okumin commented on code in PR #5707:
URL: https://github.com/apache/hive/pull/5707#discussion_r2026950629
##########
ql/src/test/results/clientpositive/llap/bucket_map_join_tez3.q.out:
##########
@@ -1432,7 +1432,7 @@ STAGE PLANS:
Map-reduce partition columns: _col0 (type: date),
_col1 (type: decimal(38,0))
Statistics: Num rows: 1 Data size: 168 Basic stats:
COMPLETE Column stats: COMPLETE
tag: -1
- auto parallelism: true
+ auto parallelism: false
Review Comment:
Looks good.
Map 2 -> Reducer 3 -> Map 1(BMJ)
##########
ql/src/test/results/clientpositive/llap/bucket_map_join_tez3.q.out:
##########
@@ -480,7 +480,7 @@ STAGE PLANS:
Map-reduce partition columns: _col0 (type: date),
_col1 (type: decimal(38,0))
Statistics: Num rows: 1 Data size: 168 Basic stats:
COMPLETE Column stats: COMPLETE
tag: -1
- auto parallelism: true
+ auto parallelism: false
Review Comment:
Looks good.
Map 2 -> Reducer 3 -> Map 1(BMJ)
##########
ql/src/java/org/apache/hadoop/hive/ql/parse/GenTezUtils.java:
##########
@@ -182,6 +187,41 @@ public static ReduceWork createReduceWork(
return reduceWork;
}
+ /**
+ * Checks if there is a Bucket Map Join (BMJ) following a hierarchy:
+ * ReduceSinkOperator (RS) -> ReduceSinkOperator (RS) -> MapJoinOperator
(MJ).
+ * Ensures at most **two** RSOs before MJ.
Review Comment:
I guess it happens only when there are precisely two RSOs. Given that Map 3
-> Reducer 4 -> Reducer 5 -> Map 1(BMJ), `createReduceWork` is invoked twice.
When we create Reducer 4, the passed RSO belongs to Map 3. The process bails
out once we find the third RSO in Reducer 5.
When we create Reducer 5, we find two RSOs in Reduer 4 and Reducer 5, and
the BucketMapJoin.
A vertex with BucketMapJoin is a mapper. So, basically, RS -> MJ doesn't
happen.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
