/** * If there are configured {@link IndexSearcherWrapper} instances, the {@link IndexSearcher} of * the provided engine searcher gets wrapped and a new {@link Engine.Searcher} instances is * returned, otherwise the provided {@link Engine.Searcher} is returned. * * <p>This is invoked each time a {@link Engine.Searcher} is requested to do an operation. (for * example search) */ public final Engine.Searcher wrap(EngineConfig engineConfig, Engine.Searcher engineSearcher) throws IOException { final ElasticsearchDirectoryReader elasticsearchDirectoryReader = ElasticsearchDirectoryReader.getElasticsearchDirectoryReader( engineSearcher.getDirectoryReader()); if (elasticsearchDirectoryReader == null) { throw new IllegalStateException("Can't wrap non elasticsearch directory reader"); } NonClosingReaderWrapper nonClosingReaderWrapper = new NonClosingReaderWrapper(engineSearcher.getDirectoryReader()); DirectoryReader reader = wrap(nonClosingReaderWrapper); if (reader != nonClosingReaderWrapper) { if (reader.getCoreCacheKey() != elasticsearchDirectoryReader.getCoreCacheKey()) { throw new IllegalStateException( "wrapped directory reader doesn't delegate IndexReader#getCoreCacheKey, wrappers must override this method and delegate" + " to the original readers core cache key. Wrapped readers can't be used as cache keys since their are used only per request which would lead to subtle bugs"); } if (ElasticsearchDirectoryReader.getElasticsearchDirectoryReader(reader) != elasticsearchDirectoryReader) { // prevent that somebody wraps with a non-filter reader throw new IllegalStateException( "wrapped directory reader hides actual ElasticsearchDirectoryReader but shouldn't"); } } final IndexSearcher innerIndexSearcher = new IndexSearcher(reader); innerIndexSearcher.setQueryCache(engineConfig.getQueryCache()); innerIndexSearcher.setQueryCachingPolicy(engineConfig.getQueryCachingPolicy()); innerIndexSearcher.setSimilarity(engineConfig.getSimilarity()); // TODO: Right now IndexSearcher isn't wrapper friendly, when it becomes wrapper friendly we // should revise this extension point // For example if IndexSearcher#rewrite() is overwritten than also // IndexSearcher#createNormalizedWeight needs to be overwritten // This needs to be fixed before we can allow the IndexSearcher from Engine to be wrapped // multiple times final IndexSearcher indexSearcher = wrap(engineConfig, innerIndexSearcher); if (reader == nonClosingReaderWrapper && indexSearcher == innerIndexSearcher) { return engineSearcher; } else { return new Engine.Searcher(engineSearcher.source(), indexSearcher) { @Override public void close() throws ElasticsearchException { try { reader().close(); // we close the reader to make sure wrappers can release resources if needed.... // our NonClosingReaderWrapper makes sure that our reader is not closed } catch (IOException e) { throw new ElasticsearchException("failed to close reader", e); } finally { engineSearcher.close(); } } }; } }
private SearcherManager createSearcherManager() throws EngineException { boolean success = false; SearcherManager searcherManager = null; try { try { final DirectoryReader directoryReader = ElasticsearchDirectoryReader.wrap(DirectoryReader.open(indexWriter), shardId); searcherManager = new SearcherManager(directoryReader, searcherFactory); lastCommittedSegmentInfos = readLastCommittedSegmentInfos(searcherManager, store); success = true; return searcherManager; } catch (IOException e) { maybeFailEngine("start", e); try { indexWriter.rollback(); } catch (IOException e1) { // iw is closed below e.addSuppressed(e1); } throw new EngineCreationFailureException(shardId, "failed to open reader on writer", e); } } finally { if (success == false) { // release everything we created on a failure IOUtils.closeWhileHandlingException(searcherManager, indexWriter); } } }
public void testResetRootDocId() throws Exception { Directory directory = newDirectory(); IndexWriterConfig iwc = new IndexWriterConfig(null); iwc.setMergePolicy(NoMergePolicy.INSTANCE); RandomIndexWriter indexWriter = new RandomIndexWriter(random(), directory, iwc); List<Document> documents = new ArrayList<>(); // 1 segment with, 1 root document, with 3 nested sub docs Document document = new Document(); document.add( new Field(UidFieldMapper.NAME, "type#1", UidFieldMapper.Defaults.NESTED_FIELD_TYPE)); document.add( new Field(TypeFieldMapper.NAME, "__nested_field", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); document = new Document(); document.add( new Field(UidFieldMapper.NAME, "type#1", UidFieldMapper.Defaults.NESTED_FIELD_TYPE)); document.add( new Field(TypeFieldMapper.NAME, "__nested_field", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); document = new Document(); document.add( new Field(UidFieldMapper.NAME, "type#1", UidFieldMapper.Defaults.NESTED_FIELD_TYPE)); document.add( new Field(TypeFieldMapper.NAME, "__nested_field", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); document = new Document(); document.add(new Field(UidFieldMapper.NAME, "type#1", UidFieldMapper.Defaults.FIELD_TYPE)); document.add(new Field(TypeFieldMapper.NAME, "test", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); indexWriter.addDocuments(documents); indexWriter.commit(); documents.clear(); // 1 segment with: // 1 document, with 1 nested subdoc document = new Document(); document.add( new Field(UidFieldMapper.NAME, "type#2", UidFieldMapper.Defaults.NESTED_FIELD_TYPE)); document.add( new Field(TypeFieldMapper.NAME, "__nested_field", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); document = new Document(); document.add(new Field(UidFieldMapper.NAME, "type#2", UidFieldMapper.Defaults.FIELD_TYPE)); document.add(new Field(TypeFieldMapper.NAME, "test", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); indexWriter.addDocuments(documents); documents.clear(); // and 1 document, with 1 nested subdoc document = new Document(); document.add( new Field(UidFieldMapper.NAME, "type#3", UidFieldMapper.Defaults.NESTED_FIELD_TYPE)); document.add( new Field(TypeFieldMapper.NAME, "__nested_field", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); document = new Document(); document.add(new Field(UidFieldMapper.NAME, "type#3", UidFieldMapper.Defaults.FIELD_TYPE)); document.add(new Field(TypeFieldMapper.NAME, "test", TypeFieldMapper.Defaults.FIELD_TYPE)); documents.add(document); indexWriter.addDocuments(documents); indexWriter.commit(); indexWriter.close(); IndexService indexService = createIndex("test"); DirectoryReader directoryReader = DirectoryReader.open(directory); directoryReader = ElasticsearchDirectoryReader.wrap(directoryReader, new ShardId(indexService.index(), 0)); IndexSearcher searcher = new IndexSearcher(directoryReader); indexService .mapperService() .merge( "test", new CompressedXContent( PutMappingRequest.buildFromSimplifiedDef("test", "nested_field", "type=nested") .string()), MapperService.MergeReason.MAPPING_UPDATE, false); SearchContext searchContext = createSearchContext(indexService); AggregationContext context = new AggregationContext(searchContext); AggregatorFactories.Builder builder = AggregatorFactories.builder(); NestedAggregatorBuilder factory = new NestedAggregatorBuilder("test", "nested_field"); builder.addAggregator(factory); AggregatorFactories factories = builder.build(context, null); searchContext.aggregations(new SearchContextAggregations(factories)); Aggregator[] aggs = factories.createTopLevelAggregators(); BucketCollector collector = BucketCollector.wrap(Arrays.asList(aggs)); collector.preCollection(); // A regular search always exclude nested docs, so we use NonNestedDocsFilter.INSTANCE here // (otherwise MatchAllDocsQuery would be sufficient) // We exclude root doc with uid type#2, this will trigger the bug if we don't reset the root doc // when we process a new segment, because // root doc type#3 and root doc type#1 have the same segment docid BooleanQuery.Builder bq = new BooleanQuery.Builder(); bq.add(Queries.newNonNestedFilter(), Occur.MUST); bq.add(new TermQuery(new Term(UidFieldMapper.NAME, "type#2")), Occur.MUST_NOT); searcher.search(new ConstantScoreQuery(bq.build()), collector); collector.postCollection(); Nested nested = (Nested) aggs[0].buildAggregation(0); // The bug manifests if 6 docs are returned, because currentRootDoc isn't reset the previous // child docs from the first segment are emitted as hits. assertThat(nested.getDocCount(), equalTo(4L)); directoryReader.close(); directory.close(); }