[Regression] GridPane crashes (@1kHz update rates) - FIXED

[ennerf]

I tried to do a few long-running tests with live-charts at 1KHz and noticed that the latest dev version keeps crashing after 1-2 hours. I did some 24h+ tests last week, so this must be a recent regression.

I've tried a few different versions and it looks like this was introduced in minor: made drawing of grid on top/bottom CSS more easily configurable (4d8ebc3)

• e8aa9ff - broken
• c20f699 - broken
• dd894e5 - broken
• 4d8ebc3 - broken
• 6bb9276 - edit: failed after >5 hours :(

The full error is below. The application still runs, but the stage with the chart needs to be closed.

Exception in thread "JavaFX Application Thread" java.lang.NullPointerException
	at javafx.graphics/javafx.scene.layout.GridPane.growToMultiSpanPreferredWidths(Unknown Source)
	at javafx.graphics/javafx.scene.layout.GridPane.adjustColumnWidths(Unknown Source)
	at javafx.graphics/javafx.scene.layout.GridPane.computeWidthsToFit(Unknown Source)
	at javafx.graphics/javafx.scene.layout.GridPane.computeMinHeight(Unknown Source)
	at javafx.graphics/javafx.scene.Parent.minHeight(Unknown Source)
	at javafx.graphics/javafx.scene.layout.Region.minHeight(Unknown Source)
	at javafx.graphics/javafx.scene.layout.Region.boundedNodeSizeWithBias(Unknown Source)
	at javafx.graphics/javafx.scene.layout.Region.layoutInArea(Unknown Source)
	at javafx.graphics/javafx.scene.layout.Region.layoutInArea(Unknown Source)
	at javafx.graphics/javafx.scene.layout.Region.layoutInArea(Unknown Source)
	at javafx.graphics/javafx.scene.layout.StackPane.layoutChildren(Unknown Source)
	at javafx.graphics/javafx.scene.Parent.layout(Unknown Source)
	at javafx.graphics/javafx.scene.Parent.layout(Unknown Source)
	at javafx.graphics/javafx.scene.Parent.layout(Unknown Source)
	at javafx.graphics/javafx.scene.Parent.layout(Unknown Source)
	at de.gsi.chart.ui.HiddenSidesPane.resizeRelocate(HiddenSidesPane.java:252)
	at javafx.graphics/javafx.scene.layout.AnchorPane.layoutChildren(Unknown Source)
	at javafx.graphics/javafx.scene.Parent.layout(Unknown Source)
	at javafx.graphics/javafx.scene.Scene.doLayoutPass(Unknown Source)
	at javafx.graphics/javafx.scene.Scene$ScenePulseListener.pulse(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.Toolkit.lambda$runPulse$2(Unknown Source)
	at java.base/java.security.AccessController.doPrivileged(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.Toolkit.runPulse(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.Toolkit.firePulse(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.quantum.QuantumToolkit.pulse(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.quantum.QuantumToolkit.pulse(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.quantum.QuantumToolkit.pulseFromQueue(Unknown Source)
	at javafx.graphics/com.sun.javafx.tk.quantum.QuantumToolkit.lambda$runToolkit$11(Unknown Source)
	at javafx.graphics/com.sun.glass.ui.InvokeLaterDispatcher$Future.run(Unknown Source)
	at javafx.graphics/com.sun.glass.ui.gtk.GtkApplication._runLoop(Native Method)
	at javafx.graphics/com.sun.glass.ui.gtk.GtkApplication.lambda$runLoop$11(Unknown Source)
	at java.base/java.lang.Thread.run(Unknown Source)

[ennerf]

In this case the charts were drawn without a grid, and the only CSS styling was dataSet.setStyle(...)

Edit: tested on OpenJDK 13 + OpenJFX 13 and OpenJDK12 + OpenJFX 12.0.2

[ennerf]

Damn, 6bb9276 failed as well. I'm not sure what to try next.

[ennerf]

This may be a Linux-only issue. I couldn't reproduce it on Windows yet.

[RalphSteinhagen]

Tried to reproduce, unfortunately, as of now (~11h), cannot reproduce this feature.

We started another long-term test over the weekend using the RollingBufferSample, RollingBufferSortedTreeSample, and some other additional samples.

I noticed that you use a different JDK/OpenJFX version. If this isn't too much trouble, could you test your observation with JDK11 and OpenJFX 12.0.2?

Will keep my first test running until tomorrow morning and then start ramping up to 1 kHz notification rates. Could you send me your parameters (ie. number of samples, traces/datasets, line/marker configurations, fixed/auto-size of chart, etc.) so we can setup a test case. Alternatively, you may send in also a MVE. 😃

N.B. I updated the issue title since this seems to be specific for 1kHz update rates which we haven't tackled (until now). The usual chart update rates we test are 25, seldom 50 and max 100 Hz.

[RalphSteinhagen]

sorry, saw your "Linux" comment only now. Could you try to periodically call the garbage collector (e.g. calling System.gc() twice @0.2 .. 0.5 Hz)? We noticed that the JVM has sometimes hick-ups if it has to do too deal with too much garbage collection at once. Alternatively, one could also temporarily reduce the maximum available memory of the JVM to force more frequent garbage collections -- just for testing/checking if it is GC-related.

[ennerf]

Sorry, I've been pretty busy with other things. I'll try to get back to you soon. I haven't seen anything out of the ordinary in flight recorder and the gc logs, but I'll give it a try.

[RalphSteinhagen]

We did some longer running tests (>3 days) with no success... next time it crashes, could you provide a thread dump (e.g. using VisualVM)

Will try to write a test example with 1 kHz update: Could you provide me with your boundary parameters: numbers of samples, traces, markers on/off, renderer and/or renderer style: error bars/surface/none, chart dimensions etc.

Thanks in advance.

[RalphSteinhagen]

@ennerf could you check your issue against dev?

I exchanged the existing auto-notification suppression guard with an additional write lock guard around the recomputeLimits(..) (N.B. also by itself write lock protected) in XYChart:
https://github.com/GSI-CS-CO/chart-fx/blob/425a8ea024bbcd4cbf14106d7dc6febd0ec9622c/chartfx-chart/src/main/java/de/gsi/chart/XYChart.java#L250-L256

The modifications checks out against the new 1 kHz test and I hope so does your use-case.

Let us know if this still causes troubles.

The JavaFX UI thread wasn't really designed for 1 kHz update rate, even less having thread safety in mind. Nevertheless, we hope that chart-fx's DataSet, plugins and rendering infrastructure are reasonably thread-safe now.

N.B. the TableViewer plugin has a known thread-safety issue for >100 Hz updates rates but this seems to be related to the underlying JavaFX issue TableCellSkin.java:97.

[ennerf]

Thanks, it looks good so far. I'll let it run overnight.

[ennerf]

Ran it for 20 hours. Seems to work. Closing.

thanks!