[
https://issues.apache.org/jira/browse/HIVE-27919?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stamatis Zampetakis resolved HIVE-27919.
----------------------------------------
Fix Version/s: 4.1.0
Resolution: Fixed
Fixed in
https://github.com/apache/hive/commit/36ce858163a19e29eafe4a8d3307191bc28fc175.
Thanks for the reviews [~akshatm] [~kkasa]!
> Constant reduction in CBO does not work for FROM_UNIXTIME, DATE_ADD,
> DATE_SUB, TO_UNIX_TIMESTAMP
> ------------------------------------------------------------------------------------------------
>
> Key: HIVE-27919
> URL: https://issues.apache.org/jira/browse/HIVE-27919
> Project: Hive
> Issue Type: Improvement
> Components: CBO
> Affects Versions: 4.0.0-beta-1
> Reporter: Stamatis Zampetakis
> Assignee: Stamatis Zampetakis
> Priority: Major
> Labels: pull-request-available
> Fix For: 4.1.0
>
>
> Constant reduction (aka. constant folding) does not work in CBO for
> FROM_UNIXTIME, DATE_ADD, and DATE_SUB functions.
> Below, we outline the result of EXPLAIN CBO and plain EXPLAIN using some
> trivial constant inputs. Notice, that constant reduction is not present in
> the CBO plan but it is applied later on by the physical optimizer.
> {code:sql}
> SELECT FROM_UNIXTIME(1701088643)
> {code}
> {noformat}
> CBO PLAN:
> HiveProject(_o__c0=[FROM_UNIXTIME(1701088643)])
> HiveTableScan(table=[[_dummy_database, _dummy_table]],
> table:alias=[_dummy_table])
> STAGE PLANS:
> Stage: Stage-0
> Fetch Operator
> limit: -1
> Processor Tree:
> TableScan
> alias: _dummy_table
> Row Limit Per Split: 1
> Select Operator
> expressions: '2023-11-27 04:37:23' (type: string)
> outputColumnNames: _col0
> ListSink
> {noformat}
> {code:sql}
> SELECT DATE_ADD('2023-01-01', 1)
> {code}
> {noformat}
> CBO PLAN:
> HiveProject(_o__c0=[DATE_ADD(_UTF-16LE'2023-01-01':VARCHAR(2147483647)
> CHARACTER SET "UTF-16LE", 1)])
> HiveTableScan(table=[[_dummy_database, _dummy_table]],
> table:alias=[_dummy_table])
> STAGE PLANS:
> Stage: Stage-0
> Fetch Operator
> limit: -1
> Processor Tree:
> TableScan
> alias: _dummy_table
> Row Limit Per Split: 1
> Select Operator
> expressions: DATE'2023-01-02' (type: date)
> outputColumnNames: _col0
> ListSink
> {noformat}
> {code:sql}
> SELECT DATE_SUB('2023-01-01', 1)
> {code}
> {noformat}
> CBO PLAN:
> HiveProject(_o__c0=[DATE_SUB(_UTF-16LE'2023-01-01':VARCHAR(2147483647)
> CHARACTER SET "UTF-16LE", 1)])
> HiveTableScan(table=[[_dummy_database, _dummy_table]],
> table:alias=[_dummy_table])
> STAGE PLANS:
> Stage: Stage-0
> Fetch Operator
> limit: -1
> Processor Tree:
> TableScan
> alias: _dummy_table
> Row Limit Per Split: 1
> Select Operator
> expressions: DATE'2022-12-31' (type: date)
> outputColumnNames: _col0
> ListSink
> {noformat}
> The reason is that all the functions at the moment are (wrongly) declared as
> a dynamic functions and this by default prevents constant folding. The
> functions are not dynamic since they don't depend on context variables and
> require one or more parameters.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
