Song Jiacheng has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/20716 )

Change subject: [util] add a warning into 'heap_sample_every_n_bytes' flag 
description.
......................................................................


Patch Set 8:

(7 comments)

> Patch Set 7:
>
> (1 comment)

http://gerrit.cloudera.org:8080/#/c/20716/7//COMMIT_MSG
Commit Message:

http://gerrit.cloudera.org:8080/#/c/20716/7//COMMIT_MSG@9
PS7, Line 9: Parameter --heap_sample_every_n_byt
> nit: --heap_sample_every_n_bytes
Done


http://gerrit.cloudera.org:8080/#/c/20716/7//COMMIT_MSG@10
PS7, Line 10: to a positive value will
> nit: a positive value
Done


http://gerrit.cloudera.org:8080/#/c/20716/7//COMMIT_MSG@11
PS7, Line 11: ,
> Remove it.
Done


http://gerrit.cloudera.org:8080/#/c/20716/7//COMMIT_MSG@12
PS7, Line 12: comp
> nit: compete the lock of page heap with other thread?
Done


http://gerrit.cloudera.org:8080/#/c/20716/7//COMMIT_MSG@18
PS7, Line 18:
> nit: --heap_sample_every_n_bytes
Done


http://gerrit.cloudera.org:8080/#/c/20716/7/src/kudu/util/flags.cc
File src/kudu/util/flags.cc:

http://gerrit.cloudera.org:8080/#/c/20716/7/src/kudu/util/flags.cc@100
PS7, Line 100:              "in tcmalloc; the lower the value, the higher the 
lock contention.");
> Is it necessary to remove the former description? 524288 is a recommend val
I have tested 524288 in a high throughput environment, and the stacks show that 
the apply threads of tserver always stuck in DoSampleAllocation() or Free(). 
There are 48 apply threads per tserver, but only a few of them could work at 
the meantime.
Taking one sample once every 512KB allocation is really too often in production 
environment.


http://gerrit.cloudera.org:8080/#/c/20716/7/src/kudu/util/flags.cc@99
PS7, Line 99: Warning: Setting this flag to a positive value increases lock 
contention "
            :              "in tcmalloc; the lower the value, the higher the 
lock contention.
> It seems better to set --heap_sample_every_n_bytes more larger to lower the
I think it should depend on the memory limit and throughput of the tserver, 
like 1% of hard limit could be accepted.



--
To view, visit http://gerrit.cloudera.org:8080/20716
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I00d55126f0cad0ac6cc130576cc45808f96f0a47
Gerrit-Change-Number: 20716
Gerrit-PatchSet: 8
Gerrit-Owner: Song Jiacheng <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: Marton Greber <[email protected]>
Gerrit-Reviewer: Song Jiacheng <[email protected]>
Gerrit-Reviewer: Wang Xixu <[email protected]>
Gerrit-Reviewer: Yingchun Lai <[email protected]>
Gerrit-Comment-Date: Mon, 18 Dec 2023 07:22:00 +0000
Gerrit-HasComments: Yes

Reply via email to