Real time plotting slow: FigureCont​roller.flu​shCoalesce​r needs a lot of memory and cpu time

Dear all,

I'm trying to plot data points from a serial device as fast as possible and I think, almost all best practices have been followed. The plot concept is approx like this (it's much more complicated, because it's part of a gui in reality)

classdef SensorPlots < handle
    properties
        HFig
        AxesHandles
        LineHandles
        Data = nan(300,4) % preallocate 300 samples, 4 different plots
        Time = nan(300,4)
        ToBePlottedIdx = false(300,4) % indicator, which data needs to be plotted
        LastPlottedIdx = repmat({nan},4,1) % idx to avoid dynamic growth of XData/YData
    end
    
    methods
        function obj = SensorPlots
           % load pre configured figure
           file = "pathToFig/fig.fig";
           obj.HFig = openfig(file);
           hh = findobj(obj.HFig,'type','axes');
           
           obj.AxesHandles = hh; % in reality, there are multiple axes
           obj.LineHandles = plot(obj.AxesHandles, obj.Time, obj.Data, '.');
           
           % first performance tweaks
           hh = obj.AxesHandles;
           disableDefaultInteractivity(hh);
           hh.Toolbar.Visible = 'off';
           hh.Toolbar = [];
           hh.Interactions = [];
           hh.XAxis.LimitsMode = 'manual';
           
           
           % timer to periodically update plots
           obj.plotTimer = timer('ExecutionMode', 'fixedRate', 'Period', 0.05, 'StartDelay', 0.5, 'TimerFcn', @obj.plotData, 'ErrorFcn', @obj.errorTimer, 'StopFcn', @obj.stopPlotTimer);
        end
        
        %% Cache the data in properties Data and Time
        % Is called by a Model in MVP context, based on an C# event class. 
        % gets called every ~40 ms
        % dataPack is a struct with fields accX, accY, accZ, temp
        function addData(obj, dataPack) 
            fn = fieldnames(dataPack);
            for k = 1:length(fn)
                len = length(dataPack.(fn{k}).value);
                freeIdx = find(~obj.ToBePlotted(:,k),len,'first');
                if isempty(freeIdx)
                    %                     disp('plot too slow')
                end
                obj.Data(freeIdx,k) = cat(1, dataPack.(fn{k}).value);
                obj.Time(freeIdx,k) = cat(1, dataPack.(fn{k}).lastUpdate);
                obj.ToBePlotted(freeIdx,k) = true;
            end
        end
    
        %% PlotData callback. 
        function plotData(obj, sender, evnt)
            for iDat = 1:size(obj.Data,2) % 4 cols -> 4 lines plots
                
                plotIdx = find(obj.ToBePlotted(:,iAng));
                if isempty(plotIdx)
                    continue;
                end
                len = nnz(plotIdx);             
                
                try
                    currLastPlottedIdx = obj.LastPlottedIdx{iDat};
                    
                    % currLastPlottedIdx not initialized yet?
                    if any(isnan(currLastPlottedIdx))
                        currLastPlottedIdx = 1:len;
                    else
                        currLastPlottedIdx = currLastPlottedIdx(end) + 1:len;
                    end
                    
                    % restrict indices not to be greater than preallocated XData
                    if(currLastPlottedIdx > size(obj.Data,1))
                        currLastPlottedIdx = currLastPlottedIdx - min(currLastPlottedIdx) + 1;
                    end
                    
                    % plot data
                    obj.LineHandles(iData).XData(currLastPlottedIdx) = obj.Time(plotIdx,iDat)';
                    obj.LineHandles(iData).YData(currLastPlottedIdx) = obj.Data(plotIdx,iDat)';
                    
                    maxTime = max(obj.Time(plotIdx,iDat)');
                    obj.LastPlottedIdx{iDat} = currLastPlottedIdx;
                
                catch e
                    e
                end
                
                obj.ToBePlotted(plotIdx,iAng) = false;
                meanAng(iAng) = mean(obj.Data(plotIdx,iAng));
            end       
            
            % try not to update the axes limits all the time, but each 500 ms
            currentToc = toc(obj.StartTic);
            if (currentToc - obj.LastLimitSetTime) > 0.5
                for hh = each(obj.AxesHandles)
                    hh.XAxis.Limits = [maxTime-10,maxTime+5];
                end
                drawnow limitrate
                obj.LastLimitSetTime = currentToc;
            end 
        end
    end
end

Do you see any obvious problems with that code? I tried to avoid high level functions an preallocate everything.

The profiler shows that output:

plotData takes the most time here:

Which seems reasonable, it's a lot of data. But what happens to the FigureController? Look at the memory + cpu time

Could anybody help me out with that? Why is this happening?

Thank you all very much!

[Jan Kappen]

Same as above with Matlab 2019b. Any ideas?

[Arturo Pino]

I'm also having a very similar issue, program getting hung up using figurecontroller.flushcoalescer. Has anyone seen this issue or find a resolution? I noticed in the figurecontroller method, the following comment was added by Mathworks:

%Used by figure.flushCoalescer() to implement the short-term solution
%for drawnow property updates. When the long-term solution is
%implemented, this method can be removed; see g1658467.

It then just gets stuck in a loop checking if "this.PeerModeInfo" has been set.

[Eleanor Betton]

I am still getting this in Appdesigner on MatLan 2021a. Any advice?

[li li]

I am still getting this in Appdesigner on matlab 2020b.

[li li]

I tried to change BusyMode = 'drop' and the error no longer appears.

[Jan Kappen]

@li li where did you set the BusyMode? Which callback or listener? Sounds very interesting!