Hey Danny, Sorry for the delay. I got started on the patch to 2.3 but then had my attention diverted by a severe power outage at work ( http://arbutus.patch.com/articles/transformer-explosion-knocks-umbc-power).
I finally had a chance to finish the patch against 2.3-- It was pretty straightforward. Here's the link: http://userpages.umbc.edu/~aaronk/slurm/patches/qos_per_user_node_cpu_limits_slurm23.patch I made some slight changes from the original patch to 2.2 including changing the sacctmgr parameters from MaxUser(Node|CPU)s to Max(Node|Cpu)sPerUser and updating the man pages. Best, Aaron On Wed, Jun 1, 2011 at 3:29 PM, Auble, Danny <[email protected]> wrote: > Hey Aaron, sorry for the mistake there. It appears I was looking at the > wrong structure in slurmdb.h (My mistake). Yes, what you did appears to be > the correct way of doing it. I am not sure how I got things mistaken, > thanks for pointing it out. > > > > Thanks for working on the 2.3 patch, it probably isn’t that bad. > > > > Danny > > > > *From:* [email protected] [mailto: > [email protected]] *On Behalf Of *Aaron Knister > *Sent:* Wednesday, June 01, 2011 12:19 PM > *To:* [email protected] > *Subject:* Re: [slurm-dev] per-user node/cpu limits for QOS's > > > > Hey Danny, > > > > I'm not sure I understand what you're asking-- > > > > My understanding is that the slurmdb_uesd_limits_t struct to which I added > the 'nodes' and 'cpus' variables gets compiled into a user_limit_list that > is stored within the assoc_mgr_qos_usage_t. Since they're per-user limits it > looks like they need to be added to the slurmdb_used_limits_t struct. If > that's not correct please let me know and I'll be happy to make the > necessary adjustments to the patch. I can also work to make a patch against > 2.3 as well. > > > > Thanks! > > > > -Aaron > > > > On Wed, Jun 1, 2011 at 1:03 PM, Auble, Danny <[email protected]> wrote: > > Hey Aaron, thanks for the patch. > > > > A few questions/comments though. > > > > In the slurm/slurmdb.h file you added a ‘nodes’ and a ‘cpus’ variable there > to be packed. I only glanced at the patch (which will probably not go into > 2.3 cleanly) and was wondering if these were needed outside of the > slurmctld? Is there any way you could just use the variables in the > assoc_mgr_qos_usage_t structure, or add them there with the other tracking > variables? It would be nice to get a patch for 2.3 as well since only major > bug fixes will be added to 2.2. > > > > The additions are a good idea though. Thanks for them. Let me know if you > have any questions/concerns over my proposals. > > > > Thanks, > > Danny > > > > *From:* [email protected] [mailto: > [email protected]] *On Behalf Of *Aaron Knister > *Sent:* Tuesday, May 31, 2011 10:52 AM > *To:* slurm-dev > *Subject:* [slurm-dev] per-user node/cpu limits for QOS's > > > > Howdy, > > We have a need to limit how many nodes/CPUs a given user can use in a QOS. > I've written a patch against 2.2.5 to do this that in testing appears to > work as expected. I'm fairly sure this patch won't be able to be included in > 2.2 since it bumps the SLURMDB_VERSION number, but none the less any > feedback is appreciated. > > The patch can be found here: > http://userpages.umbc.edu/~aaronk/slurm/patches/qos_per_user_node_cpu_limits.patch. > I didn't paste it in the e-mail since it's 451 lines long :) > > -Aaron > > >
