Tim Armstrong has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/8782 )

Change subject: IMPALA-5848: Account for TCMalloc overhead in MemTracker
......................................................................


Patch Set 1:

(2 comments)

http://gerrit.cloudera.org:8080/#/c/8782/1//COMMIT_MSG
Commit Message:

http://gerrit.cloudera.org:8080/#/c/8782/1//COMMIT_MSG@13
PS1, Line 13:
Can you include an example of how this looks?


http://gerrit.cloudera.org:8080/#/c/8782/1/be/src/runtime/mem-tracker.cc
File be/src/runtime/mem-tracker.cc:

http://gerrit.cloudera.org:8080/#/c/8782/1/be/src/runtime/mem-tracker.cc@298
PS1, Line 298:       tcmalloc_overhead = curr_consumption - 
TcmallocMetric::BYTES_IN_USE->value();
This is essentially assuming that the consumption metric is   
TcmallocMetric::PhysicalBytesMetric. This isn't entirely true - it is actually 
AggregateMemoryMetrics::TOTAL_USED, which contains non-TCMalloc memory if the 
experimental option --mmap_buffers=true is on.

It would be nice to get the accounting right for that case, even though it is 
experimental.

You could compute PHYSICAL_BYTES_RESERVED - BYTES_IN_USE here, but I think if 
you're doing that it make more sense to use the same approach as setting up 
"Buffer Pool: Free Buffers", etc, in runtime/exec-env.cc where we add an extra 
MemTracker pointing to a metric.



--
To view, visit http://gerrit.cloudera.org:8080/8782
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I602e9d5e8e8d7470dcfe4addde3265057c16263a
Gerrit-Change-Number: 8782
Gerrit-PatchSet: 1
Gerrit-Owner: Bikramjeet Vig <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Comment-Date: Wed, 06 Dec 2017 23:24:26 +0000
Gerrit-HasComments: Yes

Reply via email to