Hello
I will check the logs soon.
The problem is not noticeable with small projects (both linked and not, using f.e. 7 servers we have up and running now) but i'd say 75%-90% of large projects (i mean like 4+ or more GB files etc) fails with sync or render. If somehow we have synced 4 servers we have noticed 3 ways to "finish" the render.
1st - after servers jobs are done, rendered image returns to host and is getting "glued" into one piece
2nd - renders are not returning (all of them, not piece or two) but they are rendered in dir /cr/server/UUID/file.exr
3rd - everything crashes including headless instances servers
We have found a way to render linked projects, network usage of host ethernet card is peaking at 900Mb/s+, on slaves too due to linked files. As i saw, communication between host and slaves goes through TCP and is using jumbo frames, am i right? I have never found any info about that. Can jumbo frame and/or MTU config be the problem in our case?
I’m looking forward to your opinion.
Kam