> Some of our render nodes seem to sporadically not be able to write images
> to our file server. The weird thing is we see no error messages from Lightwave;
> it says 'Frame completed.', but no image is written, and no error message is printed.
>
> We see this in rush, and sometimes we can *even* replicate from DOS.
> Seems to be machine specific, but its weird, sometimes a machine will
> render a job correctly, but a different job it won't.
>
> We've ruled out permissions issues, not sure what could cause this.
What you might be seeing is a combo of problems leading to
LW not being able to load the "Image Saver" plug-ins it needs
for writing images.
Apparently in LW, all the image saving routines ("Image Savers")
are plug-ins, one plug-in for each image file format (png, tga..)
The LW 'config' files tell it where to find the plugins, e.g.
TGA: c:\Program Files\NewTek\LightWave 3D 9.2\Plugins\input-output\tga.p
TIFF: c:\Program Files\NewTek\LightWave 3D 9.2\Plugins\input-output\tiff_logluv.p
[..]
What can happen is LW can sometimes fall back to loading old or
stale CFG files, causing it to not load the image saver plug-ins,
which cause the render to skip writing the image.
The annoying thing is when LW can't load the image saver plug-in,
it does NOT print an error.
I studied this once carefully, because it was driving me crazy;
apparently LW does the following:
1) Loads the geometry, pre-processes everything to prep rendering a scene
2) Loads the "Image Saver" plug-in (png, tiff, tga, etc)
3) Prints the "Writing RGB image to .." message to stdout
4) Writes the image
5) Prints the "Frame completed" message to stdout
If step #2 fails (can't load the plug-in) then steps #3 and #4
are skipped. This is why you don't see any error message at all.
AFAIK, there is no 'verbose mode' available in LW to make it
print plug-ins it couldn't load.. so it's /very/ hard to debug this.
The problem gets more complicated when the user doesn't specify the
'Config Directory:'; then LW then falls back to looking in the
local "Documents and Settings" directory on each render node for
CFG files, where it can find old or outdated config files
left over from previous renders.
So for instance, if Rushd is running as 'render', and the job
didn't specify a "Config Directory", then LW will load them from:
c:\documents and settings\render\*.cfg
..on each render node.
If it finds 'left over' LW config files, it will on boxes that
have stale contents.
This is why some jobs work (jobs that specify a config directory),
and others fail (ones that don't, and fall back to the stale local
files on each node)
The key, apparently, is the absence of the 'Writing RGB image to..'
message in the logs.. when that message isn't printed, chances are
the plug-in wasn't loaded.
I suppose the submit-lightwave script could check for this missing
message, but I'm not sure that message is 100% reliable for all
the image formats; each plug-in can behave differently.
|