Michael Smith has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/21687 )

Change subject: IMPALA-5443: Apply codegen rows threshold per node
......................................................................


Patch Set 3:

(1 comment)

http://gerrit.cloudera.org:8080/#/c/21687/2/fe/src/main/java/org/apache/impala/util/PlanNodeCodegenVisitor.java
File fe/src/main/java/org/apache/impala/util/PlanNodeCodegenVisitor.java:

http://gerrit.cloudera.org:8080/#/c/21687/2/fe/src/main/java/org/apache/impala/util/PlanNodeCodegenVisitor.java@58
PS2, Line 58:     if (numRowsPerInstance < rowsThreshold_) {
Is there another threshold where it makes sense to codegen if rows are not yet 
available due to a blocking operator?

That starts to trade additional work in the executor for lower query latency 
(doing codegen because the fragment is dormant, even though it will be more 
expensive than interpreting for a few rows). Can we detect when the executor 
isn't very busy?



--
To view, visit http://gerrit.cloudera.org:8080/21687
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I1b34d4f2ef0d98fcd918a8a546acde70e09d18d7
Gerrit-Change-Number: 21687
Gerrit-PatchSet: 3
Gerrit-Owner: Michael Smith <[email protected]>
Gerrit-Reviewer: Csaba Ringhofer <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Michael Smith <[email protected]>
Gerrit-Comment-Date: Mon, 07 Oct 2024 17:40:37 +0000
Gerrit-HasComments: Yes

Reply via email to