QOpenGLFunctions in Docker (headless OpenGL)

[lukicdarkoo]

We have a QT-based application and we would like to run it in Docker without xvfb and without showing GUI. We want to create a headless mode, but if possible, we want to avoid dependency on xvfb.

Approach #1: First, we tried simply by setting the QT_QPA_PLATFORM environment variable to offscreen. Although this hides the GUI on my PC it fails to create QOpenGLFunctions in Docker.

QOpenGLContext *context = new QOpenGLContext();
QOpenGLFunctions *gl = context->functions(); // Shows warning: "QOpenGLFunctions created with non-current context"

Approach #2: Then, we tried to initialize OpenGL context (without QT) according to the following two articles:

• https://developer.nvidia.com/blog/linking-opengl-server-side-rendering/
• https://developer.nvidia.com/blog/egl-eye-opengl-visualization-without-x-server/

And it works! It works in Docker and on my machine. We are able to use gl* functions without the display in Docker. Here are some details:

GL_VENDOR=Mesa/X.org
GL_VERSION=3.3 (Compatibility Profile) Mesa 20.2.6
GL_RENDERER=llvmpipe (LLVM 11.0.0, 256 bits)
GL_SHADING_LANGUAGE_VERSION=4.50

Approach #3: Therefore, we tried to use our working EGL context to create QOpenGLContext:

QOpenGLContext *context = new QOpenGLContext;
context->setNativeHandle(QVariant::fromValue(QEGLNativeContext(ourEglContext, ourEglDisplay)));
QOpenGLFunctions *gl = context->functions(); // Shows warning: "QOpenGLFunctions created with non-current context"

But again, although it works fine on my PC I get the same warning in Docker.

Do you have any suggestions? Does it make sense that we create a custom platform similar to the offscreen, or there is a simpler solution?

[mrjj]

Hi
It's outside my normal zone but Im wondering
if approach 3 is just missing
https://doc.qt.io/qt-5/qopenglcontext.html#makeCurrent
with a
https://doc.qt.io/qt-5/qoffscreensurface.html

Again, its not something i have much experience with but i used Qt to render
openGL mesh to images as a command line tool. Not exactly headless nor in docker
but no harm in mentioning it.

[lukicdarkoo]

Hi, thank you for the suggestion. You mean something like this:

QOpenGLContext *context = new QOpenGLContext;
context->setNativeHandle(QVariant::fromValue(QEGLNativeContext(eglCtx, eglDpy)));

QOffscreenSurface *surface = new QOffscreenSurface();
surface->setFormat(context->format());
context->makeCurrent(surface);

context->create();
QOpenGLFunctions *gl = context->functions(); // <-- Shows the same warning

But I am getting the same warning. Maybe I am missing something?

[mrjj]

@lukicdarkoo
Yes exactly.
But can you check if makeCurrent returns true or false ?

[lukicdarkoo]

@mrjj Oh, it returns false. I will need some time to figure out why it returns false.

[lukicdarkoo]

@mrjj I was calling the makeCurrent function before create. Here is the fixed version:

QGuiApplication a(argc, argv);

QOpenGLContext context;
context.create();
assert(context.isValid());  // <----- fails here

QOffscreenSurface surface;
surface.create();
assert(surface.isValid());

context.makeCurrent(&surface);

QOpenGLFunctions *gl = context.functions();
qDebug() << "Version: " << QString::fromLatin1((const char *)gl->glGetString(GL_VERSION));
qDebug() << "Renderer: " << QString::fromLatin1((const char *)gl->glGetString(GL_RENDERER));

But it also fails to create the OpenGL context in Docker. With xvfb-run and on my PC it works fine though.

[SGaist]

Hi and welcome to devnet,

Did you try to use the eglfs backend ? Since the Nvdia documentation mentions egl as system to use so you do not need an X server running.