AWS Thinkbox Discussion Forums

Working around application memory leaks in Advanced plugins?

I’m attempting to render with modo under Deadline. modo has a number of memory management issues that cause (even in the absence of Deadline use) memory to be accumulated as the render progresses from frame to frame. Eventually, the machine runs out of memory and things go wrong - in my current case, this only takes 40 frames to occur and since the frames render quickly, this means I’m having to manually reset things every hour or two.

I thought using Deadline, with 1 frame per task, would work, but it seems that the modo_cl process is never actually killed, but continually used just with new commands. As such, the memory accumulation isn’t avoided.

I’ve hunted through options, etc. and didn’t find a way to force a new process to be started with each render task - is this already possible? If not, is there a way to modify the plugin to make it possible?

It is definitely possible! It would just need a little tweak to the submitter to get it working every time.

We have an option on the jobs that will restart the plugin between every task. I use it mostly for when I’m developing a new plugin or testing something, but it should work great here:

I’d make your tasks much larger though to save the overhead of reloading Modo too often.

It’s controlled by this job property:

ForceReloadPlugin=true

If you modify the submitter to throw that in, you won’t have to manually set it every time. To quote our own docs:

Ahah. Got that working, including the submitter customization. It took some digging around to find the submission script for modification, but once I found it, it was straightforward enough.

Thanks very much!

Welcome!

Privacy | Site terms | Cookie preferences