Hi,
We have an odd behaviour happening with power management in Deadline 8.0.6.5
Happening only on the Windows 7 side of our render farm, pulse obeys the rules to suspend the machines from power management settings.
However, soon after suspending it will wake the machine up again.
Looking at the pulse log it seems that it might be related to Auto configuration - even though we don’t use it.
Example:
2016-08-03 17:51:46: Server Thread - Auto Configuration: Received packet on autoconfig port
2016-08-03 17:51:46: Server Thread - Auto Configuration: Picking configuration based on: renderpc01d / ::ffff:192.168.1.164
2016-08-03 17:51:46: Server Thread - Auto Configuration: No config worth sending
whichever machine receives a packet on autoconfig port it will wake up unnecessarily.
WOL settings in the repository had been set to port 9 and 4343 (if that has anything to do with it…?)
Hi,
Looking into this further, I thought that this has something to do with Pulse but it seems that turning off Pulse doesn’t make any difference.
Steps to replicate:
- remove all rulesets for Auto-Configuration in the repository config options.
- Make sure that the slaves are not in any groups and pools and no jobs are trying to render to it.
- Suspend them manually. (or Pulse’s power management will suspend them accordingly if running)
- A couple of minutes later it will wake up again by itself.
But for sure it has something to do with the auto-configuration packet which wakes the slaves up.
As from the log, the slaves will always wake up after it receives packet on autoconfig port.
I’ve tried changing the AutoconfigurationPort in Deadline.ini file but it doesn’t alter this behaviour.
I also notice a couple of things which may or may not be normal…
- In the monitor none of my slaves says that they are connected to Pulse - even though Pulse seems to be seeing them and actively doing power management on them?
- It seems that there are a couple changes in the Pulse gui - now it no longer tells you the web service and the slave throttling status. I presume that this is by design, but in the Repository Configuration settings, there’s no tick box to turn web service on and where can we see the throttling status now?
Thanks!
Hello,
So I wanted to clarify a bit on how all of this works. When a slave starts up, and the repo says there is an auto config, it checks to see if there is a known pulse machine, and if one is not found, it sends out a broadcast to see if there is a pulse machine on the network. These signals should not at all be able to start a machine that is in suspend or shutdown. Also, if auto config is off, there will be no such broadcast sent as the slave will know there is no auto config from the repo.
Do you know which machine on your repo is running a slave that you think would be sending this broadcast to cause these machines to start up? Regarding the slave log showing an auto config packet, as that is the first thing it looks for, this will be the case any time the slave starts up. You can see more about what pulse is doing by enabling verbose pulse logging in the repository options.
In Deadline 8, the Web Service is it’s own application, no longer a part of Pulse, so you would want to run that application if you want to use the Web Service. As for throttling settings, those can be found under the slave settings section of the repository options.
I think I hit all your points, but let me know if I missed anything.
Dwight
Hi Dwight,
Thanks for the explanation of the process sequence, If that’s the expected behaviour then I think there’s something weird happening with our repo. Wondering if you could help us to identify the issue? I’ll send you the pulse log via thinkbox support ticket which might give you a better idea on what’s happening.
I don’t have any rulesets for the auto config, so in theory there should be no broadcast sent to the slaves, but do I have to untick an option box to turn off auto config from the repo somewhere?
The machine that broadcasts the auto config packet is a dedicated machine for pulse - verbose logging is ticked on in the repo.
I noticed that in the monitor even though we have pulse running, in the “connected to pulse” column, all of our slaves says ‘no’
We have 2 machines dedicated for pulse - setting one or the other to primary doesn’t make a difference in behaviour.
I’ve also just updated the farm to 8.0.7.3 and no difference in behaviour.
Not sure if it has any implications, but when I installed Deadline 8 for the first time, I was running it in parallel with Deadline 7 prior to moving the whole studio completely to Deadline 8 and uninstalling the Deadline 7 clients. (I followed the instructions on the Deadline Blog so the repositories are on separate VMs) Please let me know if there are any other steps I can do to help diagnosis. Thanks again!
Thanks for sending the log in, hoping we can get this sorted on the ticket soon, and update here.