DR behaviour different to how i remember...?

[super gnu]

Ive not used DR for many years, having only one machine. I now have 2 so I was testing it out... however i remember that in the past (many years ago maybe vray 2/3 era) DR was a bit more "robust" in the sense that, if you set a render going with DR enabled, then, after, started a machine with one of the listed spawners on it, it would pick up the job, even mid-render. This was very useful if a render was taking too long, you could add machines.

I also remember that if a slave got stuck for whatever reason, it could be restarted and it would jump back on the render.

Now it seems Vray only checks once for slave availability, at the start of the render, and if there are any issues, that slave will not contribute at all until the next render.

(for example rendering again straight after a previous render, often the spawned copy of max on the slave hasn't finished restarting fully by the time the next render starts, so is not used at all, despite being ready 15 seconds later...)

its entirely possible ive invented the whole thing in my head. but if not, any way to get the old behaviour back ?

[vladimir_krastev]

Hi, thanks for posting and congrats about the second machine.

it would pick up the job, even mid-render

It does pick up jobs mid-render now as well

if a slave got stuck for whatever reason, it could be restarted and it would jump back on the render

It does jump back on a render with the modern version

for example rendering again straight after a previous render, often the spawned copy of max on the slave hasn't finished restarting fully by the time the next render starts, so is not used at all, despite being ready 15 seconds later...

This has happened previously, happens now as well. It is not that V-Ray checks for servers only once, the issue is usually somewhere else.

One big difference is that the Light Cache is now calculated on each render node and it is not distributed across the network. When exactly was this changed I don't remember. Probably in the late version 3 or with next.
The weak point with using distributed rendering is very often the network.

[super gnu]

That is very odd. I have seen a dr render node that was unavailable at the start of an image but subsequently available, sit there not doing anything (no network activity or memory footprint change) for over 5 minutes while the main machine finished the render. Nothing in the main log apart from at start of render when it says "spawner not available" or something.

This is on a 10Gb wired direct connection between 2 machines. .. if the spawner is ready and waiting when my render starts, it picks up fine.

[vladimir_krastev]

I have seen a dr render node that was unavailable at the start of an image but subsequently available ....

This is one downside of distributed rendering. There is some delay between the main PC starts rendering and the time a slave starts. This is why it is more effective to render the whole frame on a single machine but DR is fast once it gets going and in some cases this is more important than efficiency.
Make sure you have a well structured network (not only quick). Your assets should be accessible to all slaves and the spawners are online before you start the job. A 3ds Max instance is opened and used for the rendering. If this has issues opening on the slave, the slave won't join or just hang.
In the past a 5 minutes delay was not the same as now. The render times in the past were much longer due to the weaker hardware that we had. The delay is the same then and now since it does not depend on the quicker CPU but more on the whole process to unfold (opening scenes, sending assets, etc.). Also the assets are getting bigger and bigger as well as the scene sizes. Another very important point is the memory. If you do not have enough memory on the slave and you do on the main PC then the slave will just swap memory to the hard drive and slow down extremely while having no errors or warnings since technically it renders fine.
We are planning upgrades to the distributed rendering and will let you know when they are ready.