[
https://issues.apache.org/jira/browse/HIVE-25589?focusedWorklogId=773885&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-773885
]
ASF GitHub Bot logged work on HIVE-25589:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 24/May/22 06:16
Start Date: 24/May/22 06:16
Worklog Time Spent: 10m
Work Description: kgyrtkirk commented on code in PR #3266:
URL: https://github.com/apache/hive/pull/3266#discussion_r880096997
##########
ql/src/java/org/apache/hadoop/hive/ql/parse/SemanticAnalyzer.java:
##########
@@ -1839,6 +1839,15 @@ boolean doPhase1(ASTNode ast, QB qb, Phase1Ctx ctx_1,
PlannerContext plannerCtx)
doPhase1GetDistinctFuncExprs(qbp.getAggregationExprsForClause(ctx_1.dest)));
break;
+ case HiveParser.TOK_QUALIFY:
+ if (!HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_CBO_ENABLED)) {
Review Comment:
I think when we fall-back to the non-cbo path (more likely in nature) this
flag will not be changed see around:
https://github.com/apache/hive/blob/35d4532b0c08f4f5fbb5dc897c4330cba434bc7c/ql/src/java/org/apache/hadoop/hive/ql/parse/CalcitePlanner.java#L706
I think `set hive.cbo.fallback.strategy=ALWAYS` could also be used to check
if that would happen
Issue Time Tracking
-------------------
Worklog Id: (was: 773885)
Time Spent: 50m (was: 40m)
> SQL: Implement HAVING/QUALIFY predicates for ROW_NUMBER()=1
> -----------------------------------------------------------
>
> Key: HIVE-25589
> URL: https://issues.apache.org/jira/browse/HIVE-25589
> Project: Hive
> Issue Type: Improvement
> Components: CBO, SQL
> Affects Versions: 4.0.0
> Reporter: Gopal Vijayaraghavan
> Assignee: Krisztian Kasa
> Priority: Major
> Labels: pull-request-available
> Time Spent: 50m
> Remaining Estimate: 0h
>
> The insert queries which use a row_num()=1 function are inconvenient to write
> or port from an existing workload, because there is no easy way to ignore a
> column in this pattern.
> {code}
> INSERT INTO main_table
> SELECT * from duplicated_table
> QUALIFY ROW_NUMER() OVER (PARTITION BY event_id) = 1;
> {code}
> needs to be rewritten into
> {code}
> INSERT INTO main_table
> select event_id, event_ts, event_attribute, event_metric1, event_metric2,
> event_metric3, event_metric4, .., event_metric43 from
> (select *, ROW_NUMBER() OVER (PARTITION BY event_id) as rnum from
> duplicated_table)
> where rnum=1;
> {code}
> This is a time-consuming and error-prone rewrite (dealing with a messed up
> order of columns between one source and dest table).
> An alternate rewrite would be to do the same or similar syntax using HAVING.
> {code}
> INSERT INTO main_table
> SELECT * from duplicated_table
> HAVING ROW_NUMER() OVER (PARTITION BY event_id) = 1;
> {code}
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
