org.apache.beam.sdk.io.hadoop.format

Interface HadoopFormatIO.Write.PartitionedWriterBuilder<KeyT,ValueT>

Type Parameters:

KeyT - Key type to write

ValueT - Value type to write

Enclosing class:

HadoopFormatIO.Write<KeyT,ValueT>

public static interface HadoopFormatIO.Write.PartitionedWriterBuilder<KeyT,ValueT>

Builder for partitioning determining.

Method Summary

Modifier and Type	Method and Description
HadoopFormatIO.Write.ExternalSynchronizationBuilder<KeyT,ValueT>	withoutPartitioning() Writes to the sink without need to partition output into specified number of partitions.
HadoopFormatIO.Write.ExternalSynchronizationBuilder<KeyT,ValueT>	withPartitioning() Writes to the sink with partitioning by Task Id.

Method Detail

withPartitioning

HadoopFormatIO.Write.ExternalSynchronizationBuilder<KeyT,ValueT> withPartitioning()

Writes to the sink with partitioning by Task Id.

Following Hadoop configuration properties are required with this option:

• mapreduce.job.reduces: Number of reduce tasks. Value is equal to number of write tasks which will be generated.
• mapreduce.job.partitioner.class: Hadoop partitioner class which will be used for distributing of records among partitions.

Returns:

WriteBuilder for write transformation

withoutPartitioning

HadoopFormatIO.Write.ExternalSynchronizationBuilder<KeyT,ValueT> withoutPartitioning()

Writes to the sink without need to partition output into specified number of partitions.

This write operation doesn't do shuffle by the partition so it saves transfer time before write operation itself. As a consequence it generates random number of partitions.

Note: Works only for PCollection.IsBounded.BOUNDED PCollection with global WindowingStrategy.

Returns:

WriteBuilder for write transformation