/** Replay the wrapped collector, but only on a selection of buckets. */
@Override
public void prepareSelectedBuckets(long... selectedBuckets) throws IOException {
if (!finished) {
throw new IllegalStateException(
"Cannot replay yet, collection is not finished: postCollect() has not been called");
}
if (this.selectedBuckets != null) {
throw new IllegalStateException("Already been replayed");
}
final LongHash hash = new LongHash(selectedBuckets.length, BigArrays.NON_RECYCLING_INSTANCE);
for (long bucket : selectedBuckets) {
hash.add(bucket);
}
this.selectedBuckets = hash;
collector.preCollection();
boolean needsScores = collector.needsScores();
Weight weight = null;
if (needsScores) {
weight =
aggContext
.searchContext()
.searcher()
.createNormalizedWeight(aggContext.searchContext().query(), true);
}
for (Entry entry : entries) {
final LeafBucketCollector leafCollector = collector.getLeafCollector(entry.context);
DocIdSetIterator docIt = null;
if (needsScores && entry.docDeltas.size() > 0) {
Scorer scorer = weight.scorer(entry.context);
// We don't need to check if the scorer is null
// since we are sure that there are documents to replay (entry.docDeltas it not empty).
docIt = scorer.iterator();
leafCollector.setScorer(scorer);
}
final PackedLongValues.Iterator docDeltaIterator = entry.docDeltas.iterator();
final PackedLongValues.Iterator buckets = entry.buckets.iterator();
int doc = 0;
for (long i = 0, end = entry.docDeltas.size(); i < end; ++i) {
doc += docDeltaIterator.next();
final long bucket = buckets.next();
final long rebasedBucket = hash.find(bucket);
if (rebasedBucket != -1) {
if (needsScores) {
if (docIt.docID() < doc) {
docIt.advance(doc);
}
// aggregations should only be replayed on matching documents
assert docIt.docID() == doc;
}
leafCollector.collect(doc, rebasedBucket);
}
}
}
collector.postCollection();
}
@Override
public boolean needsScores() {
if (collector == null) {
throw new IllegalStateException();
}
return collector.needsScores();
}
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();
}
/** Set the deferred collectors. */
@Override
public void setDeferredCollector(Iterable<BucketCollector> deferredCollectors) {
this.collector = BucketCollector.wrap(deferredCollectors);
}