Using asset dependencies along with Mapped Drives?

Deadline
1

Hi all,

So bare (bear?) with me as I struggle to explain the situation. In short, how would I go about configuring asset dependencies on a job along with drive mapping? If a slave doesn’t map drives until it picks up a job, it can’t pick up a job because it can’t scan for the dependency because it hasn’t mapped the drive etc etc and round in a circle.

The long version: Edwin helped me modify our vrscene submitter script so that it only starts the job when it sees the vrscene file has shown up (we have two separate locations which are synced and this prevents the job for failing off the bat super quick because the vrscene file can’t be found). See more info here (forums.thinkboxsoftware.com/vie … 649#p74649)

We have also setup drive mapping on our render slaves. So when a slave picks up a job, it will automap the drives. However, this is where we come into issues. First, AWS does not like the asset dependency check (could not would not run - see forums.thinkboxsoftware.com/vie … 857#p74857). Easy enough fix, I made an AWS specific submission script. We are also finding out that slaves, when coming out of sleep state (woken by Pulse and WoL power management) will not pick up a job, because they don’t have the drives mapped, so it doesn’t see the dependent asset, so it won’t pick up the job to map the drives. This gets sorted once the tile assembly job starts as this causes all slaves to map the drives, but if we have a single job, no one will pick it up (unless by chance they already had the drives mapped and haven’t gone to sleep or restarted since).

So my question is, am I doing this a stupid way and how are others going about using asset dependency checks along with mapped drives?

2

As far as I know, this is absolutely a limitation with how the asset dependencies were designed.

One workaround might be to create a script dependency that maps the drive. I’m not sure if all of the dependency checks are run or if there’s an implicit short-circuit order (which puts us back at square one). I think what we need to do is move the drive map check into the Scheduler thread, but we’re not sure what kind of overhead that’s going to cause file servers…

3

Yup, I know its super tricky and I can’t imagine it is an easy fix - I guess I was mostly just curious if this is a system wide issue that also effects others (and seems like it is) or my own silliness (and maybe some of that too).

4

Nah, it’s definitely an unsupported use case. The big thing is that the asset dependencies aren’t widely used inside Deadline proper. Tie that in with drive mapping also not used everywhere and you’re in an unlikely config.