- slurm-users - lists.schedmd.com

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