[lucene] branch main updated: LUCENE-10106: Sort optimization wrongly skip first docs (#300)

[dn...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

dnhatn pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/lucene.git


The following commit(s) were added to refs/heads/main by this push:
     new b7a286d  LUCENE-10106: Sort optimization wrongly skip first docs (#300)
b7a286d is described below

commit b7a286dd69f33261637182cd274bc5bb35c773f6
Author: Nhat Nguyen <nh...@elastic.co>
AuthorDate: Wed Sep 15 09:21:59 2021 -0400

    LUCENE-10106: Sort optimization wrongly skip first docs (#300)
    
    The first documents of subsequent segments are mistakenly skipped when
    sort optimization is enabled. We should initialize maxDocVisited in
    NumericComparator to -1 instead of 0.
---
 .../search/comparators/NumericComparator.java      |  2 +-
 .../apache/lucene/search/TestSortOptimization.java | 84 ++++++++++++++++++++++
 2 files changed, 85 insertions(+), 1 deletion(-)

diff --git a/lucene/core/src/java/org/apache/lucene/search/comparators/NumericComparator.java b/lucene/core/src/java/org/apache/lucene/search/comparators/NumericComparator.java
index 0455cea..051d9cc 100644
--- a/lucene/core/src/java/org/apache/lucene/search/comparators/NumericComparator.java
+++ b/lucene/core/src/java/org/apache/lucene/search/comparators/NumericComparator.java
@@ -84,7 +84,7 @@ public abstract class NumericComparator<T extends Number> extends FieldComparato
 
     private DocIdSetIterator competitiveIterator;
     private long iteratorCost;
-    private int maxDocVisited = 0;
+    private int maxDocVisited = -1;
     private int updateCounter = 0;
 
     public NumericLeafComparator(LeafReaderContext context) throws IOException {
diff --git a/lucene/core/src/test/org/apache/lucene/search/TestSortOptimization.java b/lucene/core/src/test/org/apache/lucene/search/TestSortOptimization.java
index ec6ec66..4581cf1 100644
--- a/lucene/core/src/test/org/apache/lucene/search/TestSortOptimization.java
+++ b/lucene/core/src/test/org/apache/lucene/search/TestSortOptimization.java
@@ -20,6 +20,10 @@ import static org.apache.lucene.search.SortField.FIELD_DOC;
 import static org.apache.lucene.search.SortField.FIELD_SCORE;
 
 import java.io.IOException;
+import java.util.Collections;
+import java.util.List;
+import java.util.stream.Collectors;
+import java.util.stream.LongStream;
 import org.apache.lucene.document.Document;
 import org.apache.lucene.document.Field;
 import org.apache.lucene.document.FloatDocValuesField;
@@ -633,4 +637,84 @@ public class TestSortOptimization extends LuceneTestCase {
     reader.close();
     dir.close();
   }
+
+  public void testMaxDocVisited() throws IOException {
+    Directory dir = newDirectory();
+    IndexWriter writer = new IndexWriter(dir, new IndexWriterConfig());
+    int numDocs = atLeast(10000);
+    long offset = 100 + random().nextInt(100);
+    long smallestValue = 50 + random().nextInt(50);
+    boolean flushed = false;
+    for (int i = 0; i < numDocs; ++i) {
+      Document doc = new Document();
+      doc.add(new NumericDocValuesField("my_field", i + offset));
+      doc.add(new LongPoint("my_field", i + offset));
+      writer.addDocument(doc);
+      if (i >= 5000 && flushed == false) {
+        flushed = true;
+        writer.flush();
+        // Index the smallest value to the first slot of the second segment
+        doc = new Document();
+        doc.add(new NumericDocValuesField("my_field", smallestValue));
+        doc.add(new LongPoint("my_field", smallestValue));
+        writer.addDocument(doc);
+      }
+    }
+    IndexReader reader = DirectoryReader.open(writer);
+    writer.close();
+    IndexSearcher searcher = new IndexSearcher(reader);
+    SortField sortField = new SortField("my_field", SortField.Type.LONG);
+    TopFieldDocs topDocs =
+        searcher.search(new MatchAllDocsQuery(), 1 + random().nextInt(100), new Sort(sortField));
+    FieldDoc fieldDoc = (FieldDoc) topDocs.scoreDocs[0];
+    assertEquals(smallestValue, ((Long) fieldDoc.fields[0]).intValue());
+    reader.close();
+    dir.close();
+  }
+
+  public void testRandomLong() throws IOException {
+    Directory dir = newDirectory();
+    IndexWriter writer = new IndexWriter(dir, new IndexWriterConfig());
+    List<Long> seqNos = LongStream.range(0, atLeast(10_000)).boxed().collect(Collectors.toList());
+    Collections.shuffle(seqNos, random());
+    int pendingDocs = 0;
+    for (long seqNo : seqNos) {
+      Document doc = new Document();
+      doc.add(new NumericDocValuesField("seq_no", seqNo));
+      doc.add(new LongPoint("seq_no", seqNo));
+      writer.addDocument(doc);
+      pendingDocs++;
+      if (pendingDocs > 500 && random().nextInt(100) <= 5) {
+        pendingDocs = 0;
+        writer.flush();
+      }
+    }
+    writer.flush();
+    seqNos.sort(Long::compare);
+    IndexReader reader = DirectoryReader.open(writer);
+    writer.close();
+    IndexSearcher searcher = new IndexSearcher(reader);
+    SortField sortField = new SortField("seq_no", SortField.Type.LONG);
+    int visitedHits = 0;
+    ScoreDoc after = null;
+    while (visitedHits < seqNos.size()) {
+      int batch = 1 + random().nextInt(100);
+      Query query =
+          random().nextBoolean()
+              ? new MatchAllDocsQuery()
+              : LongPoint.newRangeQuery("seq_no", 0, Long.MAX_VALUE);
+      TopDocs topDocs = searcher.searchAfter(after, query, batch, new Sort(sortField));
+      int expectedHits = Math.min(seqNos.size() - visitedHits, batch);
+      assertEquals(expectedHits, topDocs.scoreDocs.length);
+      after = topDocs.scoreDocs[expectedHits - 1];
+      for (int i = 0; i < topDocs.scoreDocs.length; i++) {
+        FieldDoc fieldDoc = (FieldDoc) topDocs.scoreDocs[i];
+        long expectedSeqNo = seqNos.get(visitedHits);
+        assertEquals(expectedSeqNo, ((Long) fieldDoc.fields[0]).intValue());
+        visitedHits++;
+      }
+    }
+    reader.close();
+    dir.close();
+  }
 }