[
https://issues.apache.org/jira/browse/HIVE-28705?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17921675#comment-17921675
]
Shohei Okumiya edited comment on HIVE-28705 at 1/28/25 10:43 AM:
-----------------------------------------------------------------
I am just testing the repro query on my local machine.
{code:java}
explain select id from t1 where (case when month='2023-07' then false else true
END); {code}
With CBO, the predicate is partially evaluated as `(month = '2023-07') is not
true`.
{code:java}
+----------------------------------------------------+
| Explain |
+----------------------------------------------------+
| Plan optimized by CBO. |
| |
| Stage-0 |
| Fetch Operator |
| limit:-1 |
| Select Operator [SEL_2] |
| Output:["_col0"] |
| Filter Operator [FIL_4] |
| predicate:(month = '2023-07') is not true |
| TableScan [TS_0] |
| Output:["id"] |
| |
+----------------------------------------------------+ {code}
Output of EXPLAIN CBO.
{code:java}
+----------------------------------------------------+
| Explain |
+----------------------------------------------------+
| CBO PLAN: |
| HiveProject(id=[$0]) |
| HiveFilter(condition=[IS NOT TRUE(=($1, _UTF-16LE'2023-07'))]) |
| HiveTableScan(table=[[default, t1]], table:alias=[t1]) |
| |
+----------------------------------------------------+ {code}
Without CBO, no predicate is pushed. I remember there is an optimizer to remove
redundant predicates(PartitionPruner? I forgot).
{code:java}
+------------------------------+
| Explain |
+------------------------------+
| Stage-0 |
| Fetch Operator |
| limit:-1 |
| Select Operator [SEL_2] |
| Output:["_col0"] |
| TableScan [TS_0] |
| Output:["id"] |
| |
+------------------------------+ {code}
was (Author: okumin):
I am just testing the repro query on my local machine.
{code:java}
explain select id from t1 where (case when month='2023-07' then false else true
END); {code}
With CBO, the predicate is partially evaluated as `(month = '2023-07') is not
true`.
{code:java}
+----------------------------------------------------+
| Explain |
+----------------------------------------------------+
| Plan optimized by CBO. |
| |
| Stage-0 |
| Fetch Operator |
| limit:-1 |
| Select Operator [SEL_2] |
| Output:["_col0"] |
| Filter Operator [FIL_4] |
| predicate:(month = '2023-07') is not true |
| TableScan [TS_0] |
| Output:["id"] |
| |
+----------------------------------------------------+ {code}
Without CBO, no predicate is pushed. I remember there is an optimizer to remove
redundant predicates(PartitionPruner? I forgot).
{code:java}
+------------------------------+
| Explain |
+------------------------------+
| Stage-0 |
| Fetch Operator |
| limit:-1 |
| Select Operator [SEL_2] |
| Output:["_col0"] |
| TableScan [TS_0] |
| Output:["id"] |
| |
+------------------------------+ {code}
> Data Inconsistency due to missing 'IS NOT? TRUE/FALSE' parsing in Partition
> Filter Pruning
> ------------------------------------------------------------------------------------------
>
> Key: HIVE-28705
> URL: https://issues.apache.org/jira/browse/HIVE-28705
> Project: Hive
> Issue Type: Bug
> Reporter: chiranjeevi
> Assignee: Indhumathi Muthumurugesh
> Priority: Major
> Labels: pull-request-available
>
> create table t1(id string) partitioned by (month string);
> insert into t1 select '1','2020-12';
> set hive.cbo.enable=false;
> select id from t1 where (case when month='2023-07' then false else true END);
> -- output: 1
> set hive.cbo.enable=true;
> select id from t1 where (case when month='2023-07' then false else true END);
> -- ouput: \{no result}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
