> On Aug. 1, 2018, 2:39 a.m., Jesús Camacho Rodríguez wrote:
> > ql/src/java/org/apache/hadoop/hive/ql/exec/OperatorUtils.java
> > Lines 451 (patched)
> > <https://reviews.apache.org/r/68124/diff/3/?file=2065696#file2065696line451>
> >
> >     We can remove this first block, it does not buy us much in terms of 
> > algorithm perfomance, and method would have no restriction on start 
> > operator (plus more readable).
> 
> Deepak Jaiswal wrote:
>     No. It wont work without it. It is not for performance, it is for 
> correctness. The start in our case is the RS2, going up wont work as it will 
> stop when it encounters RS1.
>     The more generic one is in SharedWorkOptimizer, this one, I am afraid is 
> for this particular case.
> 
> Jesús Camacho Rodríguez wrote:
>     The block can be part of the caller logic, so if you have the chain:
>     SEL->GBY1->RS1->GBY2->RS2
>     then you end up passing the SEL as the start operator.
>     
>     
>     Then the method in OperatorUtils has no restriction and it is reusable: 
> given any operator, 1) output all the operators contained in the same work, 
> 2) gather all the terminal operators of that work, and 3) gather all the 
> semijoin branches of that work.

Thanks.


- Deepak


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/68124/#review206717
-----------------------------------------------------------


On Aug. 1, 2018, 12:27 a.m., Deepak Jaiswal wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/68124/
> -----------------------------------------------------------
> 
> (Updated Aug. 1, 2018, 12:27 a.m.)
> 
> 
> Review request for hive, Jesús Camacho Rodríguez and Jason Dere.
> 
> 
> Bugs: HIVE-20252
>     https://issues.apache.org/jira/browse/HIVE-20252
> 
> 
> Repository: hive-git
> 
> 
> Description
> -------
> 
> See Jira.
> 
> removeSemiJoinCyclesDueToMapsideJoins is deprecated, although it has changes. 
> I will eventually remove it and can be ignored.
> 
> 
> Diffs
> -----
> 
>   ql/src/java/org/apache/hadoop/hive/ql/exec/OperatorUtils.java 7b2ae40107 
>   ql/src/java/org/apache/hadoop/hive/ql/parse/ParseContext.java 538aa5e924 
>   ql/src/java/org/apache/hadoop/hive/ql/parse/TezCompiler.java c3eb886fd2 
> 
> 
> Diff: https://reviews.apache.org/r/68124/diff/3/
> 
> 
> Testing
> -------
> 
> 
> Thanks,
> 
> Deepak Jaiswal
> 
>

Reply via email to