- slurm-users - lists.schedmd.com

slurm send email status with no details
by Oren 07 Apr '25

07 Apr '25

Hi, We set up a slurm system with email notification, this is the slurm.conf `MailProg=/usr/sbin/sendmail` But the email that I get has not status, just an empty message: [image: image.png] no subject, no info, what are we missing? Thanks~

2 1

Run a command in Slurm with all streams and signals connected to the submitting command
by Michael Milton 07 Apr '25

07 Apr '25

I'm helping with a workflow manager that needs to submit Slurm jobs. For logging and management reasons, the job (e.g. srun python) needs to be run as though it were a regular subprocess (python): - stdin, stdout and stderr for the command should be connected to process inside the job - signals sent to the command should be sent to the job process - We don't want to use the existing job allocation, if this is run from a Slurm job - The command should only terminate when the job is finished, to avoid us needing to poll Slurm We've tried: - sbatch --wait, but then SIGTERM'ing the process doesn't kill the job - salloc, but that requires a TTY process to control it (?) - salloc srun seems to mess with the terminal when it's killed, likely because of being "designed to be executed in the foreground" - Plain srun re-uses the existing Slurm allocation, and specifying resources like --mem will just request then from the current job rather than submitting a new one What is the best solution here?

2 2

Re: Run a command in Slurm with all streams and signals connected to the submitting command
by Michael Milton 05 Apr '25

05 Apr '25

Thanks Davide, It's true that srun will create an allocation if you aren't inside a job, but if you are inside a job and you request more resources than it has, then srun will just fail. This is the key issue that I want to avoid. On Sat, Apr 5, 2025 at 11:48 AM Davide DelVento <davide.quantum(a)gmail.com> wrote: > The plain srun is probably the best bet, and if you really need the thing > to be started from another slurm job (rather than the login node) you will > need to exploit the fact that > > > If necessary, srun will first create a resource allocation in which to > run the parallel job. > > AFAIK, there is no option to for the "create a resource allocation" even > if it's not necessary. But you may try to request something that is "above > and beyond" what the current allocation provides, and that might solve your > problem. > Looking at the srun man page, I could speculate that --clusters > or --cluster-constraint might help in that regard (but I am not sure). > > Have a nice weekend > > > On Fri, Apr 4, 2025 at 6:27 AM Michael Milton via slurm-users < > slurm-users(a)lists.schedmd.com> wrote: > >> I'm helping with a workflow manager that needs to submit Slurm jobs. For >> logging and management reasons, the job (e.g. srun python) needs to be run >> as though it were a regular subprocess (python): >> >> - stdin, stdout and stderr for the command should be connected to >> process inside the job >> - signals sent to the command should be sent to the job process >> - We don't want to use the existing job allocation, if this is run >> from a Slurm job >> - The command should only terminate when the job is finished, to >> avoid us needing to poll Slurm >> >> We've tried: >> >> - sbatch --wait, but then SIGTERM'ing the process doesn't kill the job >> - salloc, but that requires a TTY process to control it (?) >> - salloc srun seems to mess with the terminal when it's killed, >> likely because of being "designed to be executed in the foreground" >> - Plain srun re-uses the existing Slurm allocation, and specifying >> resources like --mem will just request then from the current job rather >> than submitting a new one >> >> What is the best solution here? >> >> -- >> slurm-users mailing list -- slurm-users(a)lists.schedmd.com >> To unsubscribe send an email to slurm-users-leave(a)lists.schedmd.com >> >

1 0

Best Way to See GPUs in Use?
by Jason Simms 02 Apr '25

02 Apr '25

Hello all, Apologies for the basic question, but is there a straightforward, best-accepted method for using Slurm to report on which GPUs are currently in use? I've done some searching and people recommend all sorts of methods, including parsing the output of nvidia-smi (seems inefficient, especially across multiple GPU nodes), as well as using other tools such as Grafana, XDMoD, etc. We do track GPUs as a resource, so I'd expect I could get at the info with sreport or something like that, but before trying to craft my own from scratch, I'm hoping someone has something working already. Ultimately I'd like to see either which cards are available by node, or the reverse (which are in use by node). I know recent versions of Slurm supposedly added tighter integration in some way with NVIDIA cards, but I can't seem to find definitive docs on what, exactly, changed or what is now possible as a result. Warmest regards, Jason -- *Jason L. Simms, Ph.D., M.P.H.* Research Computing Manager Swarthmore College Information Technology Services (610) 328-8102

3 2

slum job sumisison using different UID/GID
by navin srivastava 02 Apr '25

02 Apr '25

Hi, Can munge depend upon the UID/GID for the job submission? There is a scenario where the cluster is running with the NIS authentication mechanism. There is a login Node created which is part of LDAP and we installed the slurm to interact with slurm master. I am able to run jobs using root but for users it is not working as the UID/GID differ. Is there a way we can overcome this issue? and be able to run jobs. Rehards Navin

1 0

slurm releases
by Patrick Begou 01 Apr '25

01 Apr '25

Hi slurm team, I would ask some clarifications with slurm releases. Why two versions of slurm are available ? I speak of 24.05.7 versus 24.11.3 on https://www.schedmd.com/slurm-support/release-announcements and announces made on this list ? I'm managing small clusters in a french public research lab and I've successfully updated (lets says re-installed as it was very old) an old one with slurm-24.05.3 some month ago. Now I'm trying to deploy a second cluster with the same approach and I've selected 24.11.3-1 (but why not using 24.05.7 ?). Moreover I have some difficulties for deploying 24.11.3 : I've built the rpm with: rpmbuild \ --define '_prefix /usr' \ --with pmix --with ucx --with mysql --with hwloc \ -ta slurm-24.11.3.tar.bz2 but some packages wont install: $ dnf list --showduplicates slurm-slurmd Last metadata expiration check: 0:33:15 ago on Tue Apr 1 18:03:07 2025. Available Packages slurm-slurmd.x86_64 24.11.3-1.el9 bb-local $ sudo dnf install slurm-slurmd Last metadata expiration check: 0:33:03 ago on Tue Apr 1 18:03:25 2025. Error: Problem: conflicting requests - nothing provides is needed by slurm-slurmd-24.11.3-1.el9.x86_64 from bb-local - nothing provides installed needed by slurm-slurmd-24.11.3-1.el9.x86_64 from bb-local - nothing provides not needed by slurm-slurmd-24.11.3-1.el9.x86_64 from bb-local (try to add '--skip-broken' to skip uninstallable packages or '--nobest' to use not only best candidate packages) I think I will try to build 24.05.7 or 24.05.3 as a next try but I'm interested in any advices. Thank you Patrick

2 1

cpus and gpus partitions and how to optimize the resource usage
by Massimo Sgaravatto 31 Mar '25

31 Mar '25

Dear all We have just installed a small SLURM cluster composed of 12 nodes: - 6 CPU only nodes: 2 Sockets=2, 96 CoresPerSocket 2, ThreadsPerCore=2, 1.5 TB of RAM - 6 nodes with also GPUS: same conf of the CPU-only node + 4 H100 per node We started with a setup with 2 partitions: - a 'onlycpus' partition which sees all the cpu-only nodes - a 'gpus' partition which sees the nodes with gpus and asked users to use the 'gpus' partition only for jobs that need gpus (for the time being we are not technically enforced that). The problem is that a job requiring a GPU usually needs only a few cores and only a few GB of RAM, which means wasting a lot of CPU cores. And having all nodes in the same partition would mean that there is the risk that a job requiring a GPU can't start if all CPU cores and/or all memory is used by CPU only jobs I went through the mailing list archive and I think that "splitting" a GPU node into two logical nodes (one to be used in the 'gpus' partition and one to be used in the 'onlycpus' partition) as discussed in [*] would help. Since that proposed solution is considered by his author a "bit of a kludge" and since I read that splitting a node into multiple logical nodes is in a general a bad idea, I'd like to understand if you could suggest other/best options. I also found this [**] thread, but I don't like too much that approach (i.e. relying on MaxCPUsPerNode) because it would mean having 3 partition (if I have got it right): two partitions for cpu only jobs and 1 partition for gpu jobs Many thanks, Massimo [*] https://groups.google.com/g/slurm-users/c/IUd7jLKME3M [**] https://groups.google.com/g/slurm-users/c/o7AiYAQ1YJ0

4 5

Preemption question
by Kamil Wilczek 31 Mar '25

31 Mar '25

Dear All, I would like to be able to preempt (SUSPEND) a single QoS of a user that blocks the queue for several days. Currently I have about 100 users on the cluster and it seems that setting the "Preempt" option to each QoS (we have personal QoSes) is not optimal. https://slurm.schedmd.com/sacctmgr.html#OPT_Preempt Is there a way to to set an option to this single problematic QoS, saying that the QoS can be preempted by any other QoS? It would be much more administrator-friendly solution ;) Kind regards -- Kamil Wilczek [https://keys.openpgp.org/] [D415917E84B8DA5A60E853B6E676ED061316B69B]

3 4

Slurm 24.05 and OpenMPI
by Matthias Leopold 28 Mar '25

28 Mar '25

Hi, I built a small Slurm 21.08 cluster with NVIDIA GPU hardware and NVIDIA deepops framework a couple of years ago. It is based on Ubuntu 20.04 and makes use of the NVIDIA pyxis/enroot container solution. For operational validation I used the nccl-tests application in a container. nccl-tests is compiled with MPI support (OpenMPI 4.1.6 or 4.1.7) and I used it also for validation of MPI jobs. Slurm jobs use "pmix" and tasks are launched via srun (not mpirun). Some of the GPUs can talk to each other via Infiniband, but MPI is rarely used at our site and I'm fully aware that my MPI knowledge is very limited. Still it worked with Slurm 21.08. Now I built a Slurm 24.05 cluster based on Ubuntu 24.04 and started to move hardware there. When I run my nccl-tests container (also with newer software) I see error messages like this: [node1:21437] OPAL ERROR: Unreachable in file ext3x_client.c at line 111 -------------------------------------------------------------------------- The application appears to have been direct launched using "srun", but OMPI was not built with SLURM's PMI support and therefore cannot execute. There are several options for building PMI support under SLURM, depending upon the SLURM version you are using: version 16.05 or later: you can use SLURM's PMIx support. This requires that you configure and build SLURM --with-pmix. Versions earlier than 16.05: you must use either SLURM's PMI-1 or PMI-2 support. SLURM builds PMI-1 by default, or you can manually install PMI-2. You must then build Open MPI using --with-pmi pointing to the SLURM PMI library location. Please configure as appropriate and try again. -------------------------------------------------------------------------- *** An error occurred in MPI_Init *** on a NULL communicator *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, *** and potentially your MPI job) [node1:21437] Local abort before MPI_INIT completed completed successfully, but am not able to aggregate error messages, and not able to guarantee that all other processes were killed! One simple question: Is this related to https://github.com/open-mpi/ompi/issues/12471? If so: is there some workaround? I'm very grateful for any comments. I know that a lot of detail information is missing, but maybe someone can still already give me a hint where to look. Thanks a lot Matthias

3 10

Doubt with SelectTypeParameters in slurm.conf
by Gestió Servidors 28 Mar '25

28 Mar '25

Hello, I'm running some tests in a very small testing environment (before applying in the real scenario). My environment is only a computer with a old Intel i4 with this "lscpu" configuration: Architecture: x86_64 CPU(s): 8 On-line CPU(s) list: 0-7 Thread(s) per core: 2 Core(s) per socket: 4 Socket(s): 1 NUMA node(s): 1 Vendor ID: GenuineIntel BIOS Vendor ID: Intel(R) Corporation CPU family: 6 Model: 26 Model name: Intel(R) Core(TM) i7 CPU 950 @ 3.07GHz "slurmd -C" returns: "CPUs=8 Boards=1 SocketsPerBoard=1 CoresPerSocket=4 ThreadsPerCore=2". My slurm.conf is configured with: SelectType=select/cons_tres SelectTypeParameters=CR_Core AccountingStorageTRES=cpu,mem,node,billing,fs/disk,vmem,pages After running a simple "helloworld" test, I have noticed that if SelectTypeParameters=CR_Core, system always reserves me an even number of CPUs (during "pending" time, I can see the real number I have requested, but when job starts "running", that number is increased to the next even number). Howerver, if I reconfigure slurm with SelectTypeParameters=CR_CPU, system reserves me the correct number of CPUs (during "pending" time and, also, during "running" time). I suppose that my system must be configured with CR_CPU, according https://slurm.schedmd.com/cons_tres.html, but could someone confirm that? Thanks.

2 1

bit_cache_init failure on the second time backup controller tries to take control
by Safdar Iqbal 28 Mar '25

28 Mar '25

Hi, We're running into an issue where slurmctld core-dumps with the following error. This happens on the backup controller, if it needs to take over from the primary, _for a second time_. slurmctld: fatal: bit_cache_init: cannot change size once set Has anyone seen this error before? Also if there are any existing discussions and/or tickets related to this, please let me know. Our slurm version is 24.11.1. ________________ Steps to reproduce: 1. On a healthy cluster, we make the primary controller unavailable. Since we're running our cluster on cloud VMs, we cause this by stopping the primary controller VM. 2. From the logs we can see the backup controller take over, and log the message "Running as primary controller" 3. We then start the primary again, making sure the IP addresses and hostnames stay consistent. Once slurmctld on the primary has started and taken back control, we can see the log "Running as primary controller" on that VM. 4. We then stop the primary controller VM again, causing the backup to try taking the control a second time. This time however the slurmctld on the backup coredumps, with following log entries from journalctl -u slurmctld: slurmctld: fatal: bit_cache_init: cannot change size once set slurmctld.service: Main process exited, code=dumped, status=6/ABRT slurmctld.service: Failed with result 'core-dump'. Thanks! - Safdar

1 0

How limit CPUs per node in a partition
by Gestió Servidors 27 Mar '25

27 Mar '25

Hello, I have a testing partition with only a node. That server has 12 CPUs (it's a very old server) (2 sockets, 6 cores per socket, 1 thread per core). That partition, called "test.q" only has that node, so by default, partition test.q has 12 CPUs (all from testing node). However, now I would like to decrease that CPU number from 12 to 4, for example. "slurmd -C" returns "CPUs=12 Boards=1 SocketsPerBoard=2 CoresPerSocket=6 ThreadsPerCore=1" as I have said to you before and if in slurm.conf I configure partition in this way "PartitionName=test.q Nodes=test-node MaxTime=10:00 State=UP AllocNodes=clus-login MaxCPUsPerNode=4", partition continues accepting jobs with 12 CPUs. Is it any wrong in my configuration? How could I get my purpose? Thanks. -- [cid:image001.jpg@01DB9F0E.FDD9A7A0]<http://www.uab.cat/> Daniel Ruiz Molina Tècnic Mitjà Informàtic Arquitectura de Computadors i Sistemes Operatius Escola d'Enginyeria Edifici Q - Despatx QC/3052 - Carrer de les Sitges Campus de la UAB · 08193 Bellaterra (Cerdanyola del Vallès) · Barcelona · Spain +34 93 581 35 44 www.uab.cat<http://www.uab.cat/> Daniel Ruiz at UAB<https://tinyurl.com/yd95zb8j> [cid:image002.jpg@01DB9F0E.FDD9A7A0]<www.linkedin.com/in/daniel-ruiz-molina-50a83b27> Aquest missatge s'adreça exclusivament a la persona destinatària i pot contenir informació privada o confidencial. Si l'heu rebut per error, comuniqueu-nos-ho i destruïu-lo, i tingueu present que no teniu autorització per fer-ne cap ús. Abans d'imprimir aquest missatge penseu en el medi ambient.

1 0

Re: Using more cores/CPUs that requested with
by Shunran Zhang 26 Mar '25

26 Mar '25

Ugh I think I did not catch up with the docs. I started with a system that defaults to cgroup v1 but the Slurm doc for that plugin is NOT available at that time. Thus I converted everything to cgroup v2. It appears that they are both supported and that documentation issue is more on the dev side than admin side. Thanks for pointing that out. I misinterpreted the "coming soon" part of cgroup v1 plugin and the "legacy" naming for "do not use". It should be fine. 2025年3月27日(木) 0:48 Williams, Jenny Avis <jenny_williams(a)unc.edu>: > “ … As cgroup is likely not supposed to be used in newer deployments of > Slurm.” > > > > I am curious about this statement. Would someone expand on this, to either > support or counter it? > > > > Jenny Williams > > UNC Chapel Hill > > > > > > *From:* Shunran Zhang via slurm-users <slurm-users(a)lists.schedmd.com> > *Sent:* Wednesday, March 26, 2025 10:52 AM > *To:* Gestió Servidors <sysadmin.caos(a)uab.cat> > *Cc:* Slurm User Community List <slurm-users(a)lists.schedmd.com> > *Subject:* [slurm-users] Re: Using more cores/CPUs that requested with > > > > If you are letting systemd taking most things over, you got systemd-cgtop > that work better than top for your case. There is also systemd-cgls for > non-interactive listing. > > > > Also mind to check if you are using cgroup2? A mount to check your cgroup > would suffice. As cgroup is likely not supposed to be used in newer > deployments of Slurm. > > > > > > 2025年3月26日(水) 17:14 Gestió Servidors via slurm-users < > slurm-users(a)lists.schedmd.com>: > > Hello, > > > > Thanks for your answers. I will try now!! One more question: is there any > way to check if Cgroups restrictions is working fine during a “running” job > or during SLURM scheduling process? > > > > Thanks again! > > > > > -- > slurm-users mailing list -- slurm-users(a)lists.schedmd.com > To unsubscribe send an email to slurm-users-leave(a)lists.schedmd.com > >

1 0

Re: Using more cores/CPUs that requested with
by Gestió Servidors 26 Mar '25

26 Mar '25

Hello, Thanks for your answers. I will try now!! One more question: is there any way to check if Cgroups restrictions is working fine during a "running" job or during SLURM scheduling process? Thanks again!

4 3

Using more cores/CPUs that requested with sbatch
by Gestió Servidors 25 Mar '25

25 Mar '25

Hello, I would like to know if there is any mechanism to avoid a user to do "cheating" when he submits a job. For example, a user submits a OpenMP job requesting 1 node (-N 1) and 2 tasks (-n 2). However, inside his script (or compiled in his binany), user uses more than 2 ntasks, for example, all free cores. For SLURM, job is only using 2, but really, node is 100% assigned to that job. SLURM can control that "abuse" in some way? Cgroups? Thanks.

3 2

Re: Using more cores/CPUs that requested with sbatch
by Megan 25 Mar '25

25 Mar '25

Slurm can restrict jobs to their allocated resources using cgroups. To use cgroups you will have to configure the TaskPlugin cgroup option and specify how the cgroups should behave in cgroup.conf. slurm.conf TaskPlugin=task/cgroup Slurm Workload Manager - slurm.conf <https://slurm.schedmd.com/slurm.conf.html#OPT_task/cgroup> cgroup.conf ConstrainCores=yes Slurm Workload Manager - cgroup.conf <https://slurm.schedmd.com/cgroup.conf.html#OPT_ConstrainCores> Regards, --Megan

1 0

slurmrestd equivalent to "srun -n 10 echo HELLO"
by Dan Healy 24 Mar '25

24 Mar '25

Hi Slurm Community, I'm starting to experiment with slurmrestd for a new app we're writing. I'm having trouble understanding one aspect of submitting jobs. When I run something like `srun -n 10 echo HELLO', I get HELLO returned to my console/stdout 10x. When I submit this command as a script to the /jobs/submit route, I get success/200, but *I cannot determine how to get the console output of HELLO 10x in any form*. It's not in my stdout log for that job even though I can verify that the job ran successfully. Any suggestions? -- Thanks, Daniel Healy

3 2

Priority/Top seems to not be working
by Dan Healy 24 Mar '25

24 Mar '25

Hi Slurm Users, I have a newer install (23.11.3) and the priority of most all jobs is 1 and other small numbers. In previous versions, I would see numbers like 2^32. I have the multifactor plugin configured and confirmed it's in-use when I show the config. When I run `scontrol top` for a given job, the priority number doesn't change from 1. I have lots of jobs running at the moment and many queued as well. I've been using `scontrol top` for years and have seen the priority successfully change. I'm repeating this now with the newer version of Slurm and now seeing it work the same. Are you seeing this too? -- Thanks, Daniel Healy

1 0

missing jobs from queue
by Felix 14 Mar '25

14 Mar '25

Hello at site RO-14-ITIM, we have a very strange problem. A ticket was raised with the following: "Hi, The new ticket (Pilots disappearing from local batch system at ITIM) has been created by "*Timo Wilken*". Group: NGIs › NGI_RO Owner: - State: in progress Information: Hi, Pilots seem to be lost by the local batch system at ITIM. >50% of pilots fail: https://apfmon.lancs.ac.uk/CERN_central_A:RO-14-ITIM-arcn-node?state=fault Example log: https://aipanda156.cern.ch/condor_logs_2/25-03-12_13/grid.15478691.1.log Error message: "ARC job failed: LRMS error: (-1) Job missing from SLURM" Cheers, Timo" The problem is I can not find any of the jobs that are failing on the site: for example: the last job: https://aipanda023.cern.ch/condor_logs_2/25-03-14_09/grid.19602200.0.log with stdlog: 000 (19602200.000.000) 2025-03-14 09:28:03 Job submitted from host: <137.138.31.125:38090?addrs=137.138.31.125-38090+[2001-1458-d00-19--75]-38090&alias=aipanda023.cern.ch> ... 027 (19602200.000.000) 2025-03-14 09:28:13 Job submitted to grid resource GridResource: arc arcn-node.itim-cj.ro:443 GridJobId: arc arcn-node.itim-cj.ro:443 PPVNDm5VGD7ngvuSSqSAreymYz3jwmOETUEm1dOSDmbfXLDmHXmz9n ... 001 (19602200.000.000) 2025-03-14 09:40:51 Job executing on host: arc arcn-node.itim-cj.ro:443 ... 012 (19602200.000.000) 2025-03-14 09:40:58 Job was held. ARC job failed: LRMS error: (-1) Job missing from SLURM Code 0 Subcode 0 ... 009 (19602200.000.000) 2025-03-14 11:03:03 Job was aborted. Python-initiated action. (by user atlpan) ... 009 (19602200.000.000) 2025-03-14 11:03:17 Job was aborted. Python-initiated action. (by user atlpan) This name: PPVNDm5VGD7ngvuSSqSAreymYz3jwmOETUEm1dOSDmbfXLDmHXmz9n is nowhere to find on my server: [root@arcn-node log]# updatedb [root@arcn-node log]# locate PPVNDm5VGD7ngvuSSqSAreymYz3jwmOETUEm1dOSDmbfXLDmHXmz9n [root@arcn-node log]# I have on my site this kind of error The ENDPOINT affected is arcn-node.itim-cj.ro (ARC-CE)It became Critical*at*2025-03-14T09:16:44Z due to *METRIC*org.nordugrid.ARC-CE-sw-gcc *Summary:*Script exited with code 127. Could not match GCC version. See /etc/arc/nagios/20-dist.ini for debugging hints. *Message:*'' which is going to be ok in the messag, one minute later The ENDPOINT affected is arcn-node.itim-cj.ro (ARC-CE) It became Ok *at* 2025-03-14T09:17:53Z due to *METRIC* org.nordugrid.ARC-CE-sw-gcc *Summary: *Found GCC version 11.5.0. *Message: *'' Can you please advice where to look for any clue in solving this mystery? Thank you Felix :-- Dr. Eng. Farcas Felix National Institute of Research and Development of Isotopic and Molecular Technology, IT - Department - Cluj-Napoca, Romania Mobile: +40742195323

1 0

Slurm version 24.05.7 is now available
by Marshall Garey 13 Mar '25

13 Mar '25

We are pleased to announce the availability of Slurm version 24.05.7. This release fixes some stability issues in 24.05, including a crash in slurmctld after updating a reservation with an empty nodelist. Downloads are available at https://www.schedmd.com/downloads.php . -- Marshall Garey Release Management, Support, and Development SchedMD LLC - Commercial Slurm Development and Support > * Changes in Slurm 24.05.7 > ========================== > -- Fix slurmctld crash when after updating a reservation with an empty > nodelist. The crash could occur after restarting slurmctld, or if > downing/draining a node in the reservation with the REPLACE or REPLACE_DOWN > flag. > -- Fix jobs being scheduled on higher weighted powered down > nodes. > -- Fix memory leak when RestrictedCoresPerGPU is enabled. > -- Prevent slurmctld deadlock in the assoc mgr.

1 0

SLURM_JOB_ACCOUNT var missing in prolog
by Jonás Arce 13 Mar '25

13 Mar '25

Hi everyone, The Slurm variable SLURM_JOB_ACCOUNT is blank in my prolog, currently i have a TaskProlog and a Prolog in my system, in TaskProlog this variable works just fine (I use TaskProlog to set env variables), but in my Prolog (which I use to create temporal directories), it's just blank, it's strange beacuse other variables such as SLURM_JOB_PARTITION, SLURM_JOB_ACCOUNT or SLURM_JOBID work just fine in both Prolog and TaskProlog, I looked into the Slurm doc and it seems that this var should work everywhere, if someone could shed some light into this or tell me another equivalent var i'd appreciate it a lot. I need it because I need to make this type of temporal directories with my Prolog: /scratch-global/$SLURM_JOB_ACCOUNT/$SLURM_JOB_USER/$SLURM_JOBID.

2 1

Job Env Vars in Slurm Core
by Bhaskar Chakraborty 13 Mar '25

13 Mar '25

Hi everyone, I have tried my best to extract custom job environment variables inside a custom select plugin.However, I am not able to get the same. I am able to get the env in a custom client filter plugin during job submission (sbatch --export) but Ineed the same in slurmctld as well. Is there any way this can be achieved. I have looked over few of the internal DS pointers (job_record, job_details etc)but none seem to provide the required functionality. Any insight is appreciated. Regards,Bhaskar.

2 1

SLURM queue/partition/qos configuration
by vittorio.confuorto＠gmail.com 12 Mar '25

12 Mar '25

Hi everyone, I need some advice on properly configuring my HPC cluster, which consists of three compute nodes (each with 256 cores and 800GB of RAM). The end users belong to three distinct groups: internal, external, and collaborator. I would like the internal users to have full (and exclusive) access to the first two compute nodes. That is, by default, if they submit a job, it should be scheduled on Node1 or Node2, depending on availability (or placed in the queue if necessary). Internal users should also have access to 100 cores and 200GB of RAM on Node3, with this limit applying as a hard cap across all their jobs. The remaining 156 cores and 600GB of RAM on Node3 should be exclusively available to external and collaborator users. Specifically, I would like the following setup: - The resources on Node3 should be split as follows: - 100 cores and 400GB of RAM for external users. - 56 cores and 200GB of RAM for collaborator users. - External users should be able to use the resources dedicated to collaborator users but with lowest priority. This means an external user can submit a job requesting the resources reserved for collaborators, but the job will only start if no collaborator jobs are in the queue. Any job submitted by a collaborator—even after an external user's job—should take priority. The same rule should apply in reverse for collaborators trying to use external resources. - There should be a group/partition/queue that includes all the remaining 156 cores and 600GB of RAM on Node3. Both external and collaborator users should be able to submit jobs to this queue, but a job submitted here should only start if there are no pending jobs in the two dedicated partitions. What is the best way to implement such an architecture? Thanks, everyone, for your support!

1 0

Slurm Job Kill intimation
by Bhaskar Chakraborty 10 Mar '25

10 Mar '25

Hello All, We wish to do some cleanup when user kills a pending slurm job using scancel. Our setup has a client filter plugin which sets some internal variables while interacting to a remote process.This needs to be undone during job termination. Is there a way to do this (using submitter credentials) ? Also, if one is to use a custom select plugin is there a way to inform select that the job is terminated.Presently, we don't find that slurmctld informs select plugin when a pending job is killed. Thanks in advance for any pointers ! Regards,Bhaskar.

1 0

Slurm upgrade using Debian packages
by Matthias Leopold 10 Mar '25

10 Mar '25

Hi, I'm building Slurm Debian packages from SchedMD sources using this tutorial https://www.schedmd.com/slurm/installation-tutorial/. Now I tried upgrading (minor release upgrade within 24.05) using these packages. https://slurm.schedmd.com/upgrades.html tells me to upgrade (a) slurmdbd (b) slurmctld (c) slurmd separately in this order, stopping each service for the upgrade. How can I follow this when the Debian packages have a dependency between slurmdbd + slurmctld that upgrades both packages at the same time? thx Matthias

4 7

allowing spyder-kernels to an interactive session without pam_slurm_adopt and older version of Slurm from OpenHPC repo; parmiko?
by Robert Kudyba 06 Mar '25

06 Mar '25

So the Spyder IDE has this great feature to connect a local/laptop/workstation to a spyder-kernels session <https://docs.spyder-ide.org/current/panes/ipythonconsole.html#using-externa…>. If a user starts an interactive, e.g., srun session, they end up on a compute node which is on one of our older clusters without the pam_slurm_adopt module. Is there a way with a library like Parmiko or this slurmjob <https://github.com/daangeijs/deepops-slurmjob> package? I see for vscode an admin posted the following option <https://github.com/microsoft/vscode-remote-release/issues/1722#issuecomment…> : if [[ -v SSH_AUTH_SOCK ]]; then if [[ ${SSH_AUTH_SOCK} =~ vscode ]]; then [ -f ~/.code-tunnel-env.bash ] && source ~/.code-tunnel-env.bash fi fi Is there a way to work around the restriction of non admins to connect to the spyder-kernel session? Is there anything similar to batchspawner for Jupyterhub <https://github.com/jupyterhub/batchspawner>? Unless anyone knows of an archive for Slurm slurm-ohpc-18.08 that includes pam_adopt?

1 0

Slurm versions 24.11.3 is now available
by Marshall Garey 06 Mar '25

06 Mar '25

We are pleased to announce the availability of Slurm version 24.11.3. 24.11.3 fixes the database cluster ID generation not being random, a regression in which slurmd -G gave no output, a long-standing crash in slurmctld after updating a reservation with an empty nodelist, and some other minor to moderate bugs. Downloads are available at https://www.schedmd.com/downloads.php . -- Marshall Garey Release Management, Support, and Development SchedMD LLC - Commercial Slurm Development and Support > * Changes in Slurm 24.11.3 > ========================== > -- Fix race condition in slurmrestd that resulted in "Requested > data_parser plugin does not support OpenAPI plugin" error being returned > for valid endpoints. > -- If multiple partitions are requested, set the SLURM_JOB_PARTITION > output environment variable to the partition in which the job is running > for salloc and srun in order to match the documentation and the behavior of > sbatch. > -- Fix regression where slurmd -G gives no output. > -- Don't print misleading errors for stepmgr enabled steps. > -- slurmrestd - Avoid connection to slurmdbd for the following > endpoints: > GET /slurm/v0.0.41/jobs > GET /slurm/v0.0.41/job/{job_id} > -- slurmrestd - Avoid connection to slurmdbd for the following > endpoints: > GET /slurm/v0.0.40/jobs > GET /slurm/v0.0.40/job/{job_id} > -- Significantly increase entropy of clusterid portion of the > sluid by seeding the random number generator > -- Avoid changing process name to "watch" from original daemon name. > This could potentially breaking some monitoring scripts. > -- Avoid slurmctld being killed by SIGALRM due to race condition > at startup. > -- Fix slurmctld crash when after updating a reservation with an empty > nodelist. The crash could occur after restarting slurmctld, or if > downing/draining a node in the reservation with the REPLACE or REPLACE_DOWN > flag. > -- Fix race between task/cgroup cpuset and jobacctgather/cgroup. > The first was removing the pid from task_X cgroup directory causing > memory limits to not be applied. > -- srun - Fixed wrongly constructed SLURM_CPU_BIND env variable > that could get propagated to downward srun calls in certain mpi > environments, causing launch failures. > -- slurmrestd - Fix possible memory leak when parsing arrays with > data_parser/v0.0.40. > -- slurmrestd - Fix possible memory leak when parsing arrays with > data_parser/v0.0.41. > -- slurmrestd - Fix possible memory leak when parsing arrays with > data_parser/v0.0.42.

1 0

broken SLURM-PMIX out-of-band communication on v24.11.0 with PMIx v5
by Bertini, Denis Dr. 06 Mar '25

06 Mar '25

Hi there, Using slurm v24.11.0 together with openMPI 5.0.7 built with openpmix v5.0.6 i am facing a systematical crash at process wiring-up phase when launching standard MPI job (OSU benchmarks ) on our new AMD compute nodes ( amd-epyc 9654, 192 phys. cores +HT ) running Rocky Linux 9.4 OS The typical error reads: slurmstepd: error: mpi/pmix_v5: pmixp_p2p_send: ccexe0094 [4]: pmixp_utils.c:469: send failed, rc=1001, exceeded the retry limit slurmstepd: error: mpi/pmix_v5: _slurm_send: ccexe0094 [4]: pmixp_server.c:1581: Cannot send message to /var/spool/slurmd/stepd.slurm.pmix.656.0, size = 46979, hostlist: (null) srun: error: Node failure on ccexe0091 after such a error as you can see the node move to state down It looks like the slurmstep pmix_server can not use the local socket at var/spool/slurmd/stepd.slurm.pmix.job_id.0 for inter-node communication . * On one AMD node ( same SLURM version, same cluster setup ) wiring up works smoothly even at core satuation (192 cores used) * On Intel node (intel,xeon,gold6248r, 48 cores ) wiring-up works even with multiple node without any problem * When the problematic AMD nodes are setup as dynamic node<https://slurm.schedmd.com/dynamic_nodes.html> the wiring-up phase with multiple nodes works perfectly, without any issue Has anybody experienced this kind of problem? Any idea what could be the reason for that? I also add that when the problematic AMD nodes are setup as dynamic node<https://slurm.schedmd.com/dynamic_nodes.html> the wiring-up phase with multiple nodes works perfectly, without any issue Cheers, Denis * --------- Denis Bertini Abteilung: CIT Ort: SB3 2.265a Tel: +49 6159 71 2240 Fax: +49 6159 71 2986 E-Mail: d.bertini(a)gsi.de GSI Helmholtzzentrum für Schwerionenforschung GmbH Planckstraße 1, 64291 Darmstadt, Germany, www.gsi.de Commercial Register / Handelsregister: Amtsgericht Darmstadt, HRB 1528 Managing Directors / Geschäftsführung: Professor Dr. Paolo Giubellino, Dr. Ulrich Breuer, Jörg Blaurock Chairman of the GSI Supervisory Board / Vorsitzender des GSI-Aufsichtsrats: Ministerialdirigent Dr. Volkmar Dietz

1 0

error in logs, does it mean anything?
by Steven Jones 06 Mar '25

06 Mar '25

[2025-03-06T00:26:06.002] error: _add_registered_cluster: trying to register a cluster (poc-cluster) with no remote port [2025-03-06T00:31:06.002] error: _add_registered_cluster: trying to register a cluster (poc-cluster) with no remote port [2025-03-06T00:36:06.001] error: _add_registered_cluster: trying to register a cluster (poc-cluster) with no remote port [2025-03-06T00:41:07.001] error: _add_registered_cluster: trying to register a cluster (poc-cluster) with no remote port regards Steven

1 0

Re: mariadb refusing access
by Steven Jones 05 Mar '25

05 Mar '25

Yes that is set, and my settings are, # slurmDBD info DbdAddr=localhost DbdHost=vuwunicoslurmd3.ods.vuw.ac.nz #DbdHost=localhost DbdPort=6819 SlurmUser=slurm #MessageTimeout=300 DebugLevel=verbose #DefaultQOS=normal,standby LogFile=/var/log/slurm/slurmdbd.log PidFile=/var/run/slurmdbd/slurmdbd.pid #PluginDir=/usr/lib/slurm #PrivateData=accounts,users,usage,jobs #TrackWCKey=yes # # Database info StorageType=accounting_storage/mysql #StorageHost=localhost StorageHost=localhost #StoragePort=1234 #StoragePort=3306 StoragePass=xxxxxxxx StorageUser=slurm StorageLoc=slurm_acct_db #ssj regards Steven ________________________________ From: Sarlo, Jeffrey S <JSarlo(a)Central.UH.EDU> Sent: Thursday, 6 March 2025 7:42 am To: Steven Jones <steven.jones(a)vuw.ac.nz> Subject: RE: [slurm-users] Re: mariadb refusing access You don't often get email from jsarlo(a)central.uh.edu. Learn why this is important<https://aka.ms/LearnAboutSenderIdentification> Do you have the correct password listed in the slurmdbd.conf file? That was toward the bottom of the Wiki that Ole sent. Jeff From: Steven Jones via slurm-users <slurm-users(a)lists.schedmd.com> Sent: Wednesday, March 5, 2025 12:39 PM To: slurm-users(a)lists.schedmd.com; Ole Holm Nielsen <Ole.H.Nielsen(a)fysik.dtu.dk> Subject: [slurm-users] Re: mariadb refusing access Sorry, I dont follow you, Pass word works fine and grant looks OK. [root@vuwunicoslurmd3 ~]# mysql -u slurm -p Enter password: Welcome to the MariaDB monitor. Commands end with ; or \g. Your MariaDB connection id is 208 Server version: 10.11.10-MariaDB MariaDB Server Copyright (c) 2000, 2018, Oracle, MariaDB Corporation Ab and others. Type 'help;' or '\h' for help. Type '\c' to clear the current input statement. MariaDB [(none)]> show grants; +--------------------------------------------------------------------------------------------------------------+ | Grants for slurm@localhost | +--------------------------------------------------------------------------------------------------------------+ | GRANT USAGE ON *.* TO `slurm`@`localhost` IDENTIFIED BY PASSWORD '*2F4EEC707189D8B5E329829B9303058EF7585569' | | GRANT ALL PRIVILEGES ON `slurm_acct_db`.* TO `slurm`@`localhost` | +--------------------------------------------------------------------------------------------------------------+ 2 rows in set (0.000 sec) MariaDB [(none)]> Isnt there something about password encryption / password hash? regards Steven ________________________________ From: Ole Holm Nielsen via slurm-users <slurm-users(a)lists.schedmd.com<mailto:slurm-users@lists.schedmd.com>> Sent: Wednesday, 5 March 2025 9:25 pm To: slurm-users(a)lists.schedmd.com<mailto:slurm-users@lists.schedmd.com> <slurm-users(a)lists.schedmd.com<mailto:slurm-users@lists.schedmd.com>> Subject: [slurm-users] Re: mariadb refusing access On 3/5/25 02:23, Steven Jones via slurm-users wrote: > In the logs I am seeing, > > root@vuwunicoslurmd3 mariadb]# tail -f mariadb.log > 2025-03-04 19:01:32 12565 [Warning] Access denied for user > 'slurm'@'localhost' (using password: YES) > 2025-03-04 19:06:19 12566 [Warning] Access denied for user > 'slurm'@'localhost' (using password: YES) > > However mysql -u slurm -p works just fine so it seems to be a config > error for slurmdbd It seems that you didn't select a suitable slurm user’s database password, see https://apc01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwiki.fysi…<https://urldefense.com/v3/__https:/apc01.safelinks.protection.outlook.com/?…> IHTH, Ole -- slurm-users mailing list -- slurm-users(a)lists.schedmd.com<mailto:slurm-users@lists.schedmd.com> To unsubscribe send an email to slurm-users-leave(a)lists.schedmd.com<mailto:slurm-users-leave@lists.schedmd.com>

1 1

mariadb refusing access
by Steven Jones 05 Mar '25

05 Mar '25

In the logs I am seeing, root@vuwunicoslurmd3 mariadb]# tail -f mariadb.log 2025-03-04 19:01:32 12565 [Warning] Access denied for user 'slurm'@'localhost' (using password: YES) 2025-03-04 19:06:19 12566 [Warning] Access denied for user 'slurm'@'localhost' (using password: YES) However mysql -u slurm -p works just fine so it seems to be a config error for slurmdbd Any ideas please? regards Steven

4 6

Tracking costs - variable costs per partition
by Jeffrey A Dusenberry 05 Mar '25

05 Mar '25

Hello - We're in a similar situation as was described here: https://groups.google.com/g/slurm-users/c/eBDslkwoFio where we want to track (and control) costs on a fairly heterogenous system with different billing weights per partition. The solution proposed seems like it would work rather well, except our use of fairshare seems to interfere with the billing values we would want to use to limit usage based on credits granted. We have PriorityDecayHalfLife set on our system, so that billing value (GrpTRESRaw) seems to drop with time. Is there a way to implement something similar on an otherwise fairshare-based system? Thanks, Jeff

1 0

orted errors
by Berg, Stephen P CIV USN NRL DET SSC MS (USA) 04 Mar '25

04 Mar '25

I'm running a small-ish slurm grid, 87 nodes with various hardware. On a few occasions lately users submitting jobs will get an orted error and the job fails. Try again a few hours later or the next day and the same job runs just fine. Google-fu indicated it might be a DNS issue if for whatever reason a node couldn't figure out the address for other nodes in the job. So I populated the /etc/hosts on each node with a complete listing of all the nodes so there wouldn't be any reliance on DNS. And that very afternoon another job failed with orted. So it seems at least in my case DNS isn't the issue. What's the best way to troubleshoot this when orted fails but doesn't give any sort of error to indicate what the root cause of the failure might be? And I also can't predictably induce the failure, just have to wait until it randomly chokes.

2 1

Slurm versions 24.11.2 and 24.05.6 are now available
by Marshall Garey 04 Mar '25

04 Mar '25

We are pleased to announce the availability of Slurm versions 24.11.2 and 24.05.6. 24.11.2 fixes a variety of minor to major bugs. Fixed regressions include loading non-default QOS on pending jobs from pre-24.11 state, pending jobs displaying QOS=(null) when not explicitly requesting a QOS, running jobs that requested multiple partitions potentially having an incorrect partition when slurmctld is restarted, and burst_buffer.lua failing if slurm.conf is in a non-standard location. This release also fixes a few crashes in slurmctld: crashing when a job that can preempt requests --test-only, crasing when the scheduler evaluates a job on nodes with suspended jobs, and crashing due to a long-standing bug causing a job record without job_resrcs. 24.05.6 fixes sattach with auth/slurm, a slurmrestd crash when using data_parser/v0.0.40, a slurmctld crash when using job suspension, a performance regression for RPCs with large amounts of data, and some other moderate severity bugs. Downloads are available at https://www.schedmd.com/downloads.php . -- Marshall Garey Release Management, Support, and Development SchedMD LLC - Commercial Slurm Development and Support > * Changes in Slurm 24.11.2 > ========================== > -- Fix segfault when submitting --test-only jobs that can preempt. > -- Fix regression introduced in 23.11 that prevented the following > flags from being added to a reservation on an update: > DAILY, HOURLY, WEEKLY, WEEKDAY, and WEEKEND. > -- Fix crash and issues evaluating job's suitability for running in > nodes with already suspended job(s) there. > -- Slurmctld will ensure that healthy nodes are not reported as > UnavailableNodes in job reason codes. > -- Fix handling of jobs submitted to a current reservation with > flags OVERLAP,FLEX or OVERLAP,ANY_NODES when it overlaps nodes with a > future maintenance reservation. When a job submission had a time limit that > overlapped with the future maintenance reservation, it was rejected. Now > the job is accepted but stays pending with the reason "ReqNodeNotAvail, > Reserved for maintenance". > -- pam_slurm_adopt - avoid errors when explicitly setting > some arguments to the default value. > -- Fix qos preemption with PreemptMode=SUSPEND > -- slurmdbd - When changing a user's name update lineage > at the same time. > -- Fix regression in 24.11 in which burst_buffer.lua does not > inherit the SLURM_CONF environment variable from slurmctld and fails to run > if slurm.conf is in a non-standard location. > -- Fix memory leak in slurmctld if select/linear and the > PreemptParameters=reclaim_licenses options are both set in slurm.conf. > Regression in 24.11.1. > -- Fix running jobs, that requested multiple partitions, from > potentially being set to the wrong partition on restart. > -- switch/hpe_slingshot - Fix compatibility with newer cxi > drivers, specifically when specifying disable_rdzv_get. > -- Add ABORT_ON_FATAL environment variable to capture a backtrace > from any fatal() message. > -- Fix printing invalid address in rate limiting log statement. > -- sched/backfill - Fix node state PLANNED not being cleared from > fully allocated nodes during a backfill cycle. > -- select/cons_tres - Fix future planning of jobs with bf_licenses. > -- Prevent redundant "on_data returned rc: Rate limit exceeded, > please retry momentarily" error message from being printed in > slurmctld logs. > -- Fix loading non-default QOS on pending jobs from pre-24.11 state. > -- Fix pending jobs displaying QOS=(null) when not explicitly > requesting a QOS. > -- Fix segfault issue from job record with no job_resrcs > -- Fix failing sacctmgr delete/modify/show account operations > with where clauses. > -- Fix regression in 24.11 in which Slurm daemons started catching > several SIGTSTP, SIGTTIN and SIGUSR1 signals and ignored them, while before > they were not ignoring them. This also caused slurmctld to not being > able to shutdown after a SIGTSTP because slurmscriptd caught the signal > and stopped while slurmctld ignored it. Unify and fix these situations and > get back to the previous behavior for these signals. > -- Document that SIGQUIT is no longer ignored by slurmctld, > slurmdbd, and slurmd in 24.11. As of 24.11.0rc1, SIGQUIT is identical to > SIGINT and SIGTERM for these daemons, but this change was not documented. > -- Fix not considering nodes marked for reboot without ASAP > in the scheduler. > -- Remove the boot^ state on unexpected node reboot after > return to service. > -- Do not allow new jobs to start on a node which is being rebooted > with the flag nextstate=resume. > -- Prevent lower priority job running after cancelling an ASAP reboot. > -- Fix srun jobs starting on nextstate=resume rebooting nodes. > > * Changes in Slurm 24.05.6 > ========================== > -- data_parser/v0.0.40 - Prevent a segfault in the slurmrestd when > dumping data with v0.0.40+complex data parser. > -- Fix sattach when using auth/slurm. > -- scrun - Add support '--all' argument for kill subcommand. > -- Fix performance regression while packing larger RPCs. > -- Fix crash and issues evaluating job's suitability for running in > nodes with already suspended job(s) there. > -- Fixed a job requeuing issue that merged job entries into the > same SLUID when all nodes in a job failed simultaneously. > -- switch/hpe_slingshot - Fix compatibility with newer cxi > drivers, specifically when specifying disable_rdzv_get. > -- Add ABORT_ON_FATAL environment variable to capture a backtrace > from any fatal() message.

7 6

Re: how to set slurmdbd.conf if using two slurmdb node with HA database?
by taleintervenor＠sjtu.edu.cn 27 Feb '25

27 Feb '25

Do you mean the second configuration scheme? I think configuring `dbdhost=localhost` is the same as configuring ` DbdAddr =nodeC` and ` DbdAddr =nodeD` on the two nodes respectively. The key point is whether we should set the DbdBackupHost option and how it work? 发件人: Daniel Letai <dani(a)letai.org.il> 发送时间: 2025年2月19日 18:21 收件人: slurm-users(a)lists.schedmd.com 主题: [slurm-users] Re: how to set slurmdbd.conf if using two slurmdb node with HA database? I'm not sure it will work, didn't test it, but could you just do `dbdhost=localhost` to solve this? On 18/02/2025 11:59, hermes via slurm-users wrote: The deployment scenario is as follows: nodeA nodeB (slurmctld) (backup slurmctld) | \-------------------------------/ | | / \ | nodeC nodeD (slurmdbd) (backup slurmdbd) (mysql) <--multi master replica--> (mysql) Since the database is multi-master replicated, the slurmdbd should only talk to the mysql on its own node. In such case, how should we set the slurmdbd.conf? The conf file contains options “DbdAddr”, “DbdHost” and “DbdBackupHost”. Should they be consistent between nodeA-2 and nodeB-2? Such as: DbdAddr = nodeC | DbdAddr = nodeC DbdHost = nodeC | DbdHost = nodeC DbdBackupHost = nodeD | DbdBackupHost = nodeD StorageHost = nodeC | StorageHost = nodeD Or maybe just set different conf and don’t use the “DbdBackupHost” like: DbdAddr = nodeC | DbdAddr = nodeD DbdHost = nodeC | DbdHost = nodeD StorageHost = nodeC | StorageHost = nodeD I’m quite confused about the usage of DbdAddr and DbdHost. What is the difference between them and why only DbdHost has the backup one? Another confusing point is how DbdBackupHost work. I guess It is slurmctld that is responsible for selecting the available slurmdbd. Since the slurm.conf already contains “AccountingStorageHost” and “AccountingStorageBackupHost”, why we need set backupdbd again on slurmdbd side?

4 10

Limit CPUs per job (but not per user, partition or node)
by Herbert Fruchtl 26 Feb '25

26 Feb '25

We have a cluster with multi-core nodes (168) that can be shared by multiple jobs at the same time. How do I configure a partition such that it only accepts jobs requesting up to (say) 8 cores, but will run multiple jobs at the same time? The following is apparently not working: PartitionName=debug Nodes=node01 MaxTime=02:00:00 DefMemPerCPU=1000 MaxCPUsPerNode=8 Default=NO It allows one job using 8 cores, but a second one will not start because the limit is apparently for the partition as a whole. Thanks in advance, Herbert -- Herbert Fruchtl (he/him) Senior Scientific Computing Officer / HPC Administrator School of Chemistry, IT Services University of St Andrews -- The University of St Andrews is a charity registered in Scotland: No SC013532

3 2

Re: MinTRES in QoS and power saving
by Patryk Bełzak 26 Feb '25

26 Feb '25

Hi, there was this issue raised some time ago: https://www.mail-archive.com/slurm-users@lists.schedmd.com/msg10799.html We're experiencing exactly the same issue now with GPU nodes in power saving, some (but not all) jobs doesn't start because of that, and it's annoying users - badly. Anyone experienced something similiar? Or maybe Stefan has a solution? (we're on version 24.05.5) Best regards Patryk Belzak. -- HPC Administrator Wroclaw Centre for Networking and Supercomputing

1 0

Shard conf weirdness
by Reed Dier 24 Feb '25

24 Feb '25

Hoping someone can help me pin down the weirdness I’m experiencing. There are actually two issues, I’ve run into, the root issue, and then something odd when trying to work around the root issue. v23.11.10 - Ubuntu 22.04 - slurm-smd debs built <https://slurm.schedmd.com/archive/slurm-23.11.10/quickstart_admin.html#debu…> from the tarball. I have 2 slurmctld daemons, and 1 slurmdbd daemon. The slurm.conf is consistent across the cluster, all ctld,dbd,slurmd, have the same shasum hash. The backup slurmctld does not appear to like the gres configured on my gpu nodes, despite the primary slurmctld having no issues. When failing over with scontrol takeover, I get the following log messages on the secondary slurmctld, where it complains about a reported, but not configured typed gres/shard. > [2025-02-21T05:02:02.017] error: Setting node $HOST2 state to INVAL with reason:gres/shard type (p100) reported but not configured > [2025-02-21T05:02:02.018] drain_nodes: node $HOST2 state set to DRAIN > [2025-02-21T05:02:02.018] error: _slurm_rpc_node_registration node=$HOST2: Invalid argument > [2025-02-21T05:02:02.020] error: Setting node $HOST3 state to INVAL with reason:gres/shard type (p40) reported but not configured > [2025-02-21T05:02:02.020] drain_nodes: node $HOST3 state set to DRAIN > [2025-02-21T05:02:02.020] error: _slurm_rpc_node_registration node=$HOST3: Invalid argument > [2025-02-21T05:02:02.023] error: Setting node $HOST1 state to INVAL with reason:gres/shard type (t4) reported but not configured > [2025-02-21T05:02:02.020] drain_nodes: node $HOST1 state set to DRAIN > [2025-02-21T05:02:02.023] error: _slurm_rpc_node_registration node=$HOST1: Invalid argument And looking at those hosts in the slurm.conf, the shards are not typed, but generic. > NodeName=$HOST1 [SNIP] State=UNKNOWN Gres=gpu:t4:2,shard:8 > NodeName=$HOST2 [SNIP] State=UNKNOWN Gres=gpu:p100:2,shard:8 > NodeName=$HOST3 [SNIP] State=UNKNOWN Gres=gpu:p40:2,gpu:p100:2,shard:16 I should also point out at this point that my gres.conf is just AutoDetect=nvml. I am not explicitly mapping any devices. So this is problem 1, the slurmctld behaving differently between daemons, where the primary has zero issue with the configuration, but the secondary balking and draining the nodes consistently. Moving on to problem 2, which is me trying to solve the issue, and running into a different issue. I decided to try to add the typed gres/shards to both the AccountingStorageTRES, and the NodeName lists. > NodeName=$HOST1 [SNIP] State=UNKNOWN Gres=gpu:t4:2,shard:t4:8 > NodeName=$HOST2 [SNIP] State=UNKNOWN Gres=gpu:p100:2,shard:p100:8 > NodeName=$HOST3 [SNIP] State=UNKNOWN Gres=gpu:p40:2,gpu:p100:2,shard:p40:8,shard:p100:8 Now the secondary slurmctld is happy and no longer immediately drains the gpu nodes for the invalid gres. However, the node ($HOST3) with a mixed set of gpus is now complaining. > [2025-02-23T21:41:05.311] gpu/nvml: _get_system_gpu_list_nvml: 4 GPU system device(s) detected > [2025-02-23T21:41:05.311] fatal: _build_shared_list: bad configuration, multiple configurations without "File" No matter what I tried, any line of the NodeName with multiple shard:$type:$num generates the _build_shared_list error above. I tried re-ordering the list as gpu,shard,gpu,shard to no avail, I tried having untyped shards, with typed shards after that, but then it complained about too many shards (double the intended number, $untyped + $typed = too many) So my current solution is to have the slurm conf in sync everywhere BUT the host(s) with multiple gpu models in the same host, where I had to revert to untyped gres on those slurmd’s, but on the slurmctlds they are typed gres. Hopefully I’ve done a decent job of explaining the corner case, enough that someone can point me in the direction of figuring out whats going on, and what is the “correct” way of doing this. I tried increasing the slurmd and slurmctld logging to debug2, to nothing that stood out beyond what was already gathered. > [2025-02-23T21:46:44.895] debug: GRES[shard] Type:p100 Count:8 Cores(88):(null) Links:(null) Flags:HAS_TYPE File:(null) UniqueId:(null) > [2025-02-23T21:46:44.895] debug: GRES[shard] Type:p40 Count:8 Cores(88):(null) Links:(null) Flags:HAS_TYPE File:(null) UniqueId:(null) > [2025-02-23T21:46:44.895] fatal: _build_shared_list: bad configuration, multiple configurations without "File" Any ideas are greatly appreciated, Reed

1 0

Plese help [CPUs=24 Boards=1 SocketsPerBoard=1 CoresPerSocket=16 ThreadsPerCore=1]
by Hugo Solís 24 Feb '25

24 Feb '25

I am installing slurm on a small cluster of 3 identical computers at the university. It turns out that I get the message when starting: error: NodeNames=hugo-big-1 CPUs=24 match no Sockets, Sockets*CoresPerSocket or Sockets*CoresPerSocket*ThreadsPerCore. Resetting CPUs. error: NodeNames=hugo-big-2 CPUs=24 match no Sockets, Sockets*CoresPerSocket or Sockets*CoresPerSocket*ThreadsPerCore. Resetting CPUs. error: NodeNames=hugo-big-3 CPUs=24 match no Sockets, Sockets*CoresPerSocket or Sockets*CoresPerSocket*ThreadsPerCore. Resetting CPUs. if I run sudo slurmd -C I get NodeName=hugo-big-1 CPUs=24 Boards=1 SocketsPerBoard=1 CoresPerSocket=16 ThreadsPerCore=1 RealMemory=128586 reading many post it seems there is a bug starting in version 18. However I had modify the slurm.conf file NodeName=hugo-big-1 CPUs=24 Boards=1 SocketsPerBoard=1 CoresPerSocket=16 ThreadsPerCore=1 RealMemory=128586 Gres=gpu:nvidia_geforce_rtx_3090:2 to several different version where I do not specify CPUs or sockets and I failed everytime. I do need help, any idea? warm regards Hugo Solis BTW. if I run lscpu I got that any product of Thread, cores or sockets do not result in 24. Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Address sizes: 46 bits physical, 48 bits virtual Byte Order: Little Endian CPU(s): 24 On-line CPU(s) list: 0-23 Vendor ID: GenuineIntel Model name: 12th Gen Intel(R) Core(TM) i9-12900K CPU family: 6 Model: 151 Thread(s) per core: 2 Core(s) per socket: 16 Socket(s): 1 Stepping: 2 CPU max MHz: 5200.0000 CPU min MHz: 800.0000 BogoMIPS: 6374.40 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_ti mer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault cat_l2 cdp_l2 ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdt_a rdseed adx smap clflushopt clwb intel_pt sha_ni xsaveopt xsavec xgetbv1 xsaves split_lock_detect user_shstk avx_vnni dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_e pp hwp_pkg_req hfi vnmi umip pku ospke waitpkg gfni vaes vpclmulqdq tme rdpid movdiri movdir64b fsrm md_clear serialize pconfig arch_lbr ibt flush_l1d arch_capabilities Virtualization features: Virtualization: VT-x Caches (sum of all): L1d: 640 KiB (16 instances) L1i: 768 KiB (16 instances) L2: 14 MiB (10 instances) L3: 30 MiB (1 instance) NUMA: NUMA node(s): 1 NUMA node0 CPU(s): 0-23

2 2

Please help - Building Slurm-24.11.1 Failed
by Zhang, Yuan 23 Feb '25

23 Feb '25

Hello, I got errors about missing perl modules when building slurm24.11.1 rpm packages. Has anyone seen this error before? And how to fix it? Here are the error messages: -- Processing files: slurm-perlapi-24.11.1-1.el8.x86_64 error: File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/Slurm/Bitstr.pm error: File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/Slurm/Constant.pm error: File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/Slurm/Hostlist.pm error: File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/auto/Slurm/Slurm.so error: File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/auto/Slurmdb/Slurmdb.so error: File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/auto/Slurmdb/autosplit.ix RPM build errors: Macro expanded in comment on line 31: %_prefix path install path for commands, libraries, etc. Macro expanded in comment on line 240: %define _unpackaged_files_terminate_build 0 File listed twice: /usr/lib/.build-id/49/9394dac8b96530e1a29d0e35608ef9acedec70 File listed twice: /usr/lib/.build-id/88/4e3a854911920ae0c77c4b32d5af70dad538a9 File listed twice: /usr/lib/.build-id/c1/72dbf51fe1c17868a4e0dbf0aa55a7923f6d02 File listed twice: /usr/lib/.build-id/dc/67c93438d36986971444d218350d27b413c2cb File listed twice: /usr/lib/.build-id/ee/e79f70a62bf62e74705a75daa87d26857452ef Deprecated external dependency generator is used! File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/Slurm/Bitstr.pm File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/Slurm/Constant.pm File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/Slurm/Hostlist.pm File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/auto/Slurm/Slurm.so File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/auto/Slurmdb/Slurmdb.so File not found: /home/gdhpcgroup/yxz006/rpmbuild/BUILDROOT/slurm-24.11.1-1.el8.x86_64/usr/lib64/perl5/auto/Slurmdb/autosplit.ix -- Thanks! Yuan -- Yuan Zhang Senior HPC System Programmer Nationwide Children’s Hospital Research Institute Columbus, Ohio 43215

3 5

Running SLURM in a laptop
by Antonio Rius 19 Feb '25

19 Feb '25

For testing purposes I am using SLURM installed in a lenovo laptop running ubuntu 24-04. I am trying this configuration as part of my slurm learning process before using a real HPC facility. In my trial I want to submit an array of 9 independent processes. The CPU information of such machine is: Total Logical CPUs: 16 Physical Cores: 12 Threads per Core: 2 ================================= and the slurm script (attached) that I am using contain the sentences : #SBATCH --job-name=process_pkl #SBATCH --array=1-9 #SBATCH --output=/tmp/process_pkl_%A_%a.out #SBATCH --error=/tmp/process_pkl_%A_%a.err #SBATCH --ntasks=1 #SBATCH --cpus-per-task=1 #SBATCH --mem=2G =================================== but in practice, using htop, only 4 processes are running concurrently, and there are other 12 CPUs without activity. My question is: What am I missing in my setup?

4 3

slurmrestd health check
by Hagdorn, Magnus Karl Moritz 19 Feb '25

19 Feb '25

Hi there, we use haproxy to distribute SLURM REST API requests to multiple instances of slurmrestd. For the haproxy we need a health check. At the moment we are just checking that we get a 401 status. This works but we are ending up with a lot of noise in the log files. It would be very nice if there was an unauthenticated REST API endpoint that can be used to check the daemon is up. Has anybody solved this issue? Refards magnus -- Dr. Magnus Hagdorn Charité – Universitätsmedizin Berlin Geschäftsbereich IT | Scientific Computing

1 0

/etc/passwd sync?
by mark.w.moorcroft＠ama-inc.com 19 Feb '25

19 Feb '25

If you set up slurm elastic cloud in EC2 without LDAP, what is the recommended method for sync of the passwd/group files? Is this necessary to get openmpi jobs to run. I would swear I had this working last week without synced passwd on two nodes. But thinking about it now I'm not sure how this could have worked. My home directories are in an NFS mount, but the user accounts don't exist on the node AMI. I'm using ansible/packer to manage the AMI's. When I ran OpenHPC / Slurm on bare metal there was a sync process. This is my first AWS Slurm cluster rodeo. I can't use the Amazon Parallel Computing tools because we are forced to be in GovCloud. I started with "ClusterInTheCloud", but it's all 4 years old, and semi-broken out of the box. My manager had me ditch a lot of it (including LDAP). So I'm building out a fork that is getting heavily modded for our situation. An ORTE daemon has unexpectedly failed after launch and before communicating back to mpirun. This could be caused by a number of factors, including an inability to create a connection back to mpirun due to a lack of common network interfaces and/or no route found between them. Please check network connectivity

4 4

how to set slurmdbd.conf if using two slurmdb node with HA database?
by taleintervenor＠sjtu.edu.cn 19 Feb '25

19 Feb '25

The deployment scenario is as follows: nodeA nodeB (slurmctld) (backup slurmctld) | \-------------------------------/ | | / \ | nodeC nodeD (slurmdbd) (backup slurmdbd) (mysql) <--multi master replica--> (mysql) Since the database is multi-master replicated, the slurmdbd should only talk to the mysql on its own node. In such case, how should we set the slurmdbd.conf? The conf file contains options "DbdAddr", "DbdHost" and "DbdBackupHost". Should they be consistent between nodeA-2 and nodeB-2? Such as: DbdAddr = nodeC | DbdAddr = nodeC DbdHost = nodeC | DbdHost = nodeC DbdBackupHost = nodeD | DbdBackupHost = nodeD StorageHost = nodeC | StorageHost = nodeD Or maybe just set different conf and don't use the "DbdBackupHost" like: DbdAddr = nodeC | DbdAddr = nodeD DbdHost = nodeC | DbdHost = nodeD StorageHost = nodeC | StorageHost = nodeD I'm quite confused about the usage of DbdAddr and DbdHost. What is the difference between them and why only DbdHost has the backup one? Another confusing point is how DbdBackupHost work. I guess It is slurmctld that is responsible for selecting the available slurmdbd. Since the slurm.conf already contains "AccountingStorageHost" and "AccountingStorageBackupHost", why we need set backupdbd again on slurmdbd side?

2 1

Run only one time on a node
by John Hearns 19 Feb '25

19 Feb '25

I am running single node tests on a cluster. I can select named nodes using the -2 flag with sbatch. However - if I want to submit perhaps 20 test jobs is there any smart way to run only one time on a node? I know I could touch a file with the hostname and test for that file. I am just wondering if there is a smarter way to do this.\ I should point out that the tests take a few minutes to run, so if a node finishes a test run it could become idle and run the tests again.

3 2

Create filenames based on slurm hosts
by John Hearns 15 Feb '25

15 Feb '25

I am working on power logging of a GPU cluster I am working with. I am running jobs on multiple hosts. I wanst to create a file , one for each host, which has a unique filename containing the host name. Something like clush -w $SLURM_JOB_NODELIST "touch file$(hostname)" My foo is weak today. Help me Ole Wan Neilsen or any $Jedi

3 3

avoid using same GPU by the interactive job
by navin srivastava 13 Feb '25

13 Feb '25

hi, facing an issue in my environment where the batch job and the interactive job use the same gpu. Each server has 2 gpu. When 2 batch jobs are running it works fine and use the 2 different gpu's. but if one batch job is running and another job is submitted interactively then it uses the same GPU . Is there a way to avoid this? GresTypes=gpu NodeName=node[01-02] NodeAddr=node[01-02] CPUs=48 Boards=1 SocketsPerBoard=2 CoresPerSocket=24 ThreadsPerCore=1 TmpDisk=6000000 RealMemory=515634 Feature=A100 Gres=gpu:2 PartitionName=onprem Nodes=node[01-10] Default=YES MaxTime=21-00:00:00 DefaultTime=3-00:00:00 State=UP Shared=YES OverSubscribe=NO gres.conf: Name=gpu File=/dev/nvidia0 Name=gpu File=/dev/nvidia1 Any suggestions on this. Regards Navin

3 3

error no error
by Ricardo Román-Brenes 13 Feb '25

13 Feb '25

Hello. Could someone enlighten me as to what this error message is? Feb 13 10:02:00 gpu1 slurmd[573705]: slurmd: error: slurm_msg_sendto: address:port=192.168.9.1:36698 msg_type=8001: No error

1 0

Re: /etc/passwd sync?
by Mark W. Moorcroft 13 Feb '25

13 Feb '25

I made some progress without need for the /etc/passwd sync. Sbatch is working fine on multi-node jobs it appears. Now only salloc runs fail, which I guess is expected behavior without user account sync and ssh key setup. On bare metal OpenHPC, warewulf handles all that for me. So presumably I have MPI keys, but not user account sync and keys. Again, on OpenHPC they populated keys on account creation, and passwd was synced to the boot image. Can anyone tell me what this is about? slurmctld[46449]: slurmctld: error: failed to generate json for resume job/node list Mark Moorcroft Senior Linux Administrator Analytical Mechanics Associates

1 0

Re: /etc/passwd sync?
by Mark W. Moorcroft 12 Feb '25

12 Feb '25

The only service found in gov cloud with a search for “parallel” is Batch. And the awscli commands are absent there as well. In commercial cloud “parallel” yields many services. Mark Moorcroft Senior Linux Administrator Analytical Mechanics Associates e. mark.w.moorcroft(a)ama-inc.com<mailto:mark.w.moorcroft@ama-inc.com> w. ama-inc.com<https://ama-inc.com/> t. +1 408-290-5455 _____________________________________ Confidentiality Notice: This e-mail is intended only for the addressee named above. It contains information that is privileged, confidential or otherwise protected from use and disclosure. If you are not the intended recipient, you are hereby notified that any review, disclosure, copying, or dissemination of this transmission, or taking of any action in reliance on its contents, or other use is strictly prohibited. If you have received this transmission in error, please reply to the sender listed above immediately and permanently delete this message from your inbox. Thank you for your cooperation.

2 1

2025

2024