org.apache.hadoop.dynamodb.preader

Class AbstractReadManager

java.lang.Object

org.apache.hadoop.dynamodb.preader.AbstractReadManager

Direct Known Subclasses:

QueryReadManager, ScanReadManager

public abstract class AbstractReadManager
extends Object

The ReadManager is responsible for deciding the required number of ReadWorkers to achieve the target throughput rate. It will keep track of RCUs achieved and increase or decrease the worker count as necessary.

Constructor Summary

Constructors

Constructor and Description
AbstractReadManager(RateController rateController, AbstractTimeSource time, DynamoDBRecordReaderContext context)

Method Summary

Modifier and Type	Method and Description
AbstractRecordReadRequest	dequeueReadRequest()
void	enqueueReadRequestToHead(AbstractRecordReadRequest req)
void	enqueueReadRequestToTail(AbstractRecordReadRequest req)
void	report(double permittedReadUnits, double consumedReadUnits, int items, int retries)
void	shutdown() Signal workers to stop work.

Methods inherited from class java.lang.Object

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

Constructor Detail

AbstractReadManager

public AbstractReadManager(RateController rateController,
                           AbstractTimeSource time,
                           DynamoDBRecordReaderContext context)

Method Detail

enqueueReadRequestToTail

public void enqueueReadRequestToTail(AbstractRecordReadRequest req)

enqueueReadRequestToHead

public void enqueueReadRequestToHead(AbstractRecordReadRequest req)

dequeueReadRequest

public AbstractRecordReadRequest dequeueReadRequest()

report

public void report(double permittedReadUnits,
                   double consumedReadUnits,
                   int items,
                   int retries)

Parameters:

permittedReadUnits - How many RCU the worker was allocated by the rate controller

consumedReadUnits - How many RCU the worker actually consumed

items - Items returned

retries - Number of throttles

shutdown

public void shutdown()

Signal workers to stop work. Note that this does not block and wait for them to complete. If they have pending work they may still push work to the multiplexer.