Fix Iceberg read when a batch doesn't have matching delete positions

Co-authored-by: Ying Su <[email protected]>
facebookincubator · Jul 18, 2024 · 1f44741 · 1f44741
1 parent 49b57ca
commit 1f44741
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 1 deletion.
diff --git a/velox/connectors/hive/iceberg/PositionalDeleteFileReader.cpp b/velox/connectors/hive/iceberg/PositionalDeleteFileReader.cpp
@@ -231,7 +231,9 @@ void PositionalDeleteFileReader::updateDeleteBitmap(
   // the deleteBitmapBuffer should be the largest position among all delte files
   deleteBitmapBuffer->setSize(std::max(
       (uint64_t)deleteBitmapBuffer->size(),
-      deletePositionsOffset_ == 0
+      deletePositionsOffset_ == 0 ||
+              (deletePositionsOffset_ < deletePositionsVector->size() &&
+               deletePositions[deletePositionsOffset_] > rowNumberUpperBound)
           ? 0
           : bits::nbytes(
                 deletePositions[deletePositionsOffset_ - 1] + 1 - offset)));

diff --git a/velox/connectors/hive/iceberg/tests/IcebergReadTest.cpp b/velox/connectors/hive/iceberg/tests/IcebergReadTest.cpp
@@ -538,6 +538,8 @@ TEST_F(HiveIcebergTest, singleBaseFileMultiplePositionalDeleteFiles) {
   // Delete the first and last row in each batch (10000 rows per batch).
   assertSingleBaseFileMultipleDeleteFiles({{0}, {9999}, {10000}, {19999}});
 
+  assertSingleBaseFileMultipleDeleteFiles({{500, 21000}});
+
   assertSingleBaseFileMultipleDeleteFiles(
       {makeRandomIncreasingValues(0, 10000),
        makeRandomIncreasingValues(10000, 20000),