[lkml]   [2014]   [Jul]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v2 0/3] perf/sdt : Support for SDT markers
Hi Masami and Hemant,

On Sun, 20 Jul 2014 12:16:46 +0900, Masami Hiramatsu wrote:
> (2014/07/20 2:32), Hemant Kumar wrote:
>>>> We have lots of applications which use SDT markers today, like:
>>>> Postgresql, MySql, Mozilla, Perl, Python, Java, Ruby, libvirt, QEMU, glib
>>>> To add SDT markers into user applications:
>>>> We need to have this header sys/sdt.h present.
>>>> sys/sdt.h used is version 3.
>>>> If not present, install systemtap-sdt-devel package (for fedora-18).
>>>> Please refer to the Documentation patch to see how the SDT markers are added into
>>>> a program.
>>>> With this patchset,
>>>> - Use perf to list the markers in the app:
>>>> # perf list sdt ./user_app
>>>> ./user_app :
>>>> %user_app:foo_start
>>>> %user_app:fun_start
>>>> - Also, we can see the SDT markers present in our system in the usual binaries.
>>>> These usual binaries are libraries (dsos) listed by ldconfig --print-cache and some
>>>> binaries present in PATH environment variable.
>>>> First, scan the binaries using :
>>>> # perf list sdt --scan
>>> At a glance, maybe we'd better have perf sdt-cache as like as perf buildid-cache
>>> for manage sdt information. what would you think?
>> I agree with you having perf sdt-cache similar to perf buildid-cache.
>> But I think if the functionality of perf sdt-cache is only to build the
>> cache, then we can
>> go with the perf list sdt --scan. Since, "perf list sdt" is used for
>> other purposes too, it
>> should be less confusing for the users to just add another option
>> (--scan) to create/modify
>> the cache. What do you suggest?
> I think there may be some other cases, for example adding user local build
> binary to the cache, or remove/update it locally. :)
> And also, in user's mental model of perf-list, it doesn't take an "active"
> action, that always does "passive" action. So adding such "active" scan option
> will be more confusing.
> But I also think it is OK that if the sdt is never scanned, the perf-list
> automatically scans in background (without any option) or suggests user
> to run "perf sdt-cache --scan". (it depends on how long time it may take)
> To summarize it, I'd like to suggest adding below functions;
> perf list sdt : shows all cached SDT events
> perf list sdt <file> : shows SDT events in <file>
> perf sdt-cache --scan/-s : scans all system binaries/libraries + added files
> perf sdt-cache --add/-a <file(s)> : add SDT events in <file> to cache
> perf sdt-cache --remove/-r <file(s)>: remove SDT events in <file> from cache
> And if perf list can't find sdt-cache, it would suggest to run
> perf sdt-cache --scan or run it silently. :)
> What would you think about this?

I agree with Masami's idea. Maybe we can add a config option for the
auto-scan thing.

And I think this series can be split to pieces for ease review. :)
IOW, please consider make it something like below steps:

1. add raw SDT parsing code
2. support perf list to print SDT markers from a single file
3. implement perf sdt-cache like Masami suggested
4. improve perf list sdt to use cached result

I guess it's easy to make a progress on step 1 and 2. And we can
discuss further how to shape/improve step 3 and 4 later.


 \ /
  Last update: 2014-07-21 05:01    [W:0.052 / U:0.800 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site