Fix searchafter high latency when after value is out of range for seg…

…ment (#12334)
apache · May 31, 2023 · 327997b · 327997b
1 parent e134578
commit 327997b
Show file tree

Hide file tree

Showing 2 changed files with 18 additions and 11 deletions.
diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txt
@@ -44,6 +44,9 @@ Improvements
 
 * GITHUB#12325: Parallelize AbstractKnnVectorQuery rewrite across slices rather than segments. (Luca Cavanna)
 
+* GITHUB#12333: NumericLeafComparator#competitiveIterator makes better use of a "search after" value when paginating.
+  (Chaitanya Gohel)
+
 Optimizations
 ---------------------
 

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
@@ -91,8 +91,8 @@ public abstract class NumericLeafComparator implements LeafFieldComparator {
     // if skipping functionality should be enabled on this segment
     private final boolean enableSkipping;
     private final int maxDoc;
-    private final byte[] minValueAsBytes;
-    private final byte[] maxValueAsBytes;
+    private byte[] minValueAsBytes;
+    private byte[] maxValueAsBytes;
 
     private DocIdSetIterator competitiveIterator;
     private long iteratorCost = -1;
@@ -128,16 +128,10 @@ public NumericLeafComparator(LeafReaderContext context) throws IOException {
         }
         this.enableSkipping = true; // skipping is enabled when points are available
         this.maxDoc = context.reader().maxDoc();
-        this.maxValueAsBytes =
-            reverse == false ? new byte[bytesCount] : topValueSet ? new byte[bytesCount] : null;
-        this.minValueAsBytes =
-            reverse ? new byte[bytesCount] : topValueSet ? new byte[bytesCount] : null;
         this.competitiveIterator = DocIdSetIterator.all(maxDoc);
       } else {
         this.enableSkipping = false;
         this.maxDoc = 0;
-        this.maxValueAsBytes = null;
-        this.minValueAsBytes = null;
       }
     }
 
@@ -191,7 +185,9 @@ public void setHitsThresholdReached() throws IOException {
     // update its iterator to include possibly only docs that are "stronger" than the current bottom
     // entry
     private void updateCompetitiveIterator() throws IOException {
-      if (enableSkipping == false || hitsThresholdReached == false || queueFull == false) return;
+      if (enableSkipping == false
+          || hitsThresholdReached == false
+          || (queueFull == false && topValueSet == false)) return;
       // if some documents have missing points, check that missing values prohibits optimization
       if ((pointValues.getDocCount() < maxDoc) && isMissingValueCompetitive()) {
         return; // we can't filter out documents, as documents with missing values are competitive
@@ -204,13 +200,21 @@ private void updateCompetitiveIterator() throws IOException {
         return;
       }
       if (reverse == false) {
-        encodeBottom(maxValueAsBytes);
+        if (queueFull) { // bottom is avilable only when queue is full
+          maxValueAsBytes = maxValueAsBytes == null ? new byte[bytesCount] : maxValueAsBytes;
+          encodeBottom(maxValueAsBytes);
+        }
         if (topValueSet) {
+          minValueAsBytes = minValueAsBytes == null ? new byte[bytesCount] : minValueAsBytes;
           encodeTop(minValueAsBytes);
         }
       } else {
-        encodeBottom(minValueAsBytes);
+        if (queueFull) { // bottom is avilable only when queue is full
+          minValueAsBytes = minValueAsBytes == null ? new byte[bytesCount] : minValueAsBytes;
+          encodeBottom(minValueAsBytes);
+        }
         if (topValueSet) {
+          maxValueAsBytes = maxValueAsBytes == null ? new byte[bytesCount] : maxValueAsBytes;
           encodeTop(maxValueAsBytes);
         }
       }