[lttng-dev] Trace Compass in live session

Jonathan Rajotte-Julien jonathan.rajotte-julien at efficios.com
Wed Jun 13 07:57:39 EDT 2018

Hi Thierry,

----- Original Message -----
> From: "BADEUIL Thierry" <thierry.badeuil at thalesgroup.com>
> To: "Jonathan Rajotte-Julien" <jonathan.rajotte-julien at efficios.com>
> Cc: "lttng-dev" <lttng-dev at lists.lttng.org>
> Sent: Wednesday, June 13, 2018 3:48:44 AM
> Subject: RE: [lttng-dev] Trace Compass in live session

> Hi,
> The end of the goal is to debug and understand how 3 applications worked
> together on my embedded processor. That is why I am using sched_switch option
> of LTTng. I can be reproduced easily. I need to work in live session because I
> do not have enough memory for the LTTng traces and I should need a lot of
> memory.

Well in that case let's remove the usage of the lttng-live feature. It is unnecessary here.
What you need to do is to stream the session to lttng-relayd, reproduce your interaction, stop the session,
validate that no events were dropped, read the session with LTTng Scope [1]/TraceCompass.

On the embedded system:
    lttng create --set-url="net://" (you can drop the port since they are the "default" ones)
    lttng enable-event -k 'sched_*' 
    lttng start
    <do your thing here>
    lttng stop
    lttng destroy

Once the "stop" command is issued, you can read the trace one the HOST using Babeltrace, LTTng Scope [1], TraceCompass.

For you use case you do NOT need the "live" feature.

[1] https://lttng.org/beta/

> -----Message d'origine-----
> De : Jonathan Rajotte-Julien [mailto:jonathan.rajotte-julien at efficios.com]
> Envoyé : mardi 12 juin 2018 19:04
> À : BADEUIL Thierry
> Cc : lttng-dev
> Objet : Re: [lttng-dev] Trace Compass in live session
> Hi Thierry,
> Glad you got the base setup working.
> Please reply to the other thread indicating what was necessary to make it work
> on your end. This might be useful for other people in the future.
> Now my question is what is the end goal of using the live streaming mode. Is it
> really necessary?
> What are you trying to achieve overall?
> Are you trying to debug a certain scenario on you platform that can be
> reproduced easily?
> Are you simply checking out how LTTng works?
> Please provide more context so we can help you.
> See more inline.
> ----- Original Message -----
>> From: "BADEUIL Thierry" <thierry.badeuil at thalesgroup.com>
>> To: "lttng-dev" <lttng-dev at lists.lttng.org>
>> Sent: Tuesday, June 12, 2018 12:29:22 PM
>> Subject: [lttng-dev] Trace Compass in live session
>> Dear all,
>> I have difficulties to use Babeltrace in live session and I would like to
>> use Trace Compass.
> What problem are you hitting?
> TraceCompass does not support the live protocol. See [1] [2] for why.
> [1] https://bugs.eclipse.org/bugs/show_bug.cgi?id=486728
> [2] https://git.eclipse.org/r/#/c/69285/
> [TB1] : Yes I know that but when the trace files are updated, TraceCompass did
> not upated it even with F5. I do not understand why !

What do you mean by "trace files are updated"? Note that if the problem is on the TraceCompass side in the end, this is not
the mailing list to discuss it. TraceCompass have its own mailing list/irc channel see 
http://tracecompass.org/#community for more details.

>> My target activated LTTng with :
>> -          Lttng create test –live –set-url:’net://’
> This should be: lttng create test --set-url="net://" --live
> [TB1] : both solution worked on my target. Is there any difference between the
> two syntaxes ?

Not sure how --set-url":" works since it does not on my machine but if it works on yours +1.

>> -          Lttng enable-event sched_switch
>> -          Lttng start
>> My host (@ activated LTTng with :
>> -          Lttng-relayd –vv
> Since this seems to be working you can drop the verbose options (-vvv).
> [TB1] : OK !
>> Everything seems to be OK because my host received bytes from my target
>> during 1 minute.
> Does the event timestamps show a 1 minute difference on the console? Note that
> even if you see events outputted to the console during
> 1 minute does not necessarily means that a full minute of "events" is present
> since there might be latency between a "lttng stop" and a
> complete live reception. It take more time to serialize (in time) events than to
> produce them.
> [TB1] : If I set 1 minute of timestamp, I can watch only 1 minute of events on
> TraceCompass. I have got the same problem even I stopped or not lttng-relayd.
> If I set 2 minutes, I can watch 2 minutes etc...

What do you mean by "set 1 minute of timestamp"?

> [TB1] : same probleme with lttng-relayd stopped or not. Remember that
> TraceCompass is on a host and lttng session is on my target. The connection
> between is via Ethernet.

Yes this is pretty clear to me. We are talking of the same scenario.

Please check out the diagram found here: https://lttng.org/docs/v2.10/#doc-plumbing

The "Viewing System" here is NOT necessary since you will be conducting analysis "offline" from a trace gathered by the lttng-relayd process.


More information about the lttng-dev mailing list