We have been getting the following call stack pretty consistently on some extremely complex scenes. I'm attaching the log here in case someone who speaks fluent CallStack can point me in a direction. The actual cook seems ok, and we can render on the farm, but the crash seems to happen when the SceneViewer tries to update (in Houdini GL mode).

I am starting to suspect this is due to a ton of prims we're setting to render-purpose. They don't have corresponding proxy-purpose prims though. We just don't want to deal with them in the viewport at all.

It's hard to tell, since it could be a few different factors, but it could be that putting render purpose on individual prims is the cause? In general, I'd recommend setting up something like this:


Then you dont' have all that extra bookkeeping to manage individual prims. It's probably easier on USD/Hydra as well; though I can't say for certain that's technically true or not, hehe

I will say that not having the proxy relationship on the render prims should be fine; we had many assets which never quite set that up correctly. Maybe at scale it does present issues, but I'd be more suspicious of the extreme numbers of prims that have to be traversed just for hydra to know if it it should draw them in GL or not.

Maybe try moving display purpose to fewer prims? Also are these instances

Thanks, I was thinking the same, but after a few tests with the entire render-purpose setup disabled, and still getting the crash, I'm back to more shots in the dark.

I assume this is happening when you make some change to the scene and the viewport is updating from one stage state to another? That's the most common place for stuff to go wrong, as a lot of different parts all have to work together perfectly. If you can simplify a scene and send a set of reproduceable steps there is a good chance we can fix it... When simplifying a scene, the factors that usually don't matter are mesh complexity (number of polygons), or the sheer number of instances. Usually it's whether things are instanced or not (and how - pt instancer vs native instancing), and changing the purpose of a prim has been known to be problematic (though in my experience it doesn't crash, it just doesn't update properly). Thanks!

mtucker
I assume this is happening when you make some change to the scene and the viewport is updating from one stage state to another?

Yes that's exactly where the crash seems to happen. Indeed, I fully expect it's for all the reasons you mention... it's a very complex hierarchy with gobs of instances, and about 6-7 point instancers as well. I figured you guys would want a scene but I also know I can't really simplify it without at least some idea of what to remove. Which is why I posted the crash log, on the chance something seemed familiar ("oh yeah that has happened in the past when we did this thing....").