LUCENE-8875: Introduce Optimized Collector For Large Number Of Hits (#754)

This commit introduces a new collector which is optimized for
cases when the number of hits is large and/or the actual hits
collected are sparse in comparison to the number of hits
requested.

lucene/CHANGES.txt

@@ -143,6 +143,9 @@ Improvements
 * LUCENE-8845: Allow Intervals.prefix() and Intervals.wildcard() to specify
   their maximum allowed expansions (Alan Woodward)
 
+* LUCENE-8875: Introduce a Collector optimized for use cases when large
+  number of hits are requested (Atri Sharma)
+
 * LUCENE-8848 LUCENE-7757 LUCENE-8492: The UnifiedHighlighter now detects that parts of the query are not understood by
   it, and thus it should not make optimizations that result in no highlights or slow highlighting.  This generally works
   best for WEIGHT_MATCHES mode.  Consequently queries produced by ComplexPhraseQueryParser and the surround QueryParser

lucene/sandbox/src/java/org/apache/lucene/search/LargeNumHitsTopDocsCollector.java

@@ -0,0 +1,157 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.lucene.search;
+
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.Comparator;
+import java.util.List;
+
+import org.apache.lucene.index.LeafReaderContext;
+
+import static org.apache.lucene.search.TopDocsCollector.EMPTY_TOPDOCS;
+
+/**
+ * Optimized collector for large number of hits.
+ * The collector maintains an ArrayList of hits until it accumulates
+ * the requested number of hits. Post that, it builds a Priority Queue
+ * and starts filtering further hits based on the minimum competitive
+ * score.
+ */
+public final class LargeNumHitsTopDocsCollector implements Collector {
+  private final int requestedHitCount;
+  private List<ScoreDoc> hits = new ArrayList<>();
+  // package private for testing
+  HitQueue pq;
+  ScoreDoc pqTop;
+  int totalHits;
+
+  public LargeNumHitsTopDocsCollector(int requestedHitCount) {
+    this.requestedHitCount = requestedHitCount;
+    this.totalHits = 0;
+  }
+
+  // We always return COMPLETE since this collector should ideally
+  // be used only with large number of hits case
+  @Override
+  public ScoreMode scoreMode() {
+    return ScoreMode.COMPLETE;
+  }
+
+  @Override
+  public LeafCollector getLeafCollector(LeafReaderContext context) {
+    final int docBase = context.docBase;
+    return new TopScoreDocCollector.ScorerLeafCollector() {
+
+      @Override
+      public void setScorer(Scorable scorer) throws IOException {
+        super.setScorer(scorer);
+      }
+
+      @Override
+      public void collect(int doc) throws IOException {
+        float score = scorer.score();
+
+        // This collector relies on the fact that scorers produce positive values:
+        assert score >= 0; // NOTE: false for NaN
+
+        if (totalHits < requestedHitCount) {
+          hits.add(new ScoreDoc(doc + docBase, score));
+          totalHits++;
+          return;
+        } else if (totalHits == requestedHitCount) {
+          // Convert the list to a priority queue
+
+          // We should get here only when priority queue
+          // has not been built
+          assert pq == null;
+          assert pqTop == null;
+          pq = new HitQueue(requestedHitCount, false);
+
+          for (ScoreDoc scoreDoc : hits) {
+            pq.add(scoreDoc);
+          }
+
+          pqTop = pq.top();
+          hits = null;
+        }
+
+        if (score > pqTop.score) {
+          pqTop.doc = doc + docBase;
+          pqTop.score = score;
+          pqTop = pq.updateTop();
+        }
+        ++totalHits;
+      }
+    };
+  }
+
+  /** Returns the top docs that were collected by this collector. */
+  public TopDocs topDocs(int howMany) {
+
+    if (howMany <= 0 || howMany > totalHits) {
+      throw new IllegalArgumentException("Incorrect number of hits requested");
+    }
+
+    ScoreDoc[] results = new ScoreDoc[howMany];
+
+    // Get the requested results from either hits list or PQ
+    populateResults(results, howMany);
+
+    return newTopDocs(results);
+  }
+
+  /**
+   * Populates the results array with the ScoreDoc instances. This can be
+   * overridden in case a different ScoreDoc type should be returned.
+   */
+  protected void populateResults(ScoreDoc[] results, int howMany) {
+    if (pq != null) {
+      assert totalHits >= requestedHitCount;
+      for (int i = howMany - 1; i >= 0; i--) {
+        results[i] = pq.pop();
+      }
+      return;
+    }
+
+    // Total number of hits collected were less than requestedHitCount
+    assert totalHits < requestedHitCount;
+    Collections.sort(hits, Comparator.comparing((ScoreDoc scoreDoc) ->
+        scoreDoc.score).reversed().thenComparing(scoreDoc -> scoreDoc.doc));
+
+    for (int i = 0; i < howMany; i++) {
+      results[i] = hits.get(i);
+    }
+  }
+
+  /**
+   * Returns a {@link TopDocs} instance containing the given results. If
+   * <code>results</code> is null it means there are no results to return,
+   * either because there were 0 calls to collect() or because the arguments to
+   * topDocs were invalid.
+   */
+  protected TopDocs newTopDocs(ScoreDoc[] results) {
+    return results == null ? EMPTY_TOPDOCS : new TopDocs(new TotalHits(totalHits, TotalHits.Relation.EQUAL_TO), results);
+  }
+
+  /** Returns the top docs that were collected by this collector. */
+  public TopDocs topDocs() {
+    return topDocs(Math.min(totalHits, requestedHitCount));
+  }
+}

lucene/sandbox/src/test/org/apache/lucene/search/TestLargeNumHitsTopDocsCollector.java

@@ -0,0 +1,158 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.lucene.search;
+
+import java.io.IOException;
+
+import org.apache.lucene.document.Document;
+import org.apache.lucene.document.Field;
+import org.apache.lucene.index.IndexReader;
+import org.apache.lucene.index.RandomIndexWriter;
+import org.apache.lucene.index.Term;
+import org.apache.lucene.store.Directory;
+import org.apache.lucene.util.LuceneTestCase;
+
+public class TestLargeNumHitsTopDocsCollector extends LuceneTestCase {
+  private Directory dir;
+  private IndexReader reader;
+  private final Query testQuery = new BooleanQuery.Builder()
+      .add(new TermQuery(new Term("field", "5")), BooleanClause.Occur.SHOULD)
+      .add(new MatchAllDocsQuery(), BooleanClause.Occur.SHOULD)
+      .build();
+
+  @Override
+  public void setUp() throws Exception {
+    super.setUp();
+    dir = newDirectory();
+    RandomIndexWriter writer = new RandomIndexWriter(random(), dir);
+    for (int i = 0; i < 1_000; i++) {
+      Document doc = new Document();
+      doc.add(newStringField("field", "5", Field.Store.NO));
+      writer.addDocument(doc);
+      writer.addDocument(new Document());
+    }
+    reader = writer.getReader();
+    writer.close();
+  }
+
+  @Override
+  public void tearDown() throws Exception {
+    reader.close();
+    dir.close();
+    dir = null;
+    super.tearDown();
+  }
+  public void testRequestMoreHitsThanCollected() throws Exception {
+    runNumHits(150);
+  }
+
+  public void testSingleNumHit() throws Exception {
+    runNumHits(1);
+  }
+
+  public void testRequestLessHitsThanCollected() throws Exception {
+    runNumHits(25);
+  }
+
+  public void testIllegalArguments() throws IOException {
+    IndexSearcher searcher = newSearcher(reader);
+    LargeNumHitsTopDocsCollector largeCollector = new LargeNumHitsTopDocsCollector(15);
+    TopScoreDocCollector regularCollector = TopScoreDocCollector.create(15, null, Integer.MAX_VALUE);
+
+    searcher.search(testQuery, largeCollector);
+    searcher.search(testQuery, regularCollector);
+
+    assertEquals(largeCollector.totalHits, regularCollector.totalHits);
+
+    IllegalArgumentException expected  = expectThrows(IllegalArgumentException.class, () -> {
+      largeCollector.topDocs(350_000);
+    });
+
+    assertTrue(expected.getMessage().contains("Incorrect number of hits requested"));
+  }
+
+  public void testNoPQBuild() throws IOException {
+    IndexSearcher searcher = newSearcher(reader);
+    LargeNumHitsTopDocsCollector largeCollector = new LargeNumHitsTopDocsCollector(250_000);
+    TopScoreDocCollector regularCollector = TopScoreDocCollector.create(250_000, null, Integer.MAX_VALUE);
+
+    searcher.search(testQuery, largeCollector);
+    searcher.search(testQuery, regularCollector);
+
+    assertEquals(largeCollector.totalHits, regularCollector.totalHits);
+
+    assertEquals(largeCollector.pq, null);
+    assertEquals(largeCollector.pqTop, null);
+  }
+
+  public void testPQBuild() throws IOException {
+    IndexSearcher searcher = newSearcher(reader);
+    LargeNumHitsTopDocsCollector largeCollector = new LargeNumHitsTopDocsCollector(50);
+    TopScoreDocCollector regularCollector = TopScoreDocCollector.create(50, null, Integer.MAX_VALUE);
+
+    searcher.search(testQuery, largeCollector);
+    searcher.search(testQuery, regularCollector);
+
+    assertEquals(largeCollector.totalHits, regularCollector.totalHits);
+
+    assertNotEquals(largeCollector.pq, null);
+    assertNotEquals(largeCollector.pqTop, null);
+  }
+
+  public void testNoPQHitsOrder() throws IOException {
+    IndexSearcher searcher = newSearcher(reader);
+    LargeNumHitsTopDocsCollector largeCollector = new LargeNumHitsTopDocsCollector(250_000);
+    TopScoreDocCollector regularCollector = TopScoreDocCollector.create(250_000, null, Integer.MAX_VALUE);
+
+    searcher.search(testQuery, largeCollector);
+    searcher.search(testQuery, regularCollector);
+
+    assertEquals(largeCollector.totalHits, regularCollector.totalHits);
+
+    assertEquals(largeCollector.pq, null);
+    assertEquals(largeCollector.pqTop, null);
+
+    TopDocs topDocs = largeCollector.topDocs();
+
+    if (topDocs.scoreDocs.length > 0) {
+      float preScore = topDocs.scoreDocs[0].score;
+      for (ScoreDoc scoreDoc : topDocs.scoreDocs) {
+        assert scoreDoc.score <= preScore;
+        preScore = scoreDoc.score;
+      }
+    }
+  }
+
+  private void runNumHits(int numHits) throws IOException {
+    IndexSearcher searcher = newSearcher(reader);
+    LargeNumHitsTopDocsCollector largeCollector = new LargeNumHitsTopDocsCollector(numHits);
+    TopScoreDocCollector regularCollector = TopScoreDocCollector.create(numHits, null, Integer.MAX_VALUE);
+
+    searcher.search(testQuery, largeCollector);
+    searcher.search(testQuery, regularCollector);
+
+    assertEquals(largeCollector.totalHits, regularCollector.totalHits);
+
+    TopDocs firstTopDocs = largeCollector.topDocs();
+    TopDocs secondTopDocs = regularCollector.topDocs();
+
+    assertEquals(firstTopDocs.scoreDocs.length, secondTopDocs.scoreDocs.length);
+
+    CheckHits.checkEqual(testQuery, firstTopDocs.scoreDocs, secondTopDocs.scoreDocs);
+  }
+}