[
https://issues.apache.org/jira/browse/MESOS-8908?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16479553#comment-16479553
]
Benjamin Mahler edited comment on MESOS-8908 at 5/17/18 6:50 PM:
-----------------------------------------------------------------
[~bbannier] thanks, patches look good to me. I'll have to follow up and test
whether it produces better stack traces and than {{perf \-\-call-graph dwarf}}
and if so, then perhaps we should disable flame pointer omission even with
{{\-\-enable-optimize}} (if the performance impact is small).
was (Author: bmahler):
[~bbannier] thanks, patches look good to me. I'll have to follow up and test
whether it produces better stack traces and than {{perf --call-graph dwarf}}
and if so, then perhaps we should disable flame pointer omission even with
{{--enable-optimize}} (if the performance impact is small).
> Add -fno-omit-frame-pointer to improve debugging and profiling.
> ---------------------------------------------------------------
>
> Key: MESOS-8908
> URL: https://issues.apache.org/jira/browse/MESOS-8908
> Project: Mesos
> Issue Type: Improvement
> Components: build, libprocess, stout
> Reporter: Benjamin Mahler
> Assignee: Benjamin Bannier
> Priority: Major
>
> Currently, we build without {{-fno-omit-frame-pointer}} which is needed to
> get reliable stack traces. Without this, obtaining {{perf}} profiles from
> users is not reliable:
> http://www.brendangregg.com/FlameGraphs/cpuflamegraphs.html#C
> Apparently, the performance win from frame pointer omission is very minor and
> many projects opt to disable the optimization given the benefits to debugging
> and profiling.
> I tried the {{--call-graph dwarf}} "workaround" for perf but still
> experienced {{[unknown]}} stack frames and what looked to be missing frames:
> http://www.brendangregg.com/perf.html#StackTraces (see dwarf section)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
