rrC4D - Error rendering document: Not enough memory.

[RandallCurtis]

Ever since our studio went full Cinema 4D / Redshift just about every job (simple or complex) will intermittently abort with the error:

rrC4D - ERR: error rendering document: Not enough memory.  
rrC4D - ERR: Error reported, aborting render script

None of our clients have less than 32GB of RAM, and this error happens even on our new Puget render node which has 128GB. Client table also shows a lot of free memory even on the machines with 4 GPUs/4 Threads going at once. This results in a lot of jobs disabling clients and forcing us to manually re-enable them after which the queue will *eventually* finish.

Is this RR or C4D related? I was able to figure out that the error stems from the kso_c4d.pyp script (line 1233) during the call to c4d.documents.RenderDocument(). The SDK documentation doesn't really provide any additional information as to why the out of memory result would occur though. Could it be the KSO script or our config causing the issue, or is it some sort of bug in Cinema that puts us at the mercy of Maxon?

Here's an example debug export which is just one job out of just about all of them which exhibit the same behavior.

210826_1530_debugInfo_Xvivo

[RR support, Acampora]

Hi Randall,

you are correct, line 1233 of our plugin is just the render command, so any failure in rendering will report as an error in line 1233 of kso_c4d.pyp

You can try to render without the kso_script changing the renderer to "old_redshift" in rrSubmitter. Rendering should be just the same, but depending on the implementation of the plugin, we cannot exclude that there can be differences when rendering from the python API.

I have noticed that the second half of the job tasks fail more frequently, is that a widespread behaviour? Would restarting the clients allow them to complete the job? Or would the same client be able to render the first part of a different job after it fails the second part of a job? Or is the second half of that specific job heavier to render (more volumes, more action, more objects in camera, etc.)?

Based on these answers, we can make a better idea of what is going wrong. Last, I cannot find significant comsumption of RAM in the logs, but the memory to which the error refers might be the VRAM. If the clients manage to render a failed task after a reboot, it might be that the VRAM is not being emptied properly.

Kind regards,

Paolo

[RandallCurtis]

Paolo,

I'll do some testing with KSO turned off, I know for Maya I had similar issues so I would usually render with KSO disabled.

Regarding the failure rate it occurs quite sporadically with both simple and complex scenes, on different machines, and at different points in the frame sequence. Some jobs start and may fail immediately, halfway through, or only fail on a few non-specific clients. Likewise, submitting the same scene multiple times results in different failure rates on different clients. Restarting a client seems to have the same effect as simply re-enabling it, whereas it may fail immediately, not at all, or anywhere in between. With a bit of luck, some jobs only fail on 1 or 2 clients and can finish without any intervention, albeit not as fast as it should have.

With VRAM my experience is that Redshift throws its own error when running low. Usually that is something we can diagnose during scene creation as a result of heavy tesselation or a very high rendering resolution. To me it seems like the Royal Render error occurs during scene/document loading even before Redshift has started. I could be wrong with that assumption as I'm not familiar with the way the KSO script works.

We'll keep testing a few different things including Cinema S24 and disabling the few plugins that we use. I also submitted a support ticket to Maxon with the hope that they can assist as well.