Question

我想利用时间分区表的新BigQuery功能，但我不确定这在1.6版本的Dataflow SDK中是否可行。

查看BigQuery JSON API，要创建一个分区表，需要传入

"timePartitioning": { "type": "DAY" }

选项，但com.google.cloud.dataflow.sdk.io.BigQueryIO接口仅允许指定TableReference。

我想也许我可以预先创建表，并通过BigQueryIO.Write.toTableReference lambda潜入分区装饰器..？是否有其他人通过Dataflow创建/编写分区表成功？

这似乎与设置目前无法使用的table expiration time类似。

Answer 1

正如Pavan所说，绝对可以使用Dataflow写入分区表。您是使用在流模式或批处理模式下运行的DataflowPipelineRunner吗？

您提出的解决方案应该有效。具体来说，如果您预先创建一个设置了日期分区的表，那么您可以使用BigQueryIO.Write.toTableReference lambda写入日期分区。例如：

/**
 * A Joda-time formatter that prints a date in format like {@code "20160101"}.
 * Threadsafe.
 */
private static final DateTimeFormatter FORMATTER =
    DateTimeFormat.forPattern("yyyyMMdd").withZone(DateTimeZone.UTC);

// This code generates a valid BigQuery partition name:
Instant instant = Instant.now(); // any Joda instant in a reasonable time range
String baseTableName = "project:dataset.table"; // a valid BigQuery table name
String partitionName =
    String.format("%s$%s", baseTableName, FORMATTER.print(instant));

Answer 2

我采用的方法（也在流模式下工作）：

为传入记录定义自定义窗口

将窗口转换为表/分区名称

p.apply(PubsubIO.Read
            .subscription(subscription)
            .withCoder(TableRowJsonCoder.of())
        )
        .apply(Window.into(new TablePartitionWindowFn()) )
        .apply(BigQueryIO.Write
                       .to(new DayPartitionFunc(dataset, table))
                       .withSchema(schema)
                       .withWriteDisposition(BigQueryIO.Write.WriteDisposition.WRITE_APPEND)
        );

根据传入数据设置窗口，可以忽略End Instant，因为起始值用于设置分区：

public class TablePartitionWindowFn extends NonMergingWindowFn<Object, IntervalWindow> {

private IntervalWindow assignWindow(AssignContext context) {
    TableRow source = (TableRow) context.element();
    String dttm_str = (String) source.get("DTTM");

    DateTimeFormatter formatter = DateTimeFormat.forPattern("yyyy-MM-dd").withZoneUTC();

    Instant start_point = Instant.parse(dttm_str,formatter);
    Instant end_point = start_point.withDurationAdded(1000, 1);

    return new IntervalWindow(start_point, end_point);
};

@Override
public Coder<IntervalWindow> windowCoder() {
    return IntervalWindow.getCoder();
}

@Override
public Collection<IntervalWindow> assignWindows(AssignContext c) throws Exception {
    return Arrays.asList(assignWindow(c));
}

@Override
public boolean isCompatible(WindowFn<?, ?> other) {
    return false;
}

@Override
public IntervalWindow getSideInputWindow(BoundedWindow window) {
    if (window instanceof GlobalWindow) {
        throw new IllegalArgumentException(
                "Attempted to get side input window for GlobalWindow from non-global WindowFn");
    }
    return null;
}

动态设置表分区：

public class DayPartitionFunc implements SerializableFunction<BoundedWindow, String> {

String destination = "";

public DayPartitionFunc(String dataset, String table) {
    this.destination = dataset + "." + table+ "$";
}

@Override
public String apply(BoundedWindow boundedWindow) {
    // The cast below is safe because CalendarWindows.days(1) produces IntervalWindows.
    String dayString = DateTimeFormat.forPattern("yyyyMMdd")
                                     .withZone(DateTimeZone.UTC)
                                     .print(((IntervalWindow) boundedWindow).start());
    return destination + dayString;
}}

有没有更好的方法来实现相同的结果？

Answer 3

我相信当你不使用流媒体时应该可以使用分区装饰器。我们正在积极致力于通过流媒体支持分区装饰器。如果您在非流媒体模式下看到任何错误，请告诉我们。

Answer 4

Apache Beam 2.0版支持分片BigQuery输出表out of the box。

Answer 5

如果以table_name_YYYYMMDD格式传递表名，则BigQuery会将其视为分片表，可以模拟分区表功能。请参阅文档：https://cloud.google.com/bigquery/docs/partitioned-tables

Answer 6

我已通过数据流将数据写入bigquery分区表中。如果该分区中的数据已经存在，那么这些写作是动态的，因此我可以追加或覆盖它。

我已经用Python编写了代码。这是对bigquery的批处理模式写操作。

client = bigquery.Client(project=projectName)
dataset_ref = client.dataset(datasetName)
table_ref = dataset_ref.table(bqTableName)       
job_config = bigquery.LoadJobConfig()
job_config.skip_leading_rows = skipLeadingRows
job_config.source_format = bigquery.SourceFormat.CSV
if tableExists(client, table_ref):            
    job_config.autodetect = autoDetect
    previous_rows = client.get_table(table_ref).num_rows
    #assert previous_rows > 0
    if allowJaggedRows is True:
        job_config.allowJaggedRows = True
    if allowFieldAddition is True:
        job_config._properties['load']['schemaUpdateOptions'] = ['ALLOW_FIELD_ADDITION']
    if isPartitioned is True:
        job_config._properties['load']['timePartitioning'] = {"type": "DAY"}
    if schemaList is not None:
        job_config.schema = schemaList            
    job_config.write_disposition = bigquery.WriteDisposition.WRITE_TRUNCATE
else:            
    job_config.autodetect = autoDetect
    job_config._properties['createDisposition'] = 'CREATE_IF_NEEDED'
    job_config.schema = schemaList
    if isPartitioned is True:             
        job_config._properties['load']['timePartitioning'] = {"type": "DAY"}
    if schemaList is not None:
        table = bigquery.Table(table_ref, schema=schemaList)            
load_job = client.load_table_from_uri(gcsFileName, table_ref, job_config=job_config)        
assert load_job.job_type == 'load'
load_job.result()       
assert load_job.state == 'DONE'

它工作正常。

通过Google Cloud Dataflow创建/写入Parititoned BigQuery表

6 个答案: