Binary doc values have stale value offset array if block contains all empty values (elastic#139922)

If all values are empty, the offsets array isn't decoded. This causes the offsets already present in the offsets array to be used. Instead need to either clear the offset array or read the compressed offsets (which are all 0s.) As follow-up, we should not send the empty offsets at all; but this will require codec version change.

Author: parkertimmins

docs/changelog/139922.yaml

@@ -0,0 +1,6 @@
+pr: 139922
+summary: Binary doc values have stale value offset array if block contains all empty
+  values
+area: Codec
+type: bug
+issues: []

server/src/main/java/org/elasticsearch/index/codec/tsdb/es819/ES819TSDBDocValuesProducer.java

@@ -55,6 +55,7 @@
 import org.elasticsearch.index.mapper.blockloader.docvalues.CustomBinaryDocValuesReader;
 
 import java.io.IOException;
+import java.util.Arrays;
 
 import static org.elasticsearch.index.codec.tsdb.es819.ES819TSDBDocValuesFormat.SKIP_INDEX_JUMP_LENGTH_PER_LEVEL;
 import static org.elasticsearch.index.codec.tsdb.es819.ES819TSDBDocValuesFormat.SKIP_INDEX_MAX_LEVEL;
@@ -537,6 +538,7 @@ private void decompressBlock(long blockId, int numDocsInBlock) throws IOExceptio
             int uncompressedBlockLength = compressedData.readVInt();
 
             if (uncompressedBlockLength == 0) {
+                Arrays.fill(uncompressedDocStarts, 0);
                 return;
             }
 

server/src/test/java/org/elasticsearch/index/codec/tsdb/es819/ES819TSDBDocValuesFormatTests.java

@@ -173,6 +173,20 @@ public void testBlockWiseBinarySmallValues() throws Exception {
         assertBinaryValues(binaryValues);
     }
 
+    public void testBlockWiseBinaryWithEmptySequences() throws Exception {
+        // Test long sequences that either have values or are all empty
+        List<String> binaryValues = new ArrayList<>();
+        int numSequences = 10;
+        for (int i = 0; i < numSequences; i++) {
+            int numInSequence = randomIntBetween(0, 3 * BLOCK_COUNT_THRESHOLD);
+            boolean emptySequence = randomBoolean();
+            for (int j = 0; j < numInSequence; j++) {
+                binaryValues.add(emptySequence ? "" : randomAlphaOfLengthBetween(0, 5));
+            }
+        }
+        assertBinaryValues(binaryValues);
+    }
+
     public void testBlockWiseBinaryLargeValues() throws Exception {
         boolean sparse = randomBoolean();
         int numBlocksBound = 5;