Package com.linkedin.venice.hadoop.input.kafka

Class KafkaInputFormatCombiner

java.lang.Object

com.linkedin.venice.hadoop.task.datawriter.AbstractDataWriterTask

com.linkedin.venice.hadoop.input.kafka.KafkaInputFormatCombiner

All Implemented Interfaces:

java.io.Closeable, java.lang.AutoCloseable, org.apache.hadoop.io.Closeable, org.apache.hadoop.mapred.JobConfigurable, org.apache.hadoop.mapred.Reducer<org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable>

public class KafkaInputFormatCombiner
extends AbstractDataWriterTask
implements org.apache.hadoop.mapred.Reducer<org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable>, org.apache.hadoop.mapred.JobConfigurable

This class is a Combiner, which is a functionality of the MR framework where we can plug a Reducer implementation to be executed within the Mapper task, on its output. This allows the Reducer to have less work to do since part of it was already done in the Mappers. In the case of the KafkaInputFormat, we cannot do all the same work that the VeniceKafkaInputReducer is doing, since that includes producing to Kafka, but the part which is relevant to shift to Mappers is the compaction. We have observed that when the input partitions are very large, then Reducers can run out of memory. By shifting the work to Mappers, we should be able to remove this bottleneck and scale better, especially if used in combination with a low value for: VenicePushJobConstants.KAFKA_INPUT_MAX_RECORDS_PER_MAPPER

Field Summary

Fields inherited from class com.linkedin.venice.hadoop.task.datawriter.AbstractDataWriterTask

TASK_ID_NOT_SET

Constructor Summary

Constructors

Constructor	Description
KafkaInputFormatCombiner()

Method Summary

Modifier and Type	Method	Description
void	close()
void	configure(org.apache.hadoop.mapred.JobConf job)
protected void	configureTask(VeniceProperties props)	Allow implementations of this class to configure task-specific stuff.
void	reduce(org.apache.hadoop.io.BytesWritable key, java.util.Iterator<org.apache.hadoop.io.BytesWritable> values, org.apache.hadoop.mapred.OutputCollector<org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable> output, org.apache.hadoop.mapred.Reporter reporter)

Methods inherited from class com.linkedin.venice.hadoop.task.datawriter.AbstractDataWriterTask

configure, getEngineTaskConfigProvider, getPartitionCount, getTaskId, isChunkingEnabled, isRmdChunkingEnabled, setChunkingEnabled

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

KafkaInputFormatCombiner

public KafkaInputFormatCombiner()

Method Detail

reduce

public void reduce(org.apache.hadoop.io.BytesWritable key,
                   java.util.Iterator<org.apache.hadoop.io.BytesWritable> values,
                   org.apache.hadoop.mapred.OutputCollector<org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable> output,
                   org.apache.hadoop.mapred.Reporter reporter)
            throws java.io.IOException

Specified by:

reduce in interface org.apache.hadoop.mapred.Reducer<org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable,org.apache.hadoop.io.BytesWritable>

Throws:

java.io.IOException

close

public void close()

Specified by:

close in interface java.lang.AutoCloseable

Specified by:

close in interface java.io.Closeable

configureTask

protected void configureTask(VeniceProperties props)

Description copied from class: AbstractDataWriterTask

Allow implementations of this class to configure task-specific stuff.

Specified by:

configureTask in class AbstractDataWriterTask

Parameters:

props - the job props that the task was configured with.

configure

public void configure(org.apache.hadoop.mapred.JobConf job)

Specified by:

configure in interface org.apache.hadoop.mapred.JobConfigurable