[slurm-users] Servers in pending state
Hello, I have a queue with 6 servers. When 4 of the servers are with heavy load, If I send new jobs to the other 2 servers which are free and under different partition and features, The jobs are still in pending mode (can take them 20 minutes to start running) If I change their priority with "scontrol update" they start to run immediately. I am guessing it take Slurm a lot of time to reschedule all jobs when there is a heavy load so the new jobs are not check until I change their priority. Is there a way to tell Slurm to check all pending jobs every 2 minutes so if there are pending jobs on a free servers they will start running? More info: SchedulerType = sched/backfill SchedulerParameters = bf_continue,bf_max_job_test=300 Thank! Roy. *** Please consider the environment before printing this email ! The information contained in this communication is proprietary to Israel Aerospace Industries Ltd. and/or third parties, may contain confidential or privileged information, and is intended only for the use of the intended addressee thereof. If you are not the intended addressee, please be aware that any use, disclosure, distribution and/or copying of this communication is strictly prohibited. If you receive this communication in error, please notify the sender immediately and delete it from your computer. Thank you. Visit us at: www.iai.co.il
Re: [slurm-users] slurmd -C showing incorrect core count
Yep, slurmd -C is obviously getting the data from somewhere, either a local file or from the master node. hence my email to the group; I was hoping that someone would just say: "yeah, modify file ". But oh well. I'll start playing with strace and gdb later this week; looking through the source might also be helpful. I'm not cloning existing virtual machines with slurm. I have access to a vmware system that from time to time isn't running at full capacity; usage is stable for blocks of a month or two at a time, so my thought/plan was to spin up a slurm compute node on it, and resize it appropriately every few months (why not put it to work). I started with 10 cores, and it looks like I can up it to 16 cores for a while, and that's when I ran into the problem. -mike *Michael Tie*Technical Director Mathematics, Statistics, and Computer Science One North College Street phn: 507-222-4067 Northfield, MN 55057 cel:952-212-8933 m...@carleton.edufax:507-222-4312 On Wed, Mar 11, 2020 at 1:15 AM Kirill 'kkm' Katsnelson wrote: > On Tue, Mar 10, 2020 at 1:41 PM mike tie wrote: > >> Here is the output of lstopo >> > >> *$* lstopo -p >> >> Machine (63GB) >> >> Package P#0 + L3 (16MB) >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#0 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#1 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#2 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#3 >> >> Package P#1 + L3 (16MB) >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#4 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#5 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#6 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#7 >> >> Package P#2 + L3 (16MB) >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#8 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#9 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#10 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#11 >> >> Package P#3 + L3 (16MB) >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#12 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#13 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#14 >> >> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#15 >> > > There is no sane way to derive the number 10 from this topology. > obviously: it has a prime factor of 5, but everything in the lstopo output > is sized in powers of 2 (4 packages, a.k.a. sockets, 4 single-threaded CPU > cores per). > > I responded yesterday but somehow managed to plop my signature into the > middle of it, so maybe you have missed inline replies? > > It's very, very likely that the number is stored *somewhere*. First to > eliminate is the hypothesis that the number is acquired from the control > daemon. That's the simplest step and the largest landgrab in the > divide-and-conquer analysis plan. Then just look where it comes from on the > VM. strace(1) will reveal all files slurmd reads. > > You are not rolling out the VMs from an image, ain't you? I'm wondering > why do you need to tweak an existing VM that is already in a weird state. > Is simply setting its snapshot aside and creating a new one from an image > hard/impossible? I did not touch VMWare for more than 10 years, so I may be > a bit naive; in the platform I'm working now (GCE), create-use-drop pattern > of VM use is much more common and simpler than create and maintain it to > either *ad infinitum* or *ad nauseam*, whichever will have been reached the > earliest. But I don't know anything about VMWare; maybe it's not possible > or feasible with it. > > -kkm > >
[slurm-users] Can some partitions be excluded from a cluster when Slurm calculates user/total usage in Fairshare algorithm?
Dear All, We have a HPC cluster with Slurm job scheduler (17.02.8). There are several private partitions (which are sponsored by several groups) and a "common" partition. Private partitions are exclusively used by those private users, and all users (including private users) have equal access to the "common" partition. Currently we have set up Multi-factor Job Priority plugin to determine job priority. Everything is working ok except that private users are less favourable in the fair-share factor than non-private users when using the "common" partition. Private users may run many jobs on their own private partitions, and occasionally run jobs on the "common" partition. When submitting jobs to the "common" partition, those private users normally get much lower fair-share factor in job priority compared with non-private users. It appears those private users are penalized because they have already used a large amount of resources on the cluster (even those resources are private owned). In Fairshare Algorithm( https://slurm.schedmd.com/classic_fair_share.html), it appears when calculating Uuser and Utotal, the consumed processor*seconds is based on the whole cluster. Can we exclude some partitions when calculating such consumed resources? Is such functionality available in Slurm? I have tried to set TRESBillingWeights="CPU=0.0,Mem=0.0,GRES/gpu=0.0" in the private partitions. It seems such settings only affect the TRES factor, NOT Fair-share factor. Any suggestion? Thanks Manhui
[slurm-users] Upgrade paths
Hi all, I have one cluster running v16.05.4 that I would like to upgrade if possible to 19.05.5; it was installed via a .deb package I created back in 2016. I have located a 17.11.7 Ubuntu PPA (https://launchpad.net/~jonathonf/+archive/ubuntu/slurm) and have myself recently put up one for 19.05.5 (https://launchpad.net/~wdennis/+archive/ubuntu/dhpc-backports). Theoretically, I believe I should be able to upgrade from the 16.05 release to 17.11, then from 17.11 to 19.05, correct? (going under the assumption that can only go forward at most 2 Slurm releases, which went 16.05 -> 17.02 -> 17.11 -> 18.08 -> 19.05, if I am correct.) Thanks, Will
Re: [slurm-users] Upgrade paths
The release notes at https://slurm.schedmd.com/archive/slurm-19.05.5/news.html indicate you can upgrade from 17.11 or 18.08 to 19.05. I didn’t find equivalent release notes for 17.11.7, but upgrades over one major release should work. > On Mar 11, 2020, at 2:01 PM, Will Dennis wrote: > > External Email Warning > This email originated from outside the university. Please use caution when > opening attachments, clicking links, or responding to requests. > Hi all, > > I have one cluster running v16.05.4 that I would like to upgrade if possible > to 19.05.5; it was installed via a .deb package I created back in 2016. I > have located a 17.11.7 Ubuntu PPA > (https://launchpad.net/~jonathonf/+archive/ubuntu/slurm) and have myself > recently put up one for 19.05.5 > (https://launchpad.net/~wdennis/+archive/ubuntu/dhpc-backports). > Theoretically, I believe I should be able to upgrade from the 16.05 release > to 17.11, then from 17.11 to 19.05, correct? (going under the assumption that > can only go forward at most 2 Slurm releases, which went 16.05 -> 17.02 -> > 17.11 -> 18.08 -> 19.05, if I am correct.) > > Thanks, > Will
Re: [slurm-users] Upgrade paths
On 11-03-2020 20:01, Will Dennis wrote: I have one cluster running v16.05.4 that I would like to upgrade if possible to 19.05.5; it was installed via a .deb package I created back in 2016. I have located a 17.11.7 Ubuntu PPA (https://launchpad.net/~jonathonf/+archive/ubuntu/slurm) and have myself recently put up one for 19.05.5 (https://launchpad.net/~wdennis/+archive/ubuntu/dhpc-backports). Theoretically, I believe I should be able to upgrade from the 16.05 release to 17.11, then from 17.11 to 19.05, correct? (going under the assumption that can only go forward at most 2 Slurm releases, which went 16.05 -> 17.02 -> 17.11 -> 18.08 -> 19.05, if I am correct.) You may find some useful information collected in my Slurm Wiki page https://wiki.fysik.dtu.dk/niflheim/Slurm_installation#upgrading-slurm /Ole
Re: [slurm-users] slurmd -C showing incorrect core count
Yup, I think if you get stuck so badly, the first thing is to make sure the node does not get the number 10 from the controller, and the second just reimage the VM fresh. It maybe not the quickest way, but at least predictable in the sense of time spent. Good luck! -kkm On Wed, Mar 11, 2020 at 7:28 AM mike tie wrote: > > Yep, slurmd -C is obviously getting the data from somewhere, either a > local file or from the master node. hence my email to the group; I was > hoping that someone would just say: "yeah, modify file ". But oh > well. I'll start playing with strace and gdb later this week; looking > through the source might also be helpful. > > I'm not cloning existing virtual machines with slurm. I have access to a > vmware system that from time to time isn't running at full capacity; usage > is stable for blocks of a month or two at a time, so my thought/plan was to > spin up a slurm compute node on it, and resize it appropriately every few > months (why not put it to work). I started with 10 cores, and it looks > like I can up it to 16 cores for a while, and that's when I ran into the > problem. > > -mike > > > > *Michael Tie*Technical Director > Mathematics, Statistics, and Computer Science > > One North College Street phn: 507-222-4067 > Northfield, MN 55057 cel:952-212-8933 > m...@carleton.edufax:507-222-4312 > > > On Wed, Mar 11, 2020 at 1:15 AM Kirill 'kkm' Katsnelson > wrote: > >> On Tue, Mar 10, 2020 at 1:41 PM mike tie wrote: >> >>> Here is the output of lstopo >>> >> >>> *$* lstopo -p >>> >>> Machine (63GB) >>> >>> Package P#0 + L3 (16MB) >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#0 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#1 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#2 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#3 >>> >>> Package P#1 + L3 (16MB) >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#4 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#5 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#6 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#7 >>> >>> Package P#2 + L3 (16MB) >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#8 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#9 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#10 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#11 >>> >>> Package P#3 + L3 (16MB) >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#0 + PU P#12 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#1 + PU P#13 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#2 + PU P#14 >>> >>> L2 (4096KB) + L1d (32KB) + L1i (32KB) + Core P#3 + PU P#15 >>> >> >> There is no sane way to derive the number 10 from this topology. >> obviously: it has a prime factor of 5, but everything in the lstopo output >> is sized in powers of 2 (4 packages, a.k.a. sockets, 4 single-threaded CPU >> cores per). >> >> I responded yesterday but somehow managed to plop my signature into the >> middle of it, so maybe you have missed inline replies? >> >> It's very, very likely that the number is stored *somewhere*. First to >> eliminate is the hypothesis that the number is acquired from the control >> daemon. That's the simplest step and the largest landgrab in the >> divide-and-conquer analysis plan. Then just look where it comes from on the >> VM. strace(1) will reveal all files slurmd reads. >> >> You are not rolling out the VMs from an image, ain't you? I'm wondering >> why do you need to tweak an existing VM that is already in a weird state. >> Is simply setting its snapshot aside and creating a new one from an image >> hard/impossible? I did not touch VMWare for more than 10 years, so I may be >> a bit naive; in the platform I'm working now (GCE), create-use-drop pattern >> of VM use is much more common and simpler than create and maintain it to >> either *ad infinitum* or *ad nauseam*, whichever will have been reached the >> earliest. But I don't know anything about VMWare; maybe it's not possible >> or feasible with it. >> >> -kkm >> >>
Re: [slurm-users] slurmd -C showing incorrect core count
On 10/3/20 1:40 pm, mike tie wrote: Here is the output of lstopo Hmm, well I believe Slurm should be using hwloc (which provides lstopo) to get its information (at least it calls the xcpuinfo_hwloc_topo_get() function for that), so if lstopo works then slurmd should too. Ah, looking a bit deeper I see in src/slurmd/common/xcpuinfo.c: if (!hwloc_xml_whole) hwloc_xml_whole = xstrdup_printf("%s/hwloc_topo_whole.xml", conf->spooldir); Do you happen to have a file called "hwloc_topo_whole.xml" in your spool directory on that node? I'm wondering if it's cached old config there. If so move it out of the way somewhere safe (just in case) and try again. All the best, Chris -- Chris Samuel : http://www.csamuel.org/ : Berkeley, CA, USA