The latter question is because it is obvious that the http/API/filecopy takes relatively long time compare to the actual data processing.
Rabit
@rabits
Hi @moozer:matrix.org , yeah for local it's quite limited, because for now supports only CPU renderer (without additional tuning on the agent side)
Are you trying to use VSE or the cycles render engine? Because for VSE it will not be so optimal currently...
Rabit
@rabits
Regarding the questions:
you can stop all started tasks and then clear all ended tasks.
the engine logic currently limited to frame-per-task. I thought to rewrite it to something more flexible via render strategies ( state-of-the-art/BlendNet#90 ) - but for now it's stopped, need to get the right mood to get into...
When I start an animation rendering, a certain number of tasks are started, and those I can can cancel. Then the next batch of tasks are started - so removing tasks does not have the effect of cancelling the render.
My solution was to go to the manager and delete the files driectly
Wrt the strategy, the "range per agent" from issue #90 would be useful for me.
I was surprised to see how slow it was to render each frame - so doing 100+ frames per task would negate the coordination overhead.
I am getting stuff like this in the logs on the agent Fra:865 Mem:554.39M (0.00M, Peak 593.96M) | Time:00:10.93 | Remaining:00:03.26 | Mem:381.22M, Peak:381.23M | Scene, View Layer | Path Tracing Sample 51/64 Fra:865 Mem:554.39M (0.00M, Peak 593.96M) | Time:00:11.18 | Remaining:00:03.05 | Mem:381.22M, Peak:381.23M | Scene, View Layer | Path Tracing Sample 52/64 Fra:865 Mem:554.39M (0.00M, Peak 593.96M) | Time:00:11.32 | Remaining:00:02.81 | Mem:381.22M, Peak:381.23M | Scene, View Layer | Path Tracing Sample 53/64
It screams that there is some processing that is being done on the agent, that is not being done when rendering "normally" on the pc.
Any suggestions? and I must read up on the details of belnder rendering
Rabit
@rabits
Yeah, I understand - the power of one render per frame was for the huge frames, which takes minutes to render on 64-core machine - with small light frames this strategy is not working unfortunately...
Until the blendnet engine will support such render strategies - there is no good answer, sorry. Regarding the steps it's doing - better to check with your blend file - they could be quite different depends on your workflow.
Oh, I see it - could you please add the google cloud sdk version you using?
awils27
@awils27
I wasn't sure if Github or here was the best place for it but i'll keep it on github
Zman350x
@Zman350x
Hello! I was wondering, do all of my Blender preferences apply to the agents, or do I have to go manage the preferences on each computer
_
Zman350x
@Zman350x
Also, all 14 are setup pog
Rabit
@rabits
Hi @Zman350x , the preferences are taken from the blend file mostly - but seriously restricted with the render script and blendnet capabilities. So if you've saved your blend file it should contain the required settings to be rendered anywhere.
Zman350x
@Zman350x
Ok! I just ran a 250 frame test across the 14 computers (each with an Intel 10700k). All I did was make the default cube spin for those frames, and the test took WAAY longer than I'd expect with this power, and now I'm left with a bunch of E:100% tasks and PENDING tasks, and no idea where the outputs are...
Zman350x
@Zman350x
Oh, found this
All of the E:100% tasks have that
Rabit
@rabits
Yeah, seems the compose did not output anything. This one need to check carefully - the logs of agents and manager, version of blender and so on. To reproduce it out on my side - could you please create a ticket with as much info as you can provide please.
Zman350x
@Zman350x
I've left the office, but I'll create one Thursday
Zman350x
@Zman350x
Ok, back at work
Lemme get all of the logs
Zman350x
@Zman350x
Reran the test to generate some logs, where are the log files saved?
Other than that manager panel which I couldn't find, I followed it
Also, I hooked everything up via ethernet instead of Wifi and reran (only 3 agents, none on client/manager PC). It got the point where all tasks said E:100.00%, none left on pending
It formatted part of the error message into an emoji lol
EM-3D
@EM-3D
hi i´m testing blendnet v.03 with blender3..0 & BoneMaster 3.1 build from graphicall, the addon can create and manage VM in Gcloud but i get this "error" every time
and the render task button gets never enable, also in the addon preferences the blender dist dropdown its empty i try to put a direct link to the blender zip installation but i dont know if that works
EM-3D
@EM-3D
sorry i tested with blender LTS 3.0 official for linux too
Rabit
@rabits
Thank you @Zman350x , will check the issue
Rabit
@rabits
Hi @EM-3D , the usual reason for that is that the Manager instance was not started up properly, because exactly the Manager node creates this certificate in the bucket for Addon to download it.
Try to follow the Wiki FirstSteps to setup your google cloud the first time.