diff --git a/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/rules/HiveSemiJoinRule.java b/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/rules/HiveSemiJoinRule.java index 7ce924aef7..4cd68e3bad 100644 --- a/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/rules/HiveSemiJoinRule.java +++ b/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/rules/HiveSemiJoinRule.java @@ -72,6 +72,26 @@ private HiveSemiJoinRule(RelOptRuleOperand operand, RelBuilderFactory relBuilder super(operand, relBuilder, null); } + private RelNode buildProject( final Aggregate aggregate, RexBuilder rexBuilder, RelBuilder relBuilder) { + assert(!aggregate.indicator && aggregate.getAggCallList().isEmpty()); + RelNode input = aggregate.getInput(); + List groupingKeys = aggregate.getGroupSet().asList(); + List projects = new ArrayList<>(); + for(Integer keys:groupingKeys) { + projects.add(rexBuilder.makeInputRef(input, keys.intValue())); + } + return relBuilder.push(aggregate.getInput()).project(projects).build(); + } + + private boolean needProject(final RelNode input, final RelNode aggregate) { + if((input instanceof HepRelVertex + && ((HepRelVertex)input).getCurrentRel() instanceof Join ) + || input.getRowType().getFieldCount() != aggregate.getRowType().getFieldCount()) { + return true; + } + return false; + } + protected void perform(RelOptRuleCall call, ImmutableBitSet topRefs, RelNode topOperator, Join join, RelNode left, Aggregate aggregate) { LOG.debug("Matched HiveSemiJoinRule"); @@ -109,28 +129,24 @@ protected void perform(RelOptRuleCall call, ImmutableBitSet topRefs, } final ImmutableIntList newRightKeys = ImmutableIntList.copyOf(newRightKeyBuilder); - final RelNode newRight = aggregate.getInput(); + RelNode input = aggregate.getInput(); + final RelNode newRight = needProject(input, aggregate) ? buildProject(aggregate, rexBuilder, call.builder()) : input; final RexNode newCondition = RelOptUtil.createEquiJoinCondition(left, joinInfo.leftKeys, newRight, - newRightKeys, rexBuilder); - - RelNode semi = null; - //HIVE-15458: we need to add a Project on top of Join since SemiJoin with Join as it's right input - // is not expected further down the pipeline. see jira for more details - if(aggregate.getInput() instanceof HepRelVertex - && ((HepRelVertex)aggregate.getInput()).getCurrentRel() instanceof Join) { - Join rightJoin = (Join)(((HepRelVertex)aggregate.getInput()).getCurrentRel()); - List projects = new ArrayList<>(); - for(int i=0; i 1) +select count(1) from part pp where pp.p_partkey IN (select p_partkey from ss); + +with ss as +(select count(1), p_partkey, p_name from + part group by p_partkey ,p_name + having count(1) > 1) +select count(1) from part pp where pp.p_partkey IN (select p_partkey from ss); diff --git a/ql/src/test/results/clientpositive/llap/semijoin.q.out b/ql/src/test/results/clientpositive/llap/semijoin.q.out index a1e4788233..22b867aa1a 100644 --- a/ql/src/test/results/clientpositive/llap/semijoin.q.out +++ b/ql/src/test/results/clientpositive/llap/semijoin.q.out @@ -3260,11 +3260,11 @@ POSTHOOK: Input: default@part #### A masked pattern was here #### CBO PLAN: HiveProject(p_partkey=[$0]) - HiveSemiJoin(condition=[=($1, $3)], joinType=[inner]) + HiveSemiJoin(condition=[=($1, $2)], joinType=[inner]) HiveProject(p_partkey=[$0], p_name=[$1]) HiveFilter(condition=[IS NOT NULL($1)]) HiveTableScan(table=[[default, part]], table:alias=[pp]) - HiveProject(p_partkey=[$0], p_name=[$1], p_mfgr=[$2], p_brand=[$3], p_type=[$4], p_size=[$5], p_container=[$6], p_retailprice=[$7], p_comment=[$8], BLOCK__OFFSET__INSIDE__FILE=[$9], INPUT__FILE__NAME=[$10], ROW__ID=[$11]) + HiveProject(p_name=[$1]) HiveFilter(condition=[IS NOT NULL($1)]) HiveTableScan(table=[[default, part]], table:alias=[part]) @@ -3302,3 +3302,50 @@ POSTHOOK: Input: default@part 85768 86428 90681 +PREHOOK: query: explain cbo +with ss as +(select count(1), p_partkey, p_name from + part group by p_partkey ,p_name + having count(1) > 1) +select count(1) from part pp where pp.p_partkey IN (select p_partkey from ss) +PREHOOK: type: QUERY +PREHOOK: Input: default@part +#### A masked pattern was here #### +POSTHOOK: query: explain cbo +with ss as +(select count(1), p_partkey, p_name from + part group by p_partkey ,p_name + having count(1) > 1) +select count(1) from part pp where pp.p_partkey IN (select p_partkey from ss) +POSTHOOK: type: QUERY +POSTHOOK: Input: default@part +#### A masked pattern was here #### +CBO PLAN: +HiveAggregate(group=[{}], agg#0=[count()]) + HiveSemiJoin(condition=[=($0, $1)], joinType=[inner]) + HiveProject(p_partkey=[$0]) + HiveFilter(condition=[IS NOT NULL($0)]) + HiveTableScan(table=[[default, part]], table:alias=[pp]) + HiveProject(p_partkey=[$0]) + HiveFilter(condition=[>($2, 1)]) + HiveAggregate(group=[{0, 1}], agg#0=[count()]) + HiveFilter(condition=[IS NOT NULL($0)]) + HiveTableScan(table=[[default, part]], table:alias=[part]) + +PREHOOK: query: with ss as +(select count(1), p_partkey, p_name from + part group by p_partkey ,p_name + having count(1) > 1) +select count(1) from part pp where pp.p_partkey IN (select p_partkey from ss) +PREHOOK: type: QUERY +PREHOOK: Input: default@part +#### A masked pattern was here #### +POSTHOOK: query: with ss as +(select count(1), p_partkey, p_name from + part group by p_partkey ,p_name + having count(1) > 1) +select count(1) from part pp where pp.p_partkey IN (select p_partkey from ss) +POSTHOOK: type: QUERY +POSTHOOK: Input: default@part +#### A masked pattern was here #### +2