[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
Re: [tracecompass-dev] Strange behavior with flame graph
|
Wow, that's really cool!
This is pure conjecture, I think we may have a problem with adding
intervals, we are adding durations but they include t0 so it may
increment by one each element.
Would you be able to open a bug with this? (Also if you can contribute a
patch to fix it, we can review it quickly. :))
For the third item, I saw that in the incubator and Genevieve fixed it,
maybe we could backport that fix to master. I think it's this.
https://git.eclipse.org/r/#/c/102326/
Thanks.
Matthew
On 17-09-28 01:00 PM, Rocky Dunlap - NOAA Affiliate wrote:
> We have created a custom trace that extends CtfTmfTrace and an
> analysis by extending CallStackAnalysis and CallStackStateProvider.
> We are using TraceCompass 3.1. The analysis appears to be working
> correctly and the call stack view looks good. I also noticed that we
> get the Flame Graph "for free" since we have extended
> CallStackAnalysis. However, we are seeing some strange behavior with
> the Flame Graph.
>
> When I first load the trace, the "per thread" Flame Graph looks
> correct. See below where there are four threads (each in a separate
> process) and notice that the function named "[GCS] run #1" has 192
> calls in each thread.
>
> Inline image 1
>
> Then, I immediately switch to "aggregate threads view" and see that
> the total number of calls for "[GCS] run #1" is 772, but I think this
> should be 192*4 = 768.
>
> Inline image 2
>
> Then, the really strange part is when I switch back to "per thread"
> view I see that one of the threads shows 1355 calls to "[GCS] run
> #1". I would expect this view to be identical to the first time
> viewing the "per thread" view.
>
> Inline image 3
>
> Now, when I switch back to the aggregate view, the total number of
> calls for this function has gone up to 2530. I am suspicious that
> this is not the expected behavior. However, since I have created a
> custom CtfTmfTrace, maybe I do not have something set up in the state
> system / attribute tree. The strange this is that the call stack view
> itself works great. Keep in mind that these four threads are in four
> different processes--no idea if that may shed light on the strange
> behavior.
>
> Any ideas?
>
> Rocky
>
>
> _______________________________________________
> 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