Can stuck bucket syndrome finally be fixed, please?

[kosso_olli]

I hate this.



Main part of the render was done in under 1 hour.

[seandunderdale]

ooof thats a hung bucket if I ever saw one! Has that got all the reflect on backside, round corners, blurry refraction bells and whistles on it? I had to start making a closeup and a wide shot setup for headlights for this reason.

[kosso_olli]

Happening on every scene we have. It is annoying us to death!

[muoto]

are you sure it's not a LAN/Ethernet/Switch problem ? I had this as well from time to time (switch changed from 1gb to 10/100mbit)

[kosso_olli]

No. The cores shown in the image are rendering at 100%.

[kosso_olli]

And another totally random bucket hanging around for ages.

[dreidesign]

+1 for this. I think this is one is one the todo list for quite some time. I would suspect that a quick fix/workaround would something like a bucket timeout option. Like when not all cores are used and bucket is rendering for longer than 5minutes, split it up. Would this be a possible option?

[matanov]

+1 for this. I think this is one is one the todo list for quite some time. I would suspect that a quick fix/workaround would something like a bucket timeout option. Like when not all cores are used and bucket is rendering for longer than 5minutes, split it up. Would this be a possible option?

If there was a "quick fix" or some easy way - it would have been done years ago. Although, you as users can help us very much by sending such scenes to test possible solutions.

[seandunderdale]

Like when not all cores are used and bucket is rendering for longer than 5minutes, split it up. Would this be a possible option?

I wouldnt even go that far. Id just say that each bucket can have a max time. When it hits that time, just give the result of whatever noise sampling it got to. Dynamic splitting each bucket after its already started sounds a lot more involved than having a set time per bucket (which as mentioned already isnt an easy fix itself). I would rather get the render through and not clog up the farm, and maybe have some tiny sampling issues in tiny areas, (which could be correct in post anyway)

[Joelaff]

Definitely splitting is the way way to go. Returning a bucket of less than the quality set by the user is not an option for us.

All unused resources (core or other machines if DR) should be put to work on the newly split buckets. If the original core finishes the bucket first them return that, if the other buckets finish first return those. In other words, if there are ever two cores free you split into four and start handle them out.

I hate to say it, but this actually WAS solved almost 30 years ago in a product called BackBurner made by Specular International to render files from their Infini-D software. Granted they didn't have GI to contend with, but it was the same issue. They put all remaining resource on the final bucket, and whichever finished first won. Theirs was even a distributed network render application (block based like VRay DR).

This Last Bucket Syndrome (or LBS as it is more commonly known) is a continual issue for VRay (or any bucket renderer) users, and really needs to be addressed. Thanks.

[seandunderdale]

To be honest, the size of the buckets in the renders provided in this thread would take seconds to fix in post. If it means having some bucket time out tool, or nothing, Ill take the timeout and get my renders through with a 10 pixel cleanup, and save several hours per render. As has been mentioned, the splitting isnt a simple issue, and not sure yet if a timeout is either, but Id rather have something than nothing.

[Joelaff]

To be honest, the size of the buckets in the renders provided in this thread would take seconds to fix in post. If it means having some bucket time out tool, or nothing, Ill take the timeout and get my renders through with a 10 pixel cleanup, and save several hours per render. As has been mentioned, the splitting isnt a simple issue, and not sure yet if a timeout is either, but Id rather have something than nothing.

I guess you work on stills. In VFX we work on moving images, where painting such things can often be quite involved. Frames might get missed, etc.

No problem having the option to just terminate the render, but it should definitely not be the default.

[seandunderdale]

Oh god, not the default at all. For sure an option to say a timed limit like progressive, or not limit at all. But an option, if possible, would save hours on high res images.

[luisgamino2]

I have the same problem with vray gpu for sketchup, and it is very frustating because these stuck buckets can take hours to resolve the image... In vray next i never have this kind of problems... I think chaos group needs to work really on escencial problems like this, because You can have a Lot of beautiful and amazing options for your render but if You can't finish You render You can't Even show your work to a client... And this is happening in vray and vray gpu and probably in all the programs... We need a solution for this!!!

[kosso_olli]

When it hits that time, just give the result of whatever noise sampling it got to.

Sometimes I do something like that in a manual way: Cancel the render, draw a small region around the missing bucket and render that area with progressive.