Page 1 of 1

Memory leak - alembic sequence

Posted: Mon Aug 22, 2016 6:02 pm
by QuinceHungaryKft
Hey guys,

I've been dealing with this issue since a long while and I've hoped it will be fixed in one of the updates. However it seems there are no more updates for v2.x. But this is a serious issue for me and my colleagues.

When we load a heavy alembic sequence - for example a water simulation mesh - set its mode to reshapeable proxy and turn the render update mode to movable proxy. Either if IPR or Batch render is started the first frame loads fine. The next time the timeline changes the memory starts to fill in and after a short while maya crashes. There are moments when in IPR mode the scene updates without problem for a while, but the crash eventually is unescapable.

Batch rendering is practically impossible in an efficient way due to this, as the only stable way to do is to turn the animation mode into full. Which - due to our large scenes of static/scatter/movable objects- is giving a heavy load time for each frame.

I've seen someone facing a similar issue in the houdini octane plugin where it was fixed.
viewtopic.php?f=103&t=48500

please take look and let me know what and when to expect.
Many thanks!

Re: Memory leak - alembic sequence

Posted: Mon Aug 22, 2016 6:24 pm
by calus
As a workaround, I would use Octane Standalone for such a case.

Re: Memory leak - alembic sequence

Posted: Wed Aug 24, 2016 8:19 am
by QuinceHungaryKft
calus wrote:As a workaround, I would use Octane Standalone for such a case.
Hey calus, thanks for replying!

I've tried this - not the first time - and as before it didn't work out again. I used export as octane animated scene:
The scene contains 2 alembic sequences and a simple transform animation. The abcs come from different sources: one realflow, and one maya cloth simulation. the scene is in 25fps and this one is never more than 2mil polygons. It is a very long scene however: 3020 frames.

It doesn't freeze which is a great sign, but the the following issues occur:
The alembic mesh (the realflow alembic, not the cloth) looses material information after some frames and after that it stays on default material even when scrubbing back to the first frame.
The timing becomes very strange, almost like the whole scene get stretched 10 times. I'm not familiar with the Octane standalone, so it might just be my misreading.

I will welcome any idea that could help me with these.
Many thanks!

Re: Memory leak - alembic sequence

Posted: Wed Aug 24, 2016 9:31 am
by calus
QuinceHungaryKft wrote: The alembic mesh (the realflow alembic, not the cloth) looses material information after some frames and after that it stays on default material even when scrubbing back to the first frame.
I remember having this issue also, I think it's because the realflow ABC use HDF5 instead of Ogawa.
HDF5 can be unstable and give memory issue (not octane specific),
So converting the HDF5 realflow ABC to Ogawa ABC using maya cache exporter, may solve your problem.
(Also you could use a utility coming with alembic build called abcconvert,
I think it's included in houdini trial version, or can be found online)

Also did you use the latest standalone version for this test ? I remember there was a fix for Alembic in one of the V3 version.

Maybe another workaround would be to export to standalone by chunk of 100 frames, did you try this ?

If nothing works you could ask for support in the standalone section or contacting the support directly. It's more likely you get support for Standalone (and sometimes even quick fix) than for the Maya V2 plugin.

Re: Memory leak - alembic sequence

Posted: Mon Aug 29, 2016 8:55 am
by QuinceHungaryKft
calus wrote:
QuinceHungaryKft wrote: The alembic mesh (the realflow alembic, not the cloth) looses material information after some frames and after that it stays on default material even when scrubbing back to the first frame.
I remember having this issue also, I think it's because the realflow ABC use HDF5 instead of Ogawa.
HDF5 can be unstable and give memory issue (not octane specific),
So converting the HDF5 realflow ABC to Ogawa ABC using maya cache exporter, may solve your problem.
(Also you could use a utility coming with alembic build called abcconvert,
I think it's included in houdini trial version, or can be found online)

Also did you use the latest standalone version for this test ? I remember there was a fix for Alembic in one of the V3 version.

Maybe another workaround would be to export to standalone by chunk of 100 frames, did you try this ?

If nothing works you could ask for support in the standalone section or contacting the support directly. It's more likely you get support for Standalone (and sometimes even quick fix) than for the Maya V2 plugin.
Hi calus,

I just want to say thank you for your support. I eventually managed to make it work in octane standalone. I was not able to export the full scene from maya to octane, but by exporting the scene setup and importing the alembics inside octane standalone works perfectly.

Cheers!

Re: Memory leak - alembic sequence

Posted: Tue Nov 15, 2016 10:37 am
by QuinceHungaryKft
This is a serious issue in our workflow. Please let me know if you need sample files to recreate.
Thank you!

Re: Memory leak - alembic sequence

Posted: Wed Nov 16, 2016 5:25 am
by calus
QuinceHungaryKft wrote:This is a serious issue in our workflow. Please let me know if you need sample files to recreate.
Thank you!
I had a thought about this,
for some reason the Octane exporter is also really slow to export Alembic loaded in Maya and seems buggy,
I suppose the rendering code share the same issue.
Maybe this is a Maya issue, ideally we should have the option to render loaded Alembic as a proxy,
I mean I think we need a way to tell Octane plugin to not use the geometry loaded in Maya but directly read the ABC file, behaving like in Standalone.



@QuinceHungaryKft,
developers always need sample scene,
You can send privately the link of your scene+abc to memberlist.php?mode=viewprofile&u=253344,
Bikram will take care that the issue is processed.

Re: Memory leak - alembic sequence

Posted: Tue Jan 10, 2017 10:25 am
by calus
I added a feature request which could solve this issue I think.
viewtopic.php?f=111&t=58521