Joe McDonnell has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/16020 )

Change subject: IMPALA-9318: Add admission control setting to cap MT_DOP
......................................................................


Patch Set 3:

(4 comments)

http://gerrit.cloudera.org:8080/#/c/16020/3//COMMIT_MSG
Commit Message:

http://gerrit.cloudera.org:8080/#/c/16020/3//COMMIT_MSG@16
PS3, Line 16: unlimited MT_DOP
> do you mean there is no limit applied like with negative value?
Yes, changed this to say that it defaults to -1 which applies no limit


http://gerrit.cloudera.org:8080/#/c/16020/3/be/src/service/impala-server.cc
File be/src/service/impala-server.cc:

http://gerrit.cloudera.org:8080/#/c/16020/3/be/src/service/impala-server.cc@914
PS3, Line 914: EnforceMaxMtDop
> since the config values can change while the query is in admission controll
A couple thoughts:

I think we want to avoid having a difference in behavior when you change the 
policy vs when it is stable. In other words, if we are usually downgrading 
queries, it is harder for users to understand if we switch to throwing an error 
for this transitional situation.

I think it is ok for us to say that the policy is evaluated before planning and 
does not get updated after that. We wouldn't be able to modify queued queries, 
but I think that is a behavior that administrators can work with. I updated the 
commit comment to mention this behavior.

Is this ok?


http://gerrit.cloudera.org:8080/#/c/16020/3/common/thrift/ImpalaInternalService.thrift
File common/thrift/ImpalaInternalService.thrift:

http://gerrit.cloudera.org:8080/#/c/16020/3/common/thrift/ImpalaInternalService.thrift@594
PS3, Line 594:   26: optional i32 overridden_mt_dop_value = -1;
> Doesn't matter too much, but would probably be more consistent with other o
Good point, changed this to use that


http://gerrit.cloudera.org:8080/#/c/16020/3/testdata/workloads/functional-query/queries/QueryTest/max-mt-dop.test
File testdata/workloads/functional-query/queries/QueryTest/max-mt-dop.test:

http://gerrit.cloudera.org:8080/#/c/16020/3/testdata/workloads/functional-query/queries/QueryTest/max-mt-dop.test@7
PS3, Line 7: select count(*) from tpch.lineitem;
> tpch_parquet.lineitem would also be significantly faster, since it turns in
I think I was trying to be smart and avoid planner downgrades, but it actually 
doesn't matter. The settings would be the same regardless of the query 
complexity.

Switched this to "select 1"



--
To view, visit http://gerrit.cloudera.org:8080/16020
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I3affb127a5dca517591323f2b1c880aa4b38badd
Gerrit-Change-Number: 16020
Gerrit-PatchSet: 3
Gerrit-Owner: Joe McDonnell <[email protected]>
Gerrit-Reviewer: Bikramjeet Vig <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Joe McDonnell <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Comment-Date: Fri, 05 Jun 2020 22:42:47 +0000
Gerrit-HasComments: Yes

Reply via email to