Tim Armstrong has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/11103 )

Change subject: IMPALA-7096: restore scanner thread memory heuristics
......................................................................


Patch Set 9: Code-Review+2

(6 comments)

Carry +2

http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.h
File be/src/runtime/scanner-mem-limiter.h:

http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.h@32
PS7, Line 32: limiting the aggregate memory consumpt
> is it to limit the number of scanner threads, or the aggregate memory consu
Done


http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.h@42
PS7, Line 42: as this objec
> garbled
Done


http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.h@43
PS7, Line 43: (i.e. as long as the below methods
> is that requirement because the instance of this class happens to also be a
Done


http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.h@71
PS7, Line 71:   /// ClaimMemoryForScannerThread() will not be called.
> do we need that? why not just iterate over the map? do we want ordering but
I started off with the vector then kept it as an optimisation to allow more 
efficient iteration in ClaimMemoryForScannerThread. Now that I look at it again 
I doubt there's a significant enough difference between this and unordered_map 
to justify the complexity.


http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.cc
File be/src/runtime/scanner-mem-limiter.cc:

http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.cc@75
PS7, Line 75: e a crude heuristic of guessing that the scan
            :       // will
> what code? before this change or before the change that removed the origina
Added a more concrete description of which commits added/removed the code.


http://gerrit.cloudera.org:8080/#/c/11103/7/be/src/runtime/scanner-mem-limiter.cc@81
PS7, Line 81:         addtl_consumption += static_cast<int64_t>((consumption * 
1.5) / num_threads);
            :       }
> I don't really understand that. Why would adding this thread increase consu
Yeah exactly. Tried to improve the comment.



--
To view, visit http://gerrit.cloudera.org:8080/11103
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: Ib9907fa8c4d2b0b85f67f4f160899c1c258ad82b
Gerrit-Change-Number: 11103
Gerrit-PatchSet: 9
Gerrit-Owner: Tim Armstrong <[email protected]>
Gerrit-Reviewer: Bikramjeet Vig <[email protected]>
Gerrit-Reviewer: Dan Hecht <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Comment-Date: Wed, 15 Aug 2018 21:20:58 +0000
Gerrit-HasComments: Yes

Reply via email to