- slurm-users - lists.schedmd.com

slurmctld HA ; backup controller doesn't schedule and start any job
by hiromasa.watanabe＠gmail.com 09 Apr '25

09 Apr '25

slurmctld HA ; backup controller doesn't schedule and start any job Hi all, I am trying a slurmctld HA configuration on two servers, using slurm version 22.05.9 of AlmaLinux 9.4. The problem is, after stopping the primary slurmctld and slurmdbd, when I submit a job with sbatch while backup slurmctld and slurmdbd are running, the job will be pending with Reason=None, and it will not be scheduled and will not start running. $ squeue JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON) 43 cpu_multi job hpc PD 0:00 1 (None) Why won't my job start running? What should I change to get the job to start running? The slurmctld.log and configuration files are shown below. Backup's slurmctld.log： [2025-04-09T15:31:17.000] debug3: Heartbeat at 1744180276 [2025-04-09T15:31:18.000] debug3: Heartbeat at 1744180277 [2025-04-09T15:31:19.022] debug3: Heartbeat at 1744180279 [2025-04-09T15:31:19.605] debug2: Processing RPC: REQUEST_SUBMIT_BATCH_JOB from UID=1000 [2025-04-09T15:31:19.605] debug3: _set_hostname: Using auth hostname for alloc_node: compute1 [2025-04-09T15:31:19.605] debug3: JobDesc: user_id=1000 JobId=N/A partition=cpu_multi name=job [2025-04-09T15:31:19.605] debug3: cpus=2-4294967294 pn_min_cpus=1 core_spec=-1 [2025-04-09T15:31:19.605] debug3: Nodes=1-[1] Sock/Node=65534 Core/Sock=65534 Thread/Core=65534 [2025-04-09T15:31:19.605] debug3: pn_min_memory_job=18446744073709551615 pn_min_tmp_disk=-1 [2025-04-09T15:31:19.605] debug3: immediate=0 reservation=(null) [2025-04-09T15:31:19.605] debug3: features=(null) batch_features=(null) cluster_features=(null) prefer=(null) [2025-04-09T15:31:19.605] debug3: req_nodes=(null) exc_nodes=(null) [2025-04-09T15:31:19.605] debug3: time_limit=-1--1 priority=-1 contiguous=0 shared=-1 [2025-04-09T15:31:19.605] debug3: kill_on_node_fail=-1 script=#! /bin/bash #SBATCH -p cpu_multi #SBATC... [2025-04-09T15:31:19.605] debug3: argv="./twocore.sh" [2025-04-09T15:31:19.605] debug3: environment=SHELL=/bin/bash,PYENV_SHELL=bash,HISTCONTROL=ignoredups,... [2025-04-09T15:31:19.605] debug3: stdin=/dev/null stdout=/misc/home/hpc/slurmtest/twocore_%J.out stderr=(null) [2025-04-09T15:31:19.605] debug3: work_dir=/misc/home/hpc/slurmtest alloc_node:sid=compute1:281600 [2025-04-09T15:31:19.605] debug3: power_flags= [2025-04-09T15:31:19.605] debug3: resp_host=(null) alloc_resp_port=0 other_port=0 [2025-04-09T15:31:19.605] debug3: dependency=(null) account=(null) qos=(null) comment=(null) [2025-04-09T15:31:19.605] debug3: mail_type=0 mail_user=(null) nice=0 num_tasks=2 open_mode=0 overcommit=-1 acctg_freq=(null) [2025-04-09T15:31:19.605] debug3: network=(null) begin=Unknown cpus_per_task=1 requeue=-1 licenses=(null) [2025-04-09T15:31:19.605] debug3: end_time= signal=0@0 wait_all_nodes=-1 cpu_freq= [2025-04-09T15:31:19.605] debug3: ntasks_per_node=-1 ntasks_per_socket=-1 ntasks_per_core=-1 ntasks_per_tres=-1 [2025-04-09T15:31:19.605] debug3: mem_bind=0:(null) plane_size:65534 [2025-04-09T15:31:19.605] debug3: array_inx=(null) [2025-04-09T15:31:19.605] debug3: burst_buffer=(null) [2025-04-09T15:31:19.605] debug3: mcs_label=(null) [2025-04-09T15:31:19.605] debug3: deadline=Unknown [2025-04-09T15:31:19.605] debug3: bitflags=0x1a00c000 delay_boot=4294967294 [2025-04-09T15:31:19.605] debug3: assoc_mgr_fill_in_user: found correct user: hpc(1000) [2025-04-09T15:31:19.605] debug5: assoc_mgr_fill_in_assoc: looking for assoc of user=hpc(1000), acct=hpc, cluster=cluster, partition=cpu_multi [2025-04-09T15:31:19.605] debug3: assoc_mgr_fill_in_assoc: found correct association of user=hpc(1000), acct=hpc, cluster=cluster, partition=cpu_multi to assoc=16 acct=hpc [2025-04-09T15:31:19.605] debug3: found correct qos [2025-04-09T15:31:19.607] debug2: priority/multifactor: priority_p_set: initial priority for job 44 is 33 [2025-04-09T15:31:19.607] debug2: found 1 usable nodes from config containing compute1 [2025-04-09T15:31:19.607] debug2: found 1 usable nodes from config containing compute2 [2025-04-09T15:31:19.607] debug3: _pick_best_nodes: JobId=44 idle_nodes 2 share_nodes 2 [2025-04-09T15:31:19.607] debug2: select/cons_tres: select_p_job_test: evaluating JobId=44 [2025-04-09T15:31:19.607] debug2: sched: JobId=44 allocated resources: NodeList=(null) [2025-04-09T15:31:19.607] _slurm_rpc_submit_batch_job: JobId=44 InitPrio=33 usec=2490 [2025-04-09T15:31:19.608] debug3: create_mmap_buf: loaded file `/var/spool/slurm/ctld/job_state` as buf_t [2025-04-09T15:31:19.609] debug3: Writing job id 45 to header record of job_state file [2025-04-09T15:31:21.000] debug3: Heartbeat at 1744180280 [2025-04-09T15:31:21.257] debug2: _slurm_connect: failed to connect to 192.168.56.11:6817: Connection refused [2025-04-09T15:31:21.257] debug2: Error connecting slurm stream socket at 192.168.56.11:6817: Connection refused [2025-04-09T15:31:22.000] debug3: Heartbeat at 1744180282 [2025-04-09T15:31:24.000] debug3: Heartbeat at 1744180283 [2025-04-09T15:31:25.004] debug3: Heartbeat at 1744180285 [2025-04-09T15:31:27.001] debug3: Heartbeat at 1744180287 [2025-04-09T15:31:29.000] debug3: Heartbeat at 1744180288 [2025-04-09T15:31:30.000] debug3: Heartbeat at 1744180289 [2025-04-09T15:31:31.006] debug3: Heartbeat at 1744180291 [2025-04-09T15:31:32.822] debug2: _slurm_connect: failed to connect to 192.168.56.11:6817: Connection refused [2025-04-09T15:31:32.822] debug2: Error connecting slurm stream socket at 192.168.56.11:6817: Connection refused [2025-04-09T15:31:33.007] debug3: Heartbeat at 1744180293 [2025-04-09T15:31:35.000] debug3: Heartbeat at 1744180294 [2025-04-09T15:31:36.002] debug3: Heartbeat at 1744180296 [2025-04-09T15:31:36.395] debug2: select/cons_tres: select_p_job_test: evaluating JobId=43 [2025-04-09T15:31:36.395] debug2: select/cons_tres: select_p_job_test: evaluating JobId=44 [2025-04-09T15:31:38.000] debug3: Heartbeat at 1744180297 [2025-04-09T15:31:38.497] debug2: Performing purge of old job records [2025-04-09T15:31:39.000] debug3: Heartbeat at 1744180298 [2025-04-09T15:31:40.000] debug3: Heartbeat at 1744180300 [2025-04-09T15:31:40.655] debug2: Testing job time limits and checkpoints [2025-04-09T15:31:42.000] debug3: Heartbeat at 1744180301 [2025-04-09T15:31:43.000] debug3: Heartbeat at 1744180302 slurm.conf: ClusterName=cluster SlurmctldHost=gateway1 #Primary(192.168.56.11) SlurmctldHost=gateway2 #Backup(192.168.56.12) MpiDefault=pmix ProctrackType=proctrack/cgroup PrologFlags=Contain ReturnToService=0 SlurmctldPidFile=/var/run/slurm/slurmctld.pid SlurmctldPort=6817 SlurmdPidFile=/var/run/slurm/slurmd.pid SlurmdPort=6818 SlurmdSpoolDir=/var/spool/slurm/d SlurmUser=slurm StateSaveLocation=/var/spool/slurm/ctld SwitchType=switch/none TaskEpilog=/etc/slurm/taskepilog.sh TaskPlugin=task/cgroup,task/affinity TaskProlog=/etc/slurm/taskprolog.sh InactiveLimit=0 KillWait=30 MinJobAge=300 SlurmctldTimeout=10 SlurmdTimeout=300 Waittime=0 DefMemPerCPU=32 SchedulerType=sched/builtin SelectType=select/cons_tres SelectTypeParameters=CR_Core_Memory PriorityType=priority/multifactor PriorityWeightPartition=1000 AccountingStorageHost=gateway1 AccountingStorageBackupHost=gateway2 AccountingStorageType=accounting_storage/slurmdbd AccountingStoreFlags=job_comment JobCompType=jobcomp/none JobAcctGatherFrequency=30 JobAcctGatherType=jobacct_gather/cgroup SlurmctldDebug=debug5 SlurmctldLogFile=/var/log/slurm/slurmctld.log SlurmdDebug=info SlurmdLogFile=/var/log/slurm/slurmd.log NodeName=compute1 CPUs=4 Boards=1 SocketsPerBoard=1 CoresPerSocket=4 ThreadsPerCore=1 RealMemory=3900 Weight=1 NodeName=compute2 CPUs=4 Boards=1 SocketsPerBoard=1 CoresPerSocket=4 ThreadsPerCore=1 RealMemory=3900 Weight=1 PartitionName=gpu_single Nodes=ALL PriorityJobFactor=30 MaxTime=INFINITE State=UP Default=YES PartitionName=cpu_single Nodes=ALL PriorityJobFactor=10 MaxTime=INFINITE State=UP PartitionName=cpu_multi Nodes=ALL MaxTime=INFINITE State=UP slurmdbd.conf: AuthType=auth/munge DebugLevel=4 DbdHost=gateway1 DbdBackupHost=gateway2 LogFile=/var/log/slurm/slurmdbd.log PidFile=/var/run/slurm/slurmdbd.pid PurgeEventAfter=1month PurgeJobAfter=1month PurgeResvAfter=1month PurgeStepAfter=1month PurgeSuspendAfter=1month PurgeTXNAfter=1month PurgeUsageAfter=1month SlurmUser=slurm StorageType=accounting_storage/mysql StorageHost=gateway1 StoragePass=mypassword StorageUser=slurm Best regards, Hiro

1 0

Slurm version 24.11.4 is now available
by Marshall Garey 08 Apr '25

08 Apr '25

We are pleased to announce the availability of Slurm version 24.11.4. This release fixes a variety of major to minor severity bugs. Some edge cases that caused jobs to pend forever are fixed. Notable stability issues that are fixed include: * slurmctld crashing upon receiving a certain heterogeneous job submission. * slurmd crashing after a communications failure with a slurmstepd. * A variety of race conditions related to receiving and processing connections, including one that resulted in the slurmd ignoring new RPC connections. Downloads are available at https://www.schedmd.com/downloads.php . -- Marshall Garey Release Management, Support, and Development SchedMD LLC - Commercial Slurm Development and Support > -- slurmctld,slurmrestd - Avoid possible race condition that could have caused > process to crash when listener socket was closed while accepting a new > connection. > -- slurmrestd - Avoid race condition that could have resulted in address > logged for a UNIX socket to be incorrect. > -- slurmrestd - Fix parameters in OpenAPI specification for the following > endpoints to have "job_id" field: > GET /slurm/v0.0.40/jobs/state/ > GET /slurm/v0.0.41/jobs/state/ > GET /slurm/v0.0.42/jobs/state/ > GET /slurm/v0.0.43/jobs/state/ > -- slurmd - Fix tracking of thread counts that could cause incoming > connections to be ignored after burst of simultaneous incoming connections > that trigger delayed response logic. > -- Stepmgr - Avoid unnecessary SRUN_TIMEOUT forwarding to stepmgr. > -- Fix jobs being scheduled on higher weighted powered down nodes. > -- Fix how backfill scheduler filters nodes from the available nodes based on > exclusive user and mcs_label requirements. > -- acct_gather_energy/{gpu,ipmi} - Fix potential energy consumption adjustment > calculation underflow. > -- acct_gather_energy/ipmi - Fix regression introduced in 24.05.5 (which > introduced the new way of preserving energy measurements through slurmd > restarts) when EnergyIPMICalcAdjustment=yes. > -- Prevent slurmctld deadlock in the assoc mgr. > -- Fix memory leak when RestrictedCoresPerGPU is enabled. > -- Fix preemptor jobs not entering execution due to wrong calculation of > accounting policy limits. > -- Fix certain job requests that were incorrectly denied with node > configuration unavailable error. > -- slurmd - Avoid crash due when slurmd has a communications failure with > slurmstepd. > -- Fix memory leak when parsing yaml input. > -- Prevent slurmctld from showing error message about PreemptMode=GANG being a > cluster-wide option for `scontrol update part` calls that don't attempt to > modify partition PreemptMode. > -- Fix setting GANG preemption on partition when updating PreemptMode with > scontrol. > -- Fix CoreSpec and MemSpec limits not being removed from previously > configured slurmd. > -- Avoid race condition that could lead to a deadlock when slurmd, slurmstepd, > slurmctld, slurmrestd or sackd have a fatal event. > -- Fix jobs using --ntasks-per-node and --mem keep pending forever when the > requested mem divided by the number of cpus will surpass the configured > MaxMemPerCPU. > -- slurmd - Fix address logged upon new incoming RPC connection from "INVALID" > to IP address. > -- Fix memory leak when retrieving reservations. This affects scontrol, sinfo, > sview, and the following slurmrestd endpoints: > 'GET /slurm/{any_data_parser}/reservation/{reservation_name}' > 'GET /slurm/{any_data_parser}/reservations' > -- Log warning instead of debuflags=conmgr gated log when deferring new > incoming connections when number of active connections exceed > conmgr_max_connections. > -- Avoid race condition that could result in worker thread pool not activating > all threads at once after a reconfigure resulting in lower utilization of > available CPU threads until enough internal activity wakes up all threads > in the worker pool. > -- Avoid theoretical race condition that could result in new incoming RPC > socket connections being ignored after reconfigure. > -- slurmd - Avoid race condition that could result in a state where new > incoming RPC connections will always be ignored. > -- Add ReconfigFlags=KeepNodeStateFuture to restore saved FUTURE node state on > restart and reconfig instead of reverting to FUTURE state. This will be > made the default in 25.05. > -- Fix case where hetjob submit would cause slurmctld to crash. > -- Fix jobs using --cpus-per-gpu and --mem keep pending forever when the > requested mem divided by the number of cpus will surpass the configured > MaxMemPerCPU. > -- Enforce that jobs using --mem and several --*-per-* options do not violate > the MaxMemPerCPU in place. > -- slurmctld - Fix use-cases of jobs incorrectly pending held when --prefer > features are not initially satisfied. > -- slurmctld - Fix jobs incorrectly held when --prefer not satisfied in some > use-cases. > -- Ensure RestrictedCoresPerGPU and CoreSpecCount don't overlap.

1 0

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

2025

2024

slurm-users ----- 2025 ----- April 2025 March 2025 February 2025 January 2025 ----- 2024 ----- December 2024 November 2024 October 2024 September 2024 August 2024 July 2024 June 2024 May 2024 April 2024 March 2024 February 2024 January 2024

slurm-users