[fix](be)confix bug of converting outer join probe block to nullable (a…

…pache#25492) _do_evaluate will add temp result column into original table block, so in order to only convert correct columns to be nullable, need call convert_block_to_null before _do_evaluate
wsjz · Nov 19, 2023 · c24b010 · c24b010
1 parent 0572117
commit c24b010
Show file tree

Hide file tree

Showing 4 changed files with 31 additions and 4 deletions.
diff --git a/be/src/pipeline/exec/hashjoin_probe_operator.cpp b/be/src/pipeline/exec/hashjoin_probe_operator.cpp
@@ -465,12 +465,13 @@ Status HashJoinProbeOperatorX::push(RuntimeState* state, vectorized::Block* inpu
  local_state._probe_columns.resize(probe_expr_ctxs_sz);
 
  std::vector<int> res_col_ids(probe_expr_ctxs_sz);
- RETURN_IF_ERROR(_do_evaluate(*input_block, local_state._probe_expr_ctxs,
- *local_state._probe_expr_call_timer, res_col_ids));
  if (_join_op == TJoinOp::RIGHT_OUTER_JOIN || _join_op == TJoinOp::FULL_OUTER_JOIN) {
  local_state._probe_column_convert_to_null =
  local_state._dependency->convert_block_to_null(*input_block);
  }
+ RETURN_IF_ERROR(_do_evaluate(*input_block, local_state._probe_expr_ctxs,
+ *local_state._probe_expr_call_timer, res_col_ids));
+
  // TODO: Now we are not sure whether a column is nullable only by ExecNode's `row_desc`
  // so we have to initialize this flag by the first probe block.
  if (!local_state._has_set_need_null_map_for_probe) {

diff --git a/be/src/vec/exec/join/vhash_join_node.cpp b/be/src/vec/exec/join/vhash_join_node.cpp
@@ -502,11 +502,11 @@ Status HashJoinNode::push(RuntimeState* /*state*/, vectorized::Block* input_bloc
  _probe_columns.resize(probe_expr_ctxs_sz);
 
  std::vector<int> res_col_ids(probe_expr_ctxs_sz);
- RETURN_IF_ERROR(
- _do_evaluate(*input_block, _probe_expr_ctxs, *_probe_expr_call_timer, res_col_ids));
  if (_join_op == TJoinOp::RIGHT_OUTER_JOIN || _join_op == TJoinOp::FULL_OUTER_JOIN) {
  _probe_column_convert_to_null = _convert_block_to_null(*input_block);
  }
+ RETURN_IF_ERROR(
+ _do_evaluate(*input_block, _probe_expr_ctxs, *_probe_expr_call_timer, res_col_ids));
  // TODO: Now we are not sure whether a column is nullable only by ExecNode's `row_desc`
  // so we have to initialize this flag by the first probe block.
  if (!_has_set_need_null_map_for_probe) {

diff --git a/regression-test/data/correctness_p0/test_outer_join_with_inline_view.out b/regression-test/data/correctness_p0/test_outer_join_with_inline_view.out
@@ -14,3 +14,6 @@
 -- !select_with_agg_in_inline_view_and_outer_join --
 1
 
+-- !select_with_outerjoin_nullable --
+2023-10-07
+
diff --git a/regression-test/suites/correctness_p0/test_outer_join_with_inline_view.groovy b/regression-test/suites/correctness_p0/test_outer_join_with_inline_view.groovy
@@ -158,4 +158,27 @@ suite("test_outer_join_with_inline_view") {
  group by
  aa.org_code;
  """
+
+ sql """set enable_nereids_planner=false;"""
+ sql """drop table if exists tableau_trans_wide_day_month_year;"""
+ sql """CREATE TABLE
+ `tableau_trans_wide_day_month_year` (
+ `business_type` varchar(200) NULL 
+ ) ENGINE = OLAP 
+ DUPLICATE KEY(`business_type`) 
+ DISTRIBUTED BY HASH(`business_type`) BUCKETS 15 
+ PROPERTIES (
+ "replication_allocation" = "tag.location.default: 1"
+ );"""
+ sql """INSERT INTO `tableau_trans_wide_day_month_year` VALUES (NULL);"""
+ qt_select_with_outerjoin_nullable """ SELECT '2023-10-07' a
+ FROM 
+ (SELECT t.business_type
+ FROM tableau_trans_wide_day_month_year t ) a full
+ JOIN 
+ (SELECT t.business_type
+ FROM tableau_trans_wide_day_month_year t
+ WHERE false ) c
+ ON nvl(a.business_type,'0')=nvl(c.business_type,'0'); """
+ sql """drop table if exists tableau_trans_wide_day_month_year;"""
 }