[HIVE-10833] RowResolver looks mangled with CBO

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.3.0
• Component/s: CBO, Query Planning
• Assignee: Pengcheng Xiong
• Reporter: Eugene Koifman
• Created: 27/May/15 17:23
• Updated: 29/May/15 01:51

Description

While working on HIVE-10828 I noticed that internal state of RowResolver looks odd when CBO is enabled.
Consider the script below.

set hive.enforce.bucketing=true;
set hive.exec.dynamic.partition.mode=nonstrict;
set hive.cbo.enable=false;

drop table if exists acid_partitioned;
create table acid_partitioned (a int, c string)
  partitioned by (p int)
  clustered by (a) into 1 buckets;
  
insert into acid_partitioned partition (p) (a,p) values(1,1);

(this test is part of ql/src/test/results/clientpositive/insert_into_with_schema2.q)
With CBO on,
if you put a break point in

SemanticAnalyzer.genSelectPlan(String dest, ASTNode selExprList, QB qb, Operator<?> input,
      Operator<?> inputForSelectStar, boolean outerLV)

at line

selectStar = selectStar && exprList.getChildCount() == posn + 1;

(currently 3865) and examine out_rwsch.rslvMap variable looks like

{null={values__tmp__table__1.tmp_values_col1=_col0: string, values__tmp__table__1.tmp_values_col2=_col1: string}}

with CBO disabled, the same out_rwsch.rslvMap looks like

{values__tmp__table__1={tmp_values_col1=_col0: string, tmp_values_col2=_col1: string}}

The out_rwsch.invRslvMap also differs in the same way.

It seems that the version you get with CBO off is the correct one since
insert into acid_partitioned partition (p) (a,p) values(1,1) is rewritten to
insert into acid_partitioned partition (p) (a,p) select * from valuestmptable1

CC ashutoshc

Issue Links

• is related to: HIVE-10828 Insert with schema and dynamic partitions NullPointerException