luxconsole suggestions

[SATtva]

Isn't that what abel suggested originally?

Yes, without the "abandoned" timeout, which will be hard to keep on large farms anyway.

[Abel]

Is this anything but a remote corner case?

Maybe but it's the situation I'm working with. At work there are two computers with Blender+LuxRender installed, plus a couple that just act as render slaves. At the moment I'm using four computers at most (2 masters + 2 slaves) so the risk of messing things up isn't too big, but it would still be nice to get some feedback when trying to add slaves that are busy.

LordCrc's "force" proposal sounds like a nice improvement to me.

[Lord Crc]

Isn't that what abel suggested originally?

With the addition that the slave must be susceptible to forcing, via the command line flag. So a combination

edit: We could then have a timeout parameter on the slave, which makes the slave accept the new session without the force parameter.

[SATtva]

...Which may bring another corner case: for example you're rendering a huge image or a normal one but on a large number of slaves, thus the walkover the whole farm could take a lot more than the timeout, and some slaves could become susceptible to the hijack. I think setting the timeout right would become a PITA in comparison with a simple command line parameter.

[mtoivo]

Just a quick side note about the hijacking scenario. I don't know if you've followed my reports on the system i've been building lately, but it takes care of this problem. I've put together a little database of all the luxrender slaves. The database keeps track of ip-changes and the overall status of the slaves (online or not). When a rendering job starts somewhere, the master sends out query to the database to reserve some slaves. The (max) amount of slaves per master is predefined. After the reservation, those slaves are not available to anyone else. So no accidental hijack is possible, even if it takes a while for the master to communicate with each slave the first time (and so reserve the slave in the luxrender-way). I have dealt with the hanging slaves-scenario with a cruel method that schedules restarts to all the slaves assigned to the master before starting the actual render job.

But after quick look to the suggestions, I think it would be enough if slaves could be started with extra commandline parameter so that they're not "bound" to specific session / master, but instead drop everything they're doing when receiving new job.

[jeanphi]

Hi,

The slave could log information about the takeover and in this case store the flm locally so that it can be salvaged.

Jeanphi

[crosley09]

Hey guys

I am also trying to use the networking feature in luxrender. In my case I am using my macbook pro as a master and amazon ec2 instances as my slaves. (I have configured an updated AMI Ubuntu 11.10 and lux 0.9, and my laptop is using Luxrender 0.9 3347 however, since 0.9 does not seem to have any added networking features i am considering moving back to 0.8 for stability )

I have been having the same problems as mtoivo but additionally, when i attempt to connect with more than a couple instances the luxrender GUI will stop responding, i believe it continues to render and write in the background but i no longer have the ability to add additional slaves. i'm pretty sure this occurs because my computer is trying to render as well as send jobs to the slaves and merge incoming data.

A solution i propose would be to add an option to luxrender so that it does not have to render when in 'master' mode. With the virtually unlimited scalability available with the Amazon Cloud it seems some efficiency could be gained by having the master solely manage the queue, collect and merge the data from the slaves, and write the flm.

[mtoivo]

Hey guys

I am also trying to use the networking feature in luxrender. In my case I am using my macbook pro as a master and amazon ec2 instances as my slaves.

Without going too much into detail of your problem, may I ask why do you use your own laptop at all? Why not just upload everything to amazon and start master from there? I think it would make more sense, since transferring flms is quite heavy on the network and would be easier if master is in the same "local" network as the slaves. What do you think?

[zeealpal]

I have noticed that when using online render farms, if you have multiple slaves (say 5) and the film transfer interval is to short, I found the interface to become unresponsive/unable to add more network slaves while a flm transfer is happening. Given that a flm might be ~120MB, and 5 slaves = 600MB, if the final slave does not transfer before the first slave is ready again, you have a backlog of flm's needing to be sent.

[Abel]

This may be obvious but better safe than sorry: if the problem is what zeealpal suspects, the workaround would be to just set the network update interval to something much bigger.