Network Rendering gets stuck at 99%

Started by DonChunior, September 25, 2012, 12:19:57 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

DonChunior

Sometimes Network Rendering gets stuck at 99%.
At this time all tiles have been rendered and the CPU usage of all slaves is 0%.
Then it takes up to 10 minutes, till the picture is completely finished (100%).  >:(

Ruckus

If you image finishes 10 miutes late, as you stated, then you need to look very closely - you will find one (or more) tiles that have not completed.  The most likely cause is that the KS real time display has not been paused, or someone is running a local render, on one of your slave machines.

That situation is frustrating, but can be avaioded by careful coordination among your team.  (I am assuming your slave machines are also general user machines)

The situation gets worse when one slave has not returned it's tile(s), and the CPUs on that slave are doing nothing.  The network rendering will never finish when this happens until that particular slave machine is rebooted, or it's slave process is stopped/killed.

DonChunior

Nobody else is doing local renderings on the slave machines. They (and the master) are placed in a server room and only get used for Network Rendering.

Ruckus

Quote from: DonChunior on September 25, 2012, 05:54:55 AM
They (and the master) are placed in a server room and only get used for Network Rendering.
Lucky you!

My next guess would be just a really complicated tile.  For us it is taillamps, where the rays pass through several layers of clear and red plastic, a bulb, and bounce back off the chrome reflectors.  If you find another reason, don't forget to let us know.

KS is claiming some sort of fix for stuck slaves will be available in the Network Rendering release that goes along with 3.3.33 (released yesterday).  I'm anxious to see how that will work, but I may not be able to test it for few a few days.

Claus Jensen

Ruckus is right. Your scene must have some complicated areas. The slaves will wait for the whole job to finish, which is why you can see some slave doing nothing and some slaves being busy. We are improving this situation in future versions by subdividing the job even further.

DonChunior

But the CPU usage of all slaves is 0%, nevertheless it takes another 5 to 10 minutes, till the picture is really finished.