Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [tracecompass-dev] Control flow view for user tracepoints?

Hi,

thanks for the explanations and the great tool :-)

Sure, I can talk about the application, it's a pretty general component-based framework using a distributed middleware, with multiple nodes communicating using messages. These nodes form a computation graph, where either incoming messages or periodic timers trigger computation, which creates new messages and in that way, data and/or control "flows" through the graph.

What I'm interested in is tracing the control flow through the graph and my main analysis question is how much waiting occurs within the framework. For example, how much delay is there between a new message coming in, and the corresponding processing being carried out.

For that, I have instrumented the core middleware library to tell me when a message is coming in, when processing starts and ends, and when messages are sent out.
 
It would be cool to visualize the threads transitioning between waiting for data and processing, and I think from the documentation I see how I would do that, so I'm going to try that next. 

Even cooler would be to have arrows like in the control flow view, to show how activation passes between threads. For example, due to internal queuing, it is it not always the case that the method invocation immediately following a received message is actually pertaining to that message. So, being able to draw an arrow from the receiving thread to the part of the chart where the actual processing for *that* message occurs, would be quite informative. I could not find out in the docs how I would go about doing that, however. Do you have a hint?

Mit freundlichen Grüßen / Best regards

Dr. Ingo Luetkebohle
Application Software (CR/AEA2) 

Tel. +49(711)811-12248
Fax +49(711)811-0
Ingo.Luetkebohle@xxxxxxxxxxxx


-----Ursprüngliche Nachricht-----
Von: tracecompass-dev-bounces@xxxxxxxxxxx [mailto:tracecompass-dev-bounces@xxxxxxxxxxx] Im Auftrag von Matthew Khouzam
Gesendet: Montag, 13. Juli 2015 18:48
An: tracecompass-dev@xxxxxxxxxxx
Betreff: Re: [tracecompass-dev] Control flow view for user tracepoints?

Hi!

A- Thanks for using Tracecompass. :)

B- The control flow view is a very specific view that depends on a state system being build, the state system is that of a linux kernel. So the CFV is a no go. However, the underlying view (TimeGraphView) can have an XML based state provider where you can tell it what states to used.
Please check out this link
http://archive.eclipse.org/tracecompass/doc/stable/org.eclipse.tracecompass.doc.user/Data-driven-analysis.html#Data_driven_analysis
on how to do that. Are you allowed to describe what your application is and how it is instrumented, we might be able to get you going in the right path.

Cheers,

Matthew

On 15-07-13 10:07 AM, Luetkebohle Ingo (CR/AEA2) wrote:
> Hi,
>
> I would like to use the control flow view for user-defined tracepoints. Simply loading a trace that includes lttng-ust traces does not work -- I see a histogram and a list of events, but no control flow view.
>
> I also tried adding procname, pid, and tid fields (with add-context), but no luck either.
>
> Am I missing something, or is this not possible?
>
> cheers,
>

_______________________________________________
tracecompass-dev mailing list
tracecompass-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit https://dev.eclipse.org/mailman/listinfo/tracecompass-dev


Back to the top