I'm writing a vdb sequence to disk, then loading that sequence into an Arnold Volume procedural node, then writing that to an ASS archive. Then I instance that ass archive onto points (using an Instance object and instancefile attribute)
It's all fine and good when everyone has the same frame. As soon as they need some variation, it becomes very unpredictable, and does not work. Pressing render a few times will yield different results each time, even without changing any parameters. Sometimes all the volume domains are pink. Sometimes a few volumes render. Demo file illustrates what I mean.volume-proc-broken.zip
Is there something wrong with this approach? Or is it a bug? It's severely impacting workflow currently. The other option I guess would be to write the volume data straight to .ass, but this is a big waste of disk space/resources.
EDIT: baking straight to .ass breaks it as well, so that's not really a solution.
Houdini 17.5.360 htoa 4.3.0 Arnold 5.4.0.1
If anyone runs into this in the future, either assign the shaders/shader assignments before baking the .ass archive, or use an Assign Material operator in ROPs. The usual shop_materialpath point attribute on instance points will not work, and neither will assigning a material at the top level of your instance object.
Maybe it's a bug. Doing the same workflow with a polymesh & Arnold Procedural works fine (shop_materialpath on points etc).