> On Aug 12, 2016, at 12:16, Kilian Cavalotti <[email protected]> > wrote: > > On Thu, Aug 11, 2016 at 9:44 PM, Ryan Novosielski <[email protected]> > wrote: >> [pid 11767] >> open("/sys/fs/cgroup/devices/slurm/uid_109366/job_5377709/devices.allow", >> O_WRONLY) = 10 >> [pid 11767] write(10, "501835 rwm", 10) = -1 EINVAL (Invalid argument) >> [pid 11767] close(10) = 0 > > You bet it's EINVAL. :) > > Question is: where does it get that 501835 number from? Looks like a > bug to me.
That number, neither 501 nor 835 (figured, maybe), appear anyplace else in either the output of find /dev -ls or the entire strace output that I’ve excerpted. I thought maybe a funny special character at the end of group_allowed_devices_file.conf or something, so I tried reordering the file just for fun. It happens after everything in that file is dealt with. My next step, I think, was going to be trying with “ConstrainDevices” disabled so that I can at least possibly introduce core and memory constraints and finish this up a little later. > I would suggest: > 1. try a newer version: you mentioned you're running 15.08.4. I would > try at least 15.08.9, or if possible, move to the 16.05 release branch. Will do. It should be possible for me to do that fairly quickly. > 2. submit a bug to http://bugs.schedmd.com (esp. if you have a support > contract, which IMHO, is worth every penny), that's the best way to > get valuable insight. Although I suspect they may suggest 1. :) I do not, but am inching ever-closer to looking into it. :-P > And just to add to what Chris said, I'm also running GPU nodes with > the cgroup devices subsystem (albeit on CentOS 6.x) and it works very > well. Thanks for the confirmation. Glad to hear it. -- ____ || \\UTGERS, |---------------------------*O*--------------------------- ||_// the State | Ryan Novosielski - [email protected] || \\ University | Sr. Technologist - 973/972.0922 (2x0922) ~*~ RBHS Campus || \\ of NJ | Office of Advanced Research Computing - MSB C630, Newark `'
