Cancel DR render takes a long time to actually cancel

[dj_buckley]

If you start up a DR render then cancel it shortly after it starting, it takes a long time to actually cancel.  This is mostly while the DR node is still in the "Scene Parsing" stage, it would be good if Corona could somehow send a 'kill' message to the DR node on pressing cancel.

If I remote desktop into the DR node and exit the server, it cancels immediately on the workstation that started the render.

Not sure if this is specific to Corona 10 or not.

[Avi]

Hi,

Can you let me know which version of Corona and what remote software you are currently using? Can you give me some repro steps so I can try? Also is this a scene-specific issue or is happening even with a super simple scene with just some cubes and cones?

[Frood]

Since I want to tentatively +1 this anyway, respectively confirm the issue:

Just try any scene that needs some decent parsing/loading time. In many stages when starting up max or loading the scene on the slave side, communication DrServer <-> slave Max is interrupted or massively delayed. As mentioned by dj_buckley, especially in parsing stage.

Starting a DR render + cancel after a few seconds usually results in a slave first loading the scene completely until (slave) Max is able to receive commands again. If you start another render at this time, the DR system gets whacky: the slave still parses the first (canceled) DR scene, even the master VFB displays "parsing scene", but what's in fact being parsed is the old/canceled scene. You have to wait for the end of parsing (can take some ten minutes sometimes) until Max is responsive to DrServer again. Unfortunately, DrServer is somewhat out of sync at this time and logs "Slave 3ds Max ended render" and the Master VFB shows "waiting for render start" forever.

So you lost the slave for the entire render session after waiting a long time for the slave to parse the wrong scene. Maybe the responsiveness to DrServer commands while parsing can be improved on the Corona side somehow, this has always been an issue of the DR system.


Good Luck


Edit: typo

[dj_buckley]

Hi,

Can you let me know which version of Corona and what remote software you are currently using? Can you give me some repro steps so I can try? Also is this a scene-specific issue or is happening even with a super simple scene with just some cubes and cones?

Max 2024.1 Corona 10 Official Release (and all other previous Corona versions that I can remember).

Using Remote Desktop but purely because I'm lazy, I could quite easily connect a monitor and check it that way.  Remote Desktop is only ever running when I want to check something, but the problem persists even without manually checking through Remote Desktop.

Repro steps.

Open a relatively heavy scene.

Start a DR render from the workstation

While the DR server is still in 'parsing' mode, try and cancel the render from the workstation.

Then repeat the task with the DR server disabled, or while the DR server is 'rendering' as opposed to 'parsing'.

The difference is night and day.

It adds a lot of time every day just waiting for the DR server to actually cancel.  When you're running lots of test renders or you spot something last minute (which is pretty much every time I render).  It becomes very frustrating waiting all the time.

[dj_buckley]

The fact I can manually remote into the server and click the X to close out of the DR server window tells me there must be a way to automate that process with some code?  I could be wrong

But is there some way of scripting or coding "When user presses 'Cancel' on workstation then force close DR server and restart it"

[Avi]

Hi,

Thanks for your input. I actually tried to follow your repro steps but for me, canceling time is not massive. Please find the troubleshooting video below. It maybe possible, that the issue is scene specific or with very big scenes and it can take sometime to troubleshoot/Repro the issue so I would like you to start a ticket with us here: https://support.chaos.com/hc/en-us/requests/new

also if I am doing anything wrong in the video or missed any step, let me know.

[Frood]

Just try any scene that needs some decent parsing/loading time.

Open a relatively heavy scene.

Start a DR render from the workstation

While the DR server is still in 'parsing' mode, try and cancel the render from the workstation.

? It's all there. Your scene seems to be quite small so

- use a larger scene
- cancel earlier, in your video you wait until the slave is already calculating displacement
- start render again, say 10 sec. after cancelling it

No, it's not scene specific. Only important feature is, that it has to take some time to load and parse. In your video, you wait until the slave starts to render. As mentioned, it this phase, the slave Max instance is already responsive again to DrServer commands. But even then: if the scene has a large cool down time (if using some larger xref scenes for example), you run into trouble as well.


Good Luck

[dj_buckley]

I'll record a video on a heavy scene

[dj_buckley]

Ok so I've uploaded a video here.  It shows multiple problems.  The video is 9 minutes long which indicates it's a heavy scene as thats how long it takes to fire up a render.

Anyway.

The first part of the video DR is enabled.  I start the render on the workstation, cancel while it's parsing on the workstation.  It takes forever to cancel while the slave catches up.  Then when it does eventually cancel, the slave continues to parse.  I then start another render on the workstation and you can see the render has pretty much started to actually render this time on the workstation before the slave responds and stops parsing the old scene and starts to work on the new render.  Again cancelling takes a long time while the slave catches up to the latest render.

I then close the DR server and start the render again.  I cancel the render during parsing, only this time it cancels quicker than it did when DR was enabled.

I try again with DR completely disabled in the scene file, and i think it's even quicker again to cancel.

Anyway video is here https://drive.google.com/file/d/1EyafpPby1He_OoY03O8R6BizX8ohnyTB/view?usp=sharing

[Avi]

Thank you for your input again.

@dj Buckley, can you share the scene with me by either starting a ticket or you can also DM me privately? I would be interested in testing the same with your scene.

[dj_buckley]

I can try, although I always have issues archiving scenes with Forest pack.

Also - i just checked the time stamps for how long it takes to cancel on the workstation.

It's actually quicker with the slave running.  The longest one is with DR disabled completely.  I think.  But I'll let you test.  Give me a few mins to try and archive the scene

[dj_buckley]

Thank you for your input again.

@dj Buckley, can you share the scene with me by either starting a ticket or you can also DM me privately? I would be interested in testing the same with your scene.

The archive seems to have worked, I'll DM you a link shortly

Edit: I've sent the link, let me know if there are any issues with the download.

[Avi]

Hi,

I have tried your scene. Here are the steps I did.

1. Started DR rendering
2. First Message, "Transforming vertices" ~ "Saving sent scene...." ~ Slave Side - "Downloading Scene"
3. Second Message - "Parsing Scene"
4. Click Cancel - (Takes about 5-8 sec to cancel)
5. There is a short buffer time on the Slave machine about 5 - 8 sec after hitting cancel on Workstation


Once you repeat the process with Disabled DR

It simply starts rendering the scene after parsing it for a few seconds.

I am unable to reproduce the same massive time, it could be due to differences in our hardware.

Also, can you let me know which version of the forest pack you are using? and from where you are running your main file? From the local drive or from the Network drive?

[Avi]

I have logged this issue in our system to be further investigated by our devs. If there will be any update on this, we will let you know.

(Internal ID=1155055490)