[Fix](partial update) Fix core when doing partial update on tables wi…

…th row column after schema change #26632 (#26695)

be/src/olap/rowset/segment_v2/segment_writer.cpp

@@ -578,8 +578,8 @@ Status SegmentWriter::fill_missing_columns(vectorized::MutableColumns& mutable_f
  read_index[id_and_pos.pos] = read_idx++;
  }
  if (has_row_column) {
- auto st = _tablet->fetch_value_through_row_column(rowset, seg_it.first, rids,
-   cids_missing, old_value_block);
+ auto st = _tablet->fetch_value_through_row_column(
+ rowset, *_tablet_schema, seg_it.first, rids, cids_missing, old_value_block);
  if (!st.ok()) {
  LOG(WARNING) << "failed to fetch value through row column";
  return st;

be/src/olap/tablet.cpp

@@ -2686,15 +2686,15 @@ void Tablet::update_max_version_schema(const TabletSchemaSPtr& tablet_schema) {
 }
 
 // fetch value by row column
-Status Tablet::fetch_value_through_row_column(RowsetSharedPtr input_rowset, uint32_t segid,
+Status Tablet::fetch_value_through_row_column(RowsetSharedPtr input_rowset,
+ const TabletSchema& tablet_schema, uint32_t segid,
  const std::vector<uint32_t>& rowids,
  const std::vector<uint32_t>& cids,
  vectorized::Block& block) {
  // read row data
  BetaRowsetSharedPtr rowset = std::static_pointer_cast<BetaRowset>(input_rowset);
  CHECK(rowset);
 
- const TabletSchemaSPtr tablet_schema = rowset->tablet_schema();
  SegmentCacheHandle segment_cache;
  RETURN_IF_ERROR(SegmentLoader::instance()->load_segments(rowset, &segment_cache, true));
  // find segment
@@ -2713,10 +2713,10 @@ Status Tablet::fetch_value_through_row_column(RowsetSharedPtr input_rowset, uint
  LOG_EVERY_N(INFO, 500) << "fetch_value_by_rowids, cost(us):" << watch.elapsed_time() / 1000
  << ", row_batch_size:" << rowids.size();
  });
- CHECK(tablet_schema->store_row_column());
+ CHECK(tablet_schema.store_row_column());
  // create _source column
  std::unique_ptr<segment_v2::ColumnIterator> column_iterator;
- RETURN_IF_ERROR(segment->new_column_iterator(tablet_schema->column(BeConsts::ROW_STORE_COL),
+ RETURN_IF_ERROR(segment->new_column_iterator(tablet_schema.column(BeConsts::ROW_STORE_COL),
  &column_iterator));
  segment_v2::ColumnIteratorOptions opt;
  OlapReaderStatistics stats;
@@ -2735,7 +2735,7 @@ Status Tablet::fetch_value_through_row_column(RowsetSharedPtr input_rowset, uint
  std::vector<std::string> default_values;
  default_values.resize(cids.size());
  for (int i = 0; i < cids.size(); ++i) {
- const TabletColumn& column = tablet_schema->column(cids[i]);
+ const TabletColumn& column = tablet_schema.column(cids[i]);
  vectorized::DataTypePtr type =
  vectorized::DataTypeFactory::instance().create_data_type(column);
  col_uid_to_idx[column.unique_id()] = i;
@@ -3253,8 +3253,8 @@ Status Tablet::read_columns_by_plan(TabletSchemaSPtr tablet_schema,
  (*read_index)[id_and_pos.pos] = read_idx++;
  }
  if (has_row_column) {
- auto st = fetch_value_through_row_column(rowset_iter->second, seg_it.first, rids,
- cids_to_read, block);
+ auto st = fetch_value_through_row_column(rowset_iter->second, *tablet_schema,
+ seg_it.first, rids, cids_to_read, block);
  if (!st.ok()) {
  LOG(WARNING) << "failed to fetch value through row column";
  return st;

be/src/olap/tablet.h

@@ -442,7 +442,11 @@ class Tablet : public BaseTablet {
  const TabletColumn& tablet_column,
  vectorized::MutableColumnPtr& dst);
 
- Status fetch_value_through_row_column(RowsetSharedPtr input_rowset, uint32_t segid,
+ // We use the TabletSchema from the caller because the TabletSchema in the rowset'meta
+ // may be outdated due to schema change. Also note that the the cids should indicate the indexes
+ // of the columns in the TabletSchema passed in.
+ Status fetch_value_through_row_column(RowsetSharedPtr input_rowset,
+ const TabletSchema& tablet_schema, uint32_t segid,
  const std::vector<uint32_t>& rowids,
  const std::vector<uint32_t>& cids,
  vectorized::Block& block);

regression-test/data/unique_with_mow_p0/partial_update/test_partial_update_schema_change_row_store.out

@@ -0,0 +1,76 @@
+-- This file is automatically generated. You should know what you did if you want to edit this
+-- !sql1 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql2 --
+1 1 1 0 0 0 0 0 0 0 0
+
+-- !sql3 --
+1 1 1 0 0 0 0 0 0 0 10
+
+-- !sql4 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql5 --
+1 1 1 0 0 0 0 0 0
+
+-- !sql6 --
+1 2 1 0 0 0 0 1 0
+
+-- !sql7 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql8 --
+1 1 1.0 0 0 0 0 0 0 0
+
+-- !sql9 --
+1
+
+-- !sql10 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql11 --
+1 1 1 0 0 0 0 0 0 0
+
+-- !sql12 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql13 --
+1 1 1 0 0 0 0 0 0 0
+
+-- !sql14 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql15 --
+1 1 1 0 0 0 0 0 0 0 0
+
+-- !sql16 --
+1 1 1 0 0 0 0 0 0 0 10
+
+-- !sql17 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql18 --
+1 1 1 0 0 0 0 0 0
+
+-- !sql19 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql20 --
+1 1 1.0 0 0 0 0 0 0 0
+
+-- !sql21 --
+1
+
+-- !sql23 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql24 --
+1 1 1 0 0 0 0 0 0 0
+
+-- !sql25 --
+1 0 0 0 0 0 0 0 0 0
+
+-- !sql26 --
+1 1 1 0 0 0 0 0 0 0
+

regression-test/suites/unique_with_mow_p0/partial_update/test_partial_update_schema_change.groovy

@@ -17,7 +17,10 @@
 // under the License.
 
 suite("test_partial_update_schema_change", "p0") {
- // test add value column
+
+ /* ============================================== light schema change cases: ============================================== */
+
+ // test add value column
  def tableName = "test_partial_update_light_schema_change_add_column"
  sql """ DROP TABLE IF EXISTS ${tableName} """
  sql """
@@ -618,6 +621,8 @@ suite("test_partial_update_schema_change", "p0") {
 
  sql """ DROP TABLE IF EXISTS ${tableName} """
 
+ /* ============================================== schema change cases: ============================================== */
+
  // test add value column
  tableName = "test_partial_update_schema_change_add_column"
  sql """ DROP TABLE IF EXISTS ${tableName} """