(rib) Batch Render produces incomplete outputs

[Yantor]

We're running Qube on a Windows 7 network hosted by a Windows Server 2008 supervisor, using Maya 2009, Renderman For Maya 3.0 and Renderman Pro-Server 14.4 for our renderers. Currently we are trying to generate RiB files to render using PRMan and we're encountering a problem. I've traced the problem to that the Maya BatchRender (rib) jobtype is not creating all the appropriate files when it generates a rib. Specifically, in the data/{frame} folder, it often fails to generate the .ptc files neccessary to render sub-surface scattering. This leads the Renderman jobtype to fail when we submit the RiBs to it.

EDIT: Two warnings that show up in the log that may or may not shed light on this matter-

Warning: (11/28 22:02) rfm Warning: GetCmdPassCmdList ${RMANTREE}/bin/ptfilter: no outfiles found, inferring from infiles
Warning: (11/28 22:02) rfm Warning: GetCmdPassCmdList ${RMANTREE}/bin/ptfilter: bad cmd pass detected:0-> 0

I re-rendered a troublesome shot with a log output, and I see lines that would suggest it is processing the appropriate data but no point cloud is written.

My question is thus- is this a permissions issue, a Renderman issue, or a Qube issue, and how do we fix it?

[jburk]

These errors are from the logs from the ribgen pass, correct?

When you say "often fails", does this mean that it occasionally succeeds?  If so, are any of the output path settings in the successful render's scene files different from the ones that fail?

Qube simply calls maya's 'render -rendererer "rib"' command to do the ribgen.  If things are outputting as expected, the first place to start would be the path settings in the RFM options.

[Yantor]

By "often fails", I mean occassionally in the same job it will successfully write a point cloud, which apparently is odd- I've just found out from the Renderman forums that point clouds are generated during the actual render, using PRMan. Which in that case means that for some reason, Qube or Renderman (or a combination thereof) can't consistently write files to our server when rendering. I've double checked some of the jobs we've tried that have failed and yes, the frames that manage to render do have output that suggests the various SSS processes run. Those that don't return this error-

ptfilter ssdiffusion WARNING: unable to open input file renderman/lit_sc02_sh05/data/0001/sc02_sh05_test_cam_sc02_sh0Shape5_SSRender_chrR_Penguin_SG_PN_LHand.001.ptc
ptfilter ssdiffusion ERROR: unable to open any input file. Aborting.
R06004 {SEVERE}  Error 1 during RiSystem call '"C:\Program Files\Pixar\RenderManProServer-14.4\/bin/ptfilter"  -filter "ssdiffusion" -partial 1 -progress 2 -threads 8 "renderman/lit_sc02_sh05/data/0001/sc02_sh05_test_cam_sc02_sh0Shape5_SSRender_chrR_Penguin_SG_PN_LHand.001.ptc" "renderman/lit_sc02_sh05/data/0001/sc02_sh05_test_cam_sc02_sh0Shape5_SSOrganize_chrR_Penguin_SG_PN_LHand.001_"'
INFO: exitcode=512

Which obviously means that it didn't write the file and can't find it...what I'd like to know is why. Ideas?

[jburk]

Do the ribs for the frames that produce point cloud files differ significantly from the ribs for frames that have no .ptc's?

I'm wondering if the instructions telling the render pass to run ptfilter are missing from the ribs, or if the prman executable is somehow failing to run them.

[Yantor]

As far as I can read RIB archives, no, they don't look different. The shorter file calls three ptfilter commands to a .ptc in relative project space (/renderman instead of //server/share/project/renderman). The longer file is an incomprehensible wall of RiB language that I can only hope does what it's supposed to (I don't know the Renderman shading language).

As of right now, I am re-running the scene through the farm based on some advice from the Renderman forum, and so far it hasn't failed. Also, another scene is working without any problems, yet, I'll know in about half an hour if it worked. But the stdout shows sub-surface scattering is being calculated, so it may be just a problem with the single scene.

EDIT: A few frames just finished without errors, so I think this might be an in-scene problem with so many "defaultRenderLayers" existing because of our reference-based pipeline.

[jburk]

Thanks for trying the Renderman forums. 

We've tried to design Qube so that it knows as little as possible about the internals of the jobs it runs.  That way, if the application works properly, Qube doesn't get in the way.

Conversely, when things go wrong, it's often in the application, not Qube. 

That being said, we're more than glad to troubleshoot with you to catch the times when it ~is~ Qube that's not behaving correctly.