Re: [PATCH 3/3] perf events: add timehist option to record and report

[Arnaldo Carvalho de Melo <acme@ghostprotocols.net>]

Em Sat, Feb 19, 2011 at 10:32:16AM +0100, Ingo Molnar escreveu:
> * David Ahern <daahern@cisco.com> wrote:
> 
> > On 01/23/11 at 03:45:67 a syslog event noted an application restart.
> > Checking the applications logs it went silent. Why? What does perf say?
> > 
> > Oh, 2 days have gone by before said event is reported to engineering,
> > and the server was been rebooted to 'clear' the problem - hence
> > resetting monotonic clock.
> 
> I think that's a valid usecase and the feature you are suggestion is very useful.
> 
> Why not extend the regular trace output to also (optionally) output GTOD timestamps?

That is how this this feature request/submission originally started.

David tried with an extra header, then a synthesized event, then after
some discussion with Peter I started working on a monotonic clock event,
but got sidetracked with other stuff, he tried picking up from where I
left and here we are.

I think the summary is: we need a clock_gettime like interface so that
users can pick and choose what clock matches best whatever app they're
trying to merge logs.

- Arnaldo