Re: Adding latency tracking to trace-cmd

[<Viktor.Rosendahl@bmw.de>]

Hi Steve,

On Wed, 2021-02-24 at 11:54 -0500, Steven Rostedt wrote:
> Hi Viktor,
> 
> While doing some tests to trace-cmd, I found that the trace-cmd record was
> broken for latency tracers, and has been for some time. That's probably
> because nobody has been using it for such.
> 
> That is, trace-cmd record is mostly used for "flyrecord" which is the
> constant recording of tracing. But if the user records one of the latency
> tracers (preemptirqsoff, wakeup_rt, etc), then it switches to "latency"
> mode and only takes a snapshot at the end of the recording. This is what I
> found to be broken, because it reset the tracing before taking the
> snapshot and lost whatever it was recording. Nobody seemed to complain
> about it, so I guess nobody cared.
> 

I have used trace-cmd but I did not know about this feature. Somehow I had
assumed that trace-cmd would not deal with those tracers that work in overwrite
mode. 

> The fix is simple, and I was about to do so, but then I thought about your
> latency-collector tool, and thought that should be exactly what trace-cmd
> should do for such tracers. That is, instead of just taking a snapshot of
> the latency at the end of the recording (which it was suppose to do now),
> it should act like your latency-collector tool, and take snapshots
> every time there's a new latency.
> 
> My question to you is, would you like to work on adding that feature to
> trace-cmd?

I would be interested to work on the feature but I would probably need
permission from my employer, also in case I do it as a spare time project.

For this reason, I will need some time come to a decision. I will get back to
you when I know the answer.

> 
>   git://git.kernel.org/pub/scm/utils/trace-cmd/trace-cmd.git
> 
> (which requires the libtracefs and libtracevent libraries).
> 
> I'm thinking if someone were to do:
> 
>  trace-cmd record -p wakeup_rt 
> 
> It would then act just like the latency-collector and record every new
> instance of a latency into the trace.dat file, where:
> 
>  trace-cmd report
> 
> would then nicely format that data for the user.
> 
> Thoughts?

It seems like trace-cmd is mostly GPL-2.0. The latency-collector is also GPL-
2.0, so I guess that it would be possible to take snippets of code and
inspiration from it.

Would you allow the hackish random algorithm to be used in trace-cmd?
I mean the "--random" option in latency-collector.

best regards,

Viktor

> 
> -- Steve