppendRows 以写入批量记录。
3. 对每个流调用 FinalizeWriteStream 以释放该流。调用此方法后，您无法再向流中写入更多行。此步骤在已提交类型下是可选步骤，但它有助于防止超出活跃流的限制。如需了解详情，请参阅限制流的创建速率。

您无法明确删除流。流遵循系统定义的存留时间 (TTL)：

• 如果提交的流中没有流量，则流的 TTL 为 4 天。
• 默认情况下，缓冲流的 TTL 为 7 天（如果流中没有流量）。

以下代码展示了如何使用已提交类型。

Java

如需了解如何安装和使用 BigQuery 客户端库，请参阅 BigQuery 客户端库。如需了解详情，请参阅 BigQuery Java API 参考文档。

如需向 BigQuery 进行身份验证，请设置应用默认凭据。 如需了解详情，请参阅为客户端库设置身份验证。

import com.google.api.core.ApiFuture;
import com.google.api.core.ApiFutureCallback;
import com.google.api.core.ApiFutures;
import com.google.api.gax.retrying.RetrySettings;
import com.google.cloud.bigquery.storage.v1.AppendRowsResponse;
import com.google.cloud.bigquery.storage.v1.BigQueryWriteClient;
import com.google.cloud.bigquery.storage.v1.CreateWriteStreamRequest;
import com.google.cloud.bigquery.storage.v1.Exceptions;
import com.google.cloud.bigquery.storage.v1.Exceptions.StorageException;
import com.google.cloud.bigquery.storage.v1.FinalizeWriteStreamResponse;
import com.google.cloud.bigquery.storage.v1.JsonStreamWriter;
import com.google.cloud.bigquery.storage.v1.TableName;
import com.google.cloud.bigquery.storage.v1.WriteStream;
import com.google.common.util.concurrent.MoreExecutors;
import com.google.protobuf.Descriptors.DescriptorValidationException;
import java.io.IOException;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Phaser;
import javax.annotation.concurrent.GuardedBy;
import org.json.JSONArray;
import org.json.JSONObject;
import org.threeten.bp.Duration;

public class WriteCommittedStream {

  public static void runWriteCommittedStream()
      throws DescriptorValidationException, InterruptedException, IOException {
    // TODO(developer): Replace these variables before running the sample.
    String projectId = "MY_PROJECT_ID";
    String datasetName = "MY_DATASET_NAME";
    String tableName = "MY_TABLE_NAME";

    writeCommittedStream(projectId, datasetName, tableName);
  }

  public static void writeCommittedStream(String projectId, String datasetName, String tableName)
      throws DescriptorValidationException, InterruptedException, IOException {
    BigQueryWriteClient client = BigQueryWriteClient.create();
    TableName parentTable = TableName.of(projectId, datasetName, tableName);

    DataWriter writer = new DataWriter();
    // One time initialization.
    writer.initialize(parentTable, client);

    try {
      // Write two batches of fake data to the stream, each with 10 JSON records.  Data may be
      // batched up to the maximum request size:
      // https://cloud.google.com/bigquery/quotas#write-api-limits
      long offset = 0;
      for (int i = 0; i < 2; i++) {
        // Create a JSON object that is compatible with the table schema.
        JSONArray jsonArr = new JSONArray();
        for (int j = 0; j < 10; j++) {
          JSONObject record = new JSONObject();
          record.put("col1", String.format("batch-record %03d-%03d", i, j));
          jsonArr.put(record);
        }
        writer.append(jsonArr, offset);
        offset += jsonArr.length();
      }
    } catch (ExecutionException e) {
      // If the wrapped exception is a StatusRuntimeException, check the state of the operation.
      // If the state is INTERNAL, CANCELLED, or ABORTED, you can retry. For more information, see:
      // https://grpc.github.io/grpc-java/javadoc/io/grpc/StatusRuntimeException.html
      System.out.println("Failed to append records. \n" + e);
    }

    // Final cleanup for the stream.
    writer.cleanup(client);
    System.out.println("Appended records successfully.");
  }

  // A simple wrapper object showing how the stateful stream writer should be used.
  private static class DataWriter {

    private JsonStreamWriter streamWriter;
    // Track the number of in-flight requests to wait for all responses before shutting down.
    private final Phaser inflightRequestCount = new Phaser(1);

    private final Object lock = new Object();

    @GuardedBy("lock")
    private RuntimeException error = null;

    void initialize(TableName parentTable, BigQueryWriteClient client)
        throws IOException, DescriptorValidationException, InterruptedException {
      // Initialize a write stream for the specified table.
      // For more information on WriteStream.Type, see:
      // https://googleapis.dev/java/google-cloud-bigquerystorage/latest/com/google/cloud/bigquery/storage/v1/WriteStream.Type.html
      WriteStream stream = WriteStream.newBuilder().setType(WriteStream.Type.COMMITTED).build();

      CreateWriteStreamRequest createWriteStreamRequest =
          CreateWriteStreamRequest.newBuilder()
              .setParent(parentTable.toString())
              .setWriteStream(stream)
              .build();
      WriteStream writeStream = client.createWriteStream(createWriteStreamRequest);

      // Configure in-stream automatic retry settings.
      // Error codes that are immediately retried:
      // * ABORTED, UNAVAILABLE, CANCELLED, INTERNAL, DEADLINE_EXCEEDED
      // Error codes that are retried with exponential backoff:
      // * RESOURCE_EXHAUSTED
      RetrySettings retrySettings =
          RetrySettings.newBuilder()
              .setInitialRetryDelay(Duration.ofMillis(500))
              .setRetryDelayMultiplier(1.1)
              .setMaxAttempts(5)
              .setMaxRetryDelay(Duration.ofMinutes(1))
              .build();

      // Use the JSON stream writer to send records in JSON format.
      // For more information about JsonStreamWriter, see:
      // https://googleapis.dev/java/google-cloud-bigquerystorage/latest/com/google/cloud/bigquery/storage/v1/JsonStreamWriter.html
      streamWriter =
          JsonStreamWriter.newBuilder(writeStream.getName(), writeStream.getTableSchema(), client)
              .setRetrySettings(retrySettings)
              .build();
    }

    public void append(JSONArray data, long offset)
        throws DescriptorValidationException, IOException, ExecutionException {
      synchronized (this.lock) {
        // If earlier appends have failed, we need to reset before continuing.
        if (this.error != null) {
          throw this.error;
        }
      }
      // Append asynchronously for increased throughput.
      ApiFuture<AppendRowsResponse> future = streamWriter.append(data, offset);
      ApiFutures.addCallback(
          future, new DataWriter.AppendCompleteCallback(this), MoreExecutors.directExecutor());
      // Increase the count of in-flight requests.
      inflightRequestCount.register();
    }

    public void cleanup(BigQueryWriteClient client) {
      // Wait for all in-flight requests to complete.
      inflightRequestCount.arriveAndAwaitAdvance();

      // Close the connection to the server.
      streamWriter.close();

      // Verify that no error occurred in the stream.
      synchronized (this.lock) {
        if (this.error != null) {
          throw this.error;
        }
      }

      // Finalize the stream.
      FinalizeWriteStreamResponse finalizeResponse =
          client.finalizeWriteStream(streamWriter.getStreamName());
      System.out.println("Rows written: " + finalizeResponse.getRowCount());
    }

    public String getStreamName() {
      return streamWriter.getStreamName();
    }

    static class AppendCompleteCallback implements ApiFutureCallback<AppendRowsResponse> {

      private final DataWriter parent;

      public AppendCompleteCallback(DataWriter parent) {
        this.parent = parent;
      }

      public void onSuccess(AppendRowsResponse response) {
        System.out.format("Append %d success\n", response.getAppendResult().getOffset().getValue());
        done();
      }

      public void onFailure(Throwable throwable) {
        synchronized (this.parent.lock) {
          if (this.parent.error == null) {
            StorageException storageException = Exceptions.toStorageException(throwable);
            this.parent.error =
                (storageException != null) ? storageException : new RuntimeException(throwable);
          }
        }
        System.out.format("Error: %s\n", throwable.toString());
        done();
      }

      private void done() {
        // Reduce the count of in-flight requests.
        this.parent.inflightRequestCount.arriveAndDeregister();
      }
    }
  }
}