[HIVE-21388] Constant UDF is not pushed to JDBCStorage Handler

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 4.0.0
• Fix Version/s: 3.2.0, 4.0.0-alpha-1
• Component/s: CBO, StorageHandler
• Assignee: Jesús Camacho Rodríguez
• Reporter: Daniel Dai
• Created: 05/Mar/19 02:28
• Updated: 27/Feb/24 22:24
• Resolved: 11/Mar/19 17:59

Description

A query involve a Hive UDF which produce a constant value does not push to JDBC table. Replacing the UDF with a constant push down works. Ideally, Hive shall first do constant folding and then push the computation.

Here is the example:

explain select PRINCIPAL_NAME from sys.TBL_PRIVS where PRINCIPAL_NAME=current_user();
+----------------------------------------------------+
|                      Explain                       |
+----------------------------------------------------+
| Plan optimized by CBO.                             |
|                                                    |
| Stage-0                                            |
|   Fetch Operator                                   |
|     limit:-1                                       |
|     Select Operator [SEL_3]                        |
|       Output:["_col0"]                             |
|       Filter Operator [FIL_2]                      |
|         predicate:(_col5 = 'hrt_qa')               |
|         Select Operator [SEL_1]                    |
|           Output:["_col5"]                         |
|           TableScan [TS_0]                         |
|             Output:["principal_name"],properties:{"hive.sql.query":"SELECT `tbl_grant_id`, `create_time`, `grant_option`, `grantor`, `grantor_type`, `principal_name`, `principal_type`, `tbl_priv`, `tbl_id`, `authorizer`\nFROM `TBL_PRIVS`","hive.sql.query.fieldNames":"tbl_grant_id,create_time,grant_option,grantor,grantor_type,principal_name,principal_type,tbl_priv,tbl_id,authorizer","hive.sql.query.fieldTypes":"bigint,int,int,string,string,string,string,string,bigint,string","hive.sql.query.split":"true"} |
|                                                    |
+----------------------------------------------------+

If I replace current_user() with a constant, the predicate is pushed to table scan.

Also, setting annotation deterministic=true and make initialize() return a ConstantObjectInspector of GenericUDFCurrentUser does not make a difference.

Issue Links

• is related to: HIVE-27291 Constant reduction in CBO does not work for UNIX_TIMESTAMP