Distributed render slaves rendering mapless buckets

[Moko3D]

Vray 3.2 3dsmax 2016 lots of distributed render problems when hosted by various of our computers. Here are some of the situations with distributed renders having the crossword puzzle effect where DR servers are not using any material maps properly.

Computer1 hosts DR job from a max session. Computers 2-10 render their buckets fine with maps.

Computer1 sends DR job to backburner (include maps for bb job). Computer4 BB server picks up job and starts hosting the DR job. Now only computer 4 renders it's own buckets with maps and the other computers render DR buckets with no maps in the materials. They worked fine with computer1 as a host....

Computer2 opens the job in a max session rather than a backburner session. Transfer assets is definitely on, but when hosting none of the other DR computers will render with maps only the host computer2 renders with maps. So computer2 can see the maps, computer2 can render the maps, but it won't send out maps despite transfer assets option being enabled?

[tsvetan.geshev]

Hello,

Please check that Save servers in the scene is checked and if it's not turn it on.

Let us know about the result.

[Moko3D]

Save servers is definitely on. The various DR servers are producing buckets. Just those buckets often have no bitmaps used in the materials, so they are useless buckets with wrong texturing being returned.

[Svetlozar Draganov]

There might be a bug with Transfer Missing Assets option.
Would you please log on remotely on Computer4 then open 3DSmax and make sure that Transfer Missing Assets option is activated. If it is not please activate it, close 3DSMax, run the same test and let us know if there is any improvement.

Another way to activate Transfer Missing Assets is to open the following file and make sure that the following command exists:
C:\Users\USER\AppData\Local\Autodesk\3dsMax\2016 - 64bit\ENU\en-US\plugcfg\vray_dr.cfg

transfer_missing_assets 1

Please note that above changes should be applied on Computer4 the one that receives BB job from Computer1 and distributes it to other machines via V-Ray DR.

[Moko3D]

Most of the render boxes e.g. computer 4 had no vray_dr.cfg file at all. What I did the other day was turn save servers in scene off - then go to every machine and manually add a vray_dr.cfg file at that location and edit every individual file so that localhost was on but that individual machine was excluded from it's own list of servers. That way a DR job can be sent out with BB to any machine and it won't try to double render it's own job (both as localhost and as a DR server which can be slow/crashy/memory issues). So it looks like:

design1 1 20204
design2 1 20204
design3 1 20204
design4 0 20204
design5 1 20204
design6 1 20204
slave1 0 20204
slave2 0 20204
slave3 1 20204
slave4 1 20204
slave5 1 20204
slave6 1 20204
slave7 0 20204
slave8 1 20204
restart_slaves 1
list_in_scene 0
max_servers 0
use_local_machine 1
transfer_missing_assets 1
use_cached_assets 1
cache_limit_type 2
cache_limit 30.000000


for an example config from one machine atm, obviously different machines each need their own self excluded. So that change is making sure transfer assets is on and have not seen mapless buckets problems since (so far... but we've been involved in other things than vray dr mostly). Obviously it's a bit of a pig to maintain with manual edits everywhere anytime you need to add a new machine or change the setup etc. But it does seem to function as a work around for the mapless buckets so far.

It would be really fantastic if there was some sort of DR (& RT DR) manager or network updater utility that options could be set for everywhere more easily, where machines that have crashed their DR or RT_DR status could be spotted, and so on.

[Svetlozar Draganov]

It seems that Transfer Missing Assets option is not taken into account when the scene is sent to BB for rendering - this is a known issue according to our system and it will be fixed for the future releases.
As for now the easiest way to ensure that Transfer Assets is enabled is to log on individually on the machine that receives BB-job and to change its vray_dr.cfg settings.
Once the bug is fixed all DR settings like Transfer Assets or Use Local Machine will be properly transferred and this manual configuration won't be needed.

[Moko3D]

OK that's good. However has the fix implementation considered it's own issues... for example with localhost on/off and double rendering the same job.

For example you send out a DR job. It gets picked up by let's say computer 6 which then tries to both render it locally, and also render it as a DR session and hand out all the assets, slowing everything down.

You're possibly left with manually configuring every job either turning on or off DR server on machines hosting jobs, manually setting which machine do the job and adjusting the list of DR servers inside the job for each job, or constantly turning on and off use localhost but that means also using double memory on the host (once for hosting the scene to hand it out for everyone to DR, once again on the same host to actually do it's own DR session) if you take the turn off localhost approach.

Plus it's less than optimal for workflow if you have to constantly go turn localhost on and off depending on whether you render a DR test frame (or region) locally on your machine.

There ideally should be a way to send out BackBurner jobs that can just go to any of the render machines, where any machine can host the job and also render locally with localhost on, but aswell knows how to exclude itself from double rendering that particular job by excluding for that job itself from the list of DR servers.

Double rendering a job can be slow, a memory hog, inefficient and prone to crashes, and can really slow down communicating with everyone else doing their DR pieces - making the whole job much slower.


If the whole process could be a bit more fire and forget rather than manually config that would be great.

[Svetlozar Draganov]

That request is directed more into render management software ranter than V-Ray.
I could make another feature request about excluding local machine from rendering so the launched VRaySpawner will not accept jobs coming from the local machine but it will accept jobs from other machines so the same double-rendering could happen again.

[tashko.zashev]

Some good news about this issue. Our developers managed to fix the missing maps with BB when "Transfer missing maps option" is enabled.
Most probably the fix will be part of the next SP release.

[gregvfx]

Hi,

I have a bitmap on a server, visible to all the computers on the network. DR buckets aren't using it, only Host is.
If I start an empty scene and try rendering a box with this exact same bitmap it works.
If I reset my scene and import from the first one and I have the same missing asset message...

What to do?
How can I isolate the bug?

[Morbid Angel]

could you post this image + sample scene where it does not work?