Package com.atlassian.confluence.internal.index

Class ConcurrentBatchIndexer

java.lang.Object

com.atlassian.confluence.internal.index.ConcurrentBatchIndexer

All Implemented Interfaces:

BatchIndexer

@Internal
@ParametersAreNonnullByDefault
public class ConcurrentBatchIndexer
extends Object
implements BatchIndexer

A multi-threaded batch indexing implementation.

Manages batches of indexable content, and provides progress updates on indexing completion.

Since:

5.10

Constructor Summary

Constructors

Constructor	Description
ConcurrentBatchIndexer(BatchIndexer delegate, Integer threadCount, ConcurrentBatchIndexerExecutorServiceFactory executorServiceFactory)

Method Summary

Modifier and Type	Method	Description
void	index(List<HibernateHandle> handles, ReindexProgress progress)
protected List<Future<?>>	submitBatches(ExecutorService executor, @Nullable List<HibernateHandle> handles, ReindexProgress progress)	This method tries to work around a number of problems we have with rebuilding the index.

Methods inherited from class java.lang.Object

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

Constructor Detail

ConcurrentBatchIndexer

public ConcurrentBatchIndexer(BatchIndexer delegate,
                              Integer threadCount,
                              ConcurrentBatchIndexerExecutorServiceFactory executorServiceFactory)

Method Detail

index

public void index(List<HibernateHandle> handles,
                  ReindexProgress progress)

Specified by:

index in interface BatchIndexer

submitBatches

protected List<Future<?>> submitBatches(ExecutorService executor,
                                        @Nullable List<HibernateHandle> handles,
                                        ReindexProgress progress)

This method tries to work around a number of problems we have with rebuilding the index. At a high level it will submit batches of work to the executor service and return a list containing the Future of BatchIndexer.

More specifically we deal with the following problems 1. Attachments are expensive (memory and CPU) 2. Progress display is Hard (a status update is very fast, compared to say an attachment). By spreading the types out a bit we can get better reporting of how long it will take. 3. OOMEs are a real possibility for naive indexing. 4. Round robin helps even out the work item batches and makes progress estimates more smooth and accurate 5. Variable batch sizes help with controlling Attachment processing and OOME avoidance

Returns:

A list of Future of BatchIndexer