[HIVE-26582] Cartesian join fails if the query has an empty table when cartesian product edge is used

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Hive, Tez
• Labels: pull-request-available
• Assignee: Soumyakanti Das
• Reporter: Sourabh Badhya
• Created: 30/Sep/22 15:14
• Updated: Yesterday 22:18

Description

The following example fails when "hive.tez.cartesian-product.enabled" is true - 
Test command - 

mvn test -Dtest=TestMiniLlapCliDriver -Dqfile=file.q -Dtest.output.overwrite=true 

Query - file.q

set hive.tez.cartesian-product.enabled=true;

create table c (a1 int) stored as orc;
create table tmp1 (a int) stored as orc;
create table tmp2 (a int) stored as orc;

insert into table c values (3);
insert into table tmp1 values (3);

with
first as (
select a1 from c where a1 = 3
),
second as (
select a from tmp1
union all
select a from tmp2
)
select a from second cross join first; 

The following stack trace is seen - 

Caused by: java.lang.IllegalArgumentException: Number of items is 0. Should be positive
        at org.apache.tez.common.Preconditions.checkArgument(Preconditions.java:38)
        at org.apache.tez.runtime.library.utils.Grouper.init(Grouper.java:41)
        at org.apache.tez.runtime.library.cartesianproduct.FairCartesianProductEdgeManager.initialize(FairCartesianProductEdgeManager.java:66)
        at org.apache.tez.runtime.library.cartesianproduct.CartesianProductEdgeManager.initialize(CartesianProductEdgeManager.java:51)
        at org.apache.tez.dag.app.dag.impl.Edge.initialize(Edge.java:213)
        ... 22 more

The following error is seen because one of the tables (tmp2 in this case) has 0 rows in it. 

The query works fine when the config hive.tez.cartesian-product.enabled is set to false.

Issue Links

• relates to: HIVE-26524 Use Calcite to remove sections of a query plan known never produces rows
• relates to: CALCITE-5314 Prune empty parts of a query by exploiting stats/metadata
• links to: GitHub Pull Request #5524