| Re: [tracecompass-dev] Use Trace compass internal storage for large trace streaming mode |
|
Hi,
Not sure what is going on and why it's not working. The signal TmfTraceRangeUpdatedSignal is the right one to notify the analysis modules that new trace data is available. It is supposed to create a request to read the trace from the provide start and end time.
In streaming mode, this signal is sent repeatedly, every time new trace data is available which you have some kind of logic implemented. In non-streaming mode, this signal is only sent once when the trace (or experiment) instance handles the TmfTraceOpenedSignal.
To figure out what is going on, I would suggest to debug and have a breakpoint when the TmfTraceRangeUpdatedSignal is sent and received. Analysis modules of type TmfStateSystemAnalysisModule handle that signal in method traceRangeUpdated() (see here
https://github.com/eclipse-tracecompass/org.eclipse.tracecompass/blob/5704617daf69f3e5951b38180cec21340ba7b99b/tmf/org.eclipse.tracecompass.tmf.core/src/org/eclipse/tracecompass/tmf/core/statesystem/TmfStateSystemAnalysisModule.java#L820)
Maybe also check if the TmfTraceOpenedSignal was handled (TmfTrace or TmfExperiment in case of an experiment)
BTW, The getStreamingInterval() method of TmfTrace / TmfExperiment determines if it's a streaming trace (value > 0) or a offline trace (value = 0).
I hope this helps.
Regards
Bernd
From: Vinod Appu <vinod.appu@xxxxxxxxxxx>
Sent: August 20, 2024 6:18 AM To: Bernd Hufmann <bernd.hufmann@xxxxxxxxxxxx>; Matthew Khouzam <matthew.khouzam@xxxxxxxxxxxx>; tracecompass-dev@xxxxxxxxxxx <tracecompass-dev@xxxxxxxxxxx> Subject: RE: Use Trace compass internal storage for large trace streaming mode Hi,
Any guidance or alternate approach suggestions?
~Vinod
From: Vinod Appu
Hi,
We already have a function flow graph using trace compass working, where we do analysis only once. What I’m trying to do is adding streaming support, when analysis in progress, new data coming and it’s getting analysed and visualized. Obviously, new data can come or cannot. Sorry to say that when you say “streaming is not supported out of the box & there is some code to achieve that”, It’s chaotic for me.
I’ve tried below, was expecting this to work. But I see no events coming for analysis, AbstractTMFStateProvider$EventProcessor#fEventsQueue has no data coming.
MyTrace extends TmfTrace{ @Override public boolean isComplete() { return false; }
@Override public long getStreamingInterval() { // TODO: What is the relevance of streaming interval? return 300; } // Update as events coming, now a timer thread doing this public void sendStreamingRangeUpdatedEvent() { if (!isTraceOpened.getAndSet(true)) { fNbEvents = functionEvents.size(); try { seek(0); final ITmfContext context = seekEvent(0); getNext(context); context.dispose(); } catch (IOException e) { Activator.log(e); } TmfSignalManager.dispatchSignal(new TmfTraceOpenedSignal(this, this, null)); } else { final TmfTimeRange timeRange = new TmfTimeRange(getStartTime(), TmfTimestamp.fromNanos(System.currentTimeMillis() * 1000)); final TmfTraceRangeUpdatedSignal rangeUpdatedsignal = new TmfTraceRangeUpdatedSignal(this, this, timeRange); broadcastAsync(rangeUpdatedsignal); } } }
Your guidance will allow us to use the framework effectively! ~Vinod
From: Bernd Hufmann <bernd.hufmann@xxxxxxxxxxxx>
Hi,
First of all, what do you mean with streaming function flow trace? Do you mean that you have stream trace data into a trace that Trace Compass is then supposed to analyze and when new trace data arrives continue with the analysis (live trace analysis)? Please clarify.
Streaming and live trace analysis is not fully supported at the moment. There is some code to achieve that and views are made to allow to display data while the trace (offline trace) is analysed. But handling streamed trace is not supported out of the box.
If you store data in heap you run into issue of running out of heap, depending on trace size and size of heap. That's why in Trace Compass intermediate storages on disk like state system exists.
I think you can use the state system to store the data and then have data providers (another Trace Compass concept) to query the state system and further analyse or visualize the data.
Since you dealing with function entries/exits you might want to look into the flame chart analysis (or also called call stack analysis) and how it uses the state system apis to store that. In particular, it use the pushAttribute and popAttribute feature of the state system.
Here is the state provider for the call stack analysis of LTTng UST traces with function entry/exit (from finstrument functions of gcc). It extends a generic CallStackStateProvider
Once you have created your own state provider extending the CallStackStateProvider then you can create an analysis module extending https://github.com/eclipse-tracecompass/org.eclipse.tracecompass/blob/master/analysis/org.eclipse.tracecompass.analysis.profiling.core/src/org/eclipse/tracecompass/analysis/profiling/core/instrumented/InstrumentedCallStackAnalysis.java. After adding the analysis module to your plugin.xml then you should be able to see the a call stack view out-of-the-box in Trace Compass
I hope this helps. If you have any more questions please let us know
Regards Bernd
From: tracecompass-dev <tracecompass-dev-bounces@xxxxxxxxxxx>
on behalf of Vinod Appu via tracecompass-dev <tracecompass-dev@xxxxxxxxxxx>
Hi,
I’m trying to implement a streaming function flow trace (function entry and exit), where the below specific case is getting handled.
Below is what I’m trying to achieve in theory.
Please share your thoughts, Could you please share the major areas I must investigate?
~Vinod |