Houdini/Redshift .exrs truncated?

Deadline
1

Deadline version: 10.0.16.5
Redshift: 2.6.13
Houdini: 16.5.473
Windows

Hi,

I have a current job rendering out .rs archives in Houdini 16.5.473 and then running the dependent Redshift job using the Redshift cmdline using 2.6.13. For every 100 frames possibly 3-4 .exrs are truncated/corrupt. These frames are random and the machines are random despite all attempts to see a pattern. The same machine will render another frame in the sequence perfectly and the same frame will re-render fine on the same machine. The .rs archives are all fine - they all render perfectly if you run RS cmdline locally.

The render logs for the corrupted frames are good - they say the whole RS process has happened correctly and the .exr has been written to disk. It is only when you try and open the .exr in Photoshop or Nuke that you get a bad image warning. If you look at the difference between a corrupt frame and the same frame re-rendered successfully the corrupt .exr is short of maybe 200 lines of binary data.

For some reason, after the whole rendering process has completed and with no errors in the logs the .exr is not written succesfully to disk. Has anyone seen anything like this before and have any suggestions?

Julian

2

I’m not sure. I’ve seen problems when we try and do clever things (like my hack to kill AE when plugins lock it up). If Redshift is exiting properly, I would think its output should save just fine.

I took a look at the Redshift plugin ("[repo]/plugins/Redshift/Redshift.py") and it’s one of the more basic plugins we have. There isn’t anything there that would force an exit, etc.

It may be worthwhile to bring it up with the Redshift guys, but not being able to reproduce outside of Deadline might make it pretty difficult to diagnose… Maybe something is lingering in the GPU between runs?

Has anyone else seen this kind of thing?

3

Thanks Edwin. I’ve posted on the RS forums to see if anyone else has had something similar…

4

Thanks! Here’s hoping someone over there knows what’s going on.