I'm trying to render out flip book frames of a simulation. When it gets to to pulling in renders from OpenGL I get the error:
“Work item ‘ropfetch_ropnet1_x’lists file ‘(output filename)’ as an expected output file, but it wasn't fond when the item cooked.”
If I render a single frame from the OpenGl node, it outputs fine. It also works fine (albeit very slowly) with Mantra and Redshift, just not with OpenGL. I've tried setting this up from scratch and get the same errors.
The output file path on your OpenGL ROP should be visc.`@wedgeindex`.`@visco`.$F4.pnginstead of visc.`@wedgeindex`.$visco.$F4.png. PDG will evaluate that parameter when create work items, in order to determine where the work items are going to write their output files for caching purposes. You need to use @visco to access that work item attribute.
Thank you, I tried that and even further simplified it to $HIP/geo2/visc.`@wedgeindex`.$F4.png The expected file names look right when I check the generated tasks but it just won't cook with OpenGL.
The messages about cache file paths are not errors – they're warnings that indicate the cache file path does not match the file reported by the job. They won't cause the work items to fail to cook and only affect caching behavior.
If there are work item failures you can use Ctrl+Middle Mouse click on the failed item to see its output log. Any work item errors will be reported there. Your file works fine for me if I change the output path. What version of Houdini are you using, and on which platform?
That error log indicates that the OpenGL ROP itself is crashing while it's cooking. There are errors regarding an invalid OpenCL device that occur before that well, however I don't know if that's related.
Since you're cooking the work item as batches, all of the frames for each wedge are running in a single Houdini process. That means that if the process crashes, any subsequent frames that haven't finish will be marked as failed.
You may want to try limiting the number of concurrent work items for that particular node to see if it resolves the crash. Right now your graph is running all 4 OpenGL batches at the same time, however I noticed those jobs are each consuming 1GB of GPU memory. You can force PDG to only run on batch at time by setting the “Single” job parm through the scheduler options, or alternatively use the “Slots per Work Item” job parameter along with the “Total Slots” parameter on the scheduler itself. More details on setting per-node scheduler parameter be found here: https://www.sidefx.com/docs/houdini/tops/schedulers.html#jobparms [www.sidefx.com]
I have the same issue with the openGL rop "failing". However, it is only failing "sort of". The frame is rendered and put to disk. The work item (on the Rop Fetch calling the OpenGL ROP) is green but the "registry" of the file doesn't seem to work. Right clicking on that work item doesn't allow you to open the resulting file. It doesn't think it is there.
The next node in line doesn't think the frame is there and errors out. This happens even with a custom scheduler (set to "single") driving the OpenGl RopFetch.
