February 2024 - slurm-users - lists.schedmd.com

job_container/tmpfs and srun.
by Phill Harvey-Smith 17 Jan '25

17 Jan '25

Hi all, On our setup we are using job_container/tmpfs to give each job it's own temp space. Since our compute nodes have reasonably sized disks for tasks that do a lot of disk I/O on user's data we have asked users to copy their data to the local disk at the beginning of the task and (if needed) copy it back at the end. This saves lots of NFS thrashing slowing down both the task and the NFS servers. However some of our users are having problems with this, their initial sbatch script will create a temp directory in their private /tmp copy their data to it and then try to srun a program. The srun will fall over as it doesn't seem to have have access to the copied data. I suspect this is because the srun task is getting it's own private /tmp. So my question is, is there a way to have the srun task inherit the /tmp of the initial sbatch? I'll include a sample of the script our user is using below. If any further information is required please feel free to ask. Cheers. Phill. #!/usr/bin/bash #SBATCH --nodes 1 #SBATCH --ntasks-per-node=1 #SBATCH --cpus-per-task=1 #SBATCH --time=00:00:10 #SBATCH --mem-per-cpu=3999 #SBATCH --output=script_out.log #SBATCH --error=script_error.log # The above options puts the STDOUT and STDERR of sbatch in # log files prefixed with 'script_'. # Create a randomly-named directory under /tmp jobtmpdir=$(mktemp -d) # Register a function to try and cleanup in case of job failure cleanup_handler() { echo "Cleaning up ${jobtmpdir}" rm -rf ${jobtmpdir} } trap 'cleanup_handler' SIGTERM EXIT # Change working directory to this directory cd ${jobtmpdir} # Copy the executable and input files from # where the job was submitted to the temporary directory. cp ${SLURM_SUBMIT_DIR}/a.out . cp ${SLURM_SUBMIT_DIR}/input.txt . # Run the executable, handling the collection of stdout # and stderr ourselves by redirecting to file srun ./a.out 2> task_error.log > task_out.log # Copy output data back to the submit directory. cp output.txt ${SLURM_SUBMIT_DIR} cp task_out.log ${SLURM_SUBMIT_DIR} cp task_error.log ${SLURM_SUBMIT_DIR} # Cleanup cd ${SLURM_SUBMIT_DIR} cleanup_handler

2 1

Can't schedule on cloud node: State=IDLE+CLOUD+POWERED_DOWN+NOT_RESPONDING
by Xaver Stiensmeier 20 Sep '24

20 Sep '24

Dear slurm-user list, I have a cloud node that is powered up and down on demand. Rarely it can happen that slurm's resumeTimeout is reached and the node is therefore powered down. We have set ReturnToService=2 in order to avoid the node being marked down, because the instance behind that node is created on demand and therefore after a failure nothing stops the system to start the node again as it is a different instance. I thought this would be enough, but apparently the node is still marked with "NOT_RESPONDING" which leads to slurm not trying to schedule on it. After a while NOT_RESPONDING is removed, but I would like to move it directly from within my fail script if possible so that the node can return to service immediately and not be blocked by "NOT_RESPONDING". Best regards, Xaver

2 3

Issues with Slurm 23.11.1
by Fokke Dijkstra 15 Mar '24

15 Mar '24

Dear all, Since the upgrade from Slurm 22.05 to 23.11.1 we are having problems with the communication between the slurmctld and slurmd processes. We are running a cluster with 183 nodes and almost 19000 cores. Unfortunately some nodes are in a different network preventing full internode communication. A network topology and setting TopologyParam RouteTree have been used to make sure no slurmd communication happens between nodes on different networks. In the new Slurm version we see the following issues, which did not appear in 22.05: 1. slurmd processes acquire many network connections in CLOSE-WAIT (or CLOSE_WAIT depending on the tool used) causing the processes to hang, when trying to restart slurmd. When checking for CLOSE-WAIT processes we see the following behaviour: Recv-Q Send-Q Local Address:Port Peer Address:Port Process 1 0 10.5.2.40:6818 10.5.0.43:58572 users:(("slurmd",pid=1930095,fd=72),("slurmd",pid=1930067,fd=72)) 1 0 10.5.2.40:6818 10.5.0.43:58284 users:(("slurmd",pid=1930095,fd=8),("slurmd",pid=1930067,fd=8)) 1 0 10.5.2.40:6818 10.5.0.43:58186 users:(("slurmd",pid=1930095,fd=22),("slurmd",pid=1930067,fd=22)) 1 0 10.5.2.40:6818 10.5.0.43:58592 users:(("slurmd",pid=1930095,fd=76),("slurmd",pid=1930067,fd=76)) 1 0 10.5.2.40:6818 10.5.0.43:58338 users:(("slurmd",pid=1930095,fd=19),("slurmd",pid=1930067,fd=19)) 1 0 10.5.2.40:6818 10.5.0.43:58568 users:(("slurmd",pid=1930095,fd=68),("slurmd",pid=1930067,fd=68)) 1 0 10.5.2.40:6818 10.5.0.43:58472 users:(("slurmd",pid=1930095,fd=69),("slurmd",pid=1930067,fd=69)) 1 0 10.5.2.40:6818 10.5.0.43:58486 users:(("slurmd",pid=1930095,fd=38),("slurmd",pid=1930067,fd=38)) 1 0 10.5.2.40:6818 10.5.0.43:58316 users:(("slurmd",pid=1930095,fd=29),("slurmd",pid=1930067,fd=29)) The first IP address is that of the compute node, the second that of the node running slurmctld. The nodes can communicate using these IP addresses just fine. 2. slurmd cannot be properly restarted [2024-01-18T10:45:26.589] slurmd version 23.11.1 started [2024-01-18T10:45:26.593] error: Error binding slurm stream socket: Address already in use [2024-01-18T10:45:26.593] fatal: Unable to bind listen port (6818): Address already in use This is probably because of the processes being in CLOSE-WAIT, which can only be killed using signal -9. 3. We see jobs stuck in completing CG state, probably due to communication issues between slurmctld and slurmd. The slurmctld sends repeated kill requests but those do not seem to be acknowledged by the client. This happens more often in large job arrays, or generally when many jobs start at the same time. However, this could be just a biased observation (i.e., it is more noticeable on large job arrays because there are more jobs to fail in the first place). 4. Since the new version we also see messages like: [2024-01-17T09:58:48.589] error: Failed to kill program loading user environment [2024-01-17T09:58:48.590] error: Failed to load current user environment variables [2024-01-17T09:58:48.590] error: _get_user_env: Unable to get user's local environment, running only with passed environment The effect of this is that the users run with the wrong environment and can’t load the modules for the software that is needed by their jobs. This leads to many job failures. The issue appears to be somewhat similar to the one described at: https://bugs.schedmd.com/show_bug.cgi?id=18561 In that case the site downgraded the slurmd clients to 22.05 which got rid of the problems. We’ve now downgraded the slurmd on the compute nodes to 23.02.7 which also seems to be a workaround for the issue. Does anyone know of a better solution? Kind regards, Fokke Dijkstra -- Fokke Dijkstra <f.dijkstra(a)rug.nl> <f.dijkstra(a)rug.nl> Team High Performance Computing Center for Information Technology, University of Groningen Postbus 11044, 9700 CA Groningen, The Netherlands

2 3

Problem building slurm with PMIx
by Patrick Begou 11 Mar '24

11 Mar '24

Hi ! I manage a small CentOS8 cluster using slurm slurm-20.11.7-1 and OpenMPI built from sources. - I know this OS is not maintained any more and I need to negotiate downtime to reinstall - I know Slurm 20.11.7 has security issue (I've built it from source some years ago with rpmbuild -ta --with mysql --with hwloc slurm-20.11.7.tar.bz) and I should update. All was running fine until I add a GPU Node and Nvidia sdk. This SDK provides an openMPI3 implementation GPU aware but I'm unable to launch an intranode parallel job with it using srun: -------------------------------------------------------------------------- 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. -------------------------------------------------------------------------- I check with "srun --mpi=list" and got no pmx. srun: MPI types are... srun: pmi2 srun: cray_shasta srun: none So I decide to build the rpms from slurm-20.11.9.tar.bz2 as I had done previously for 20.11.7 and update. I've first installed pmix-2.1.1-1 from src as I had no pmix-devel rpm in my local CentOS8 repo: rpm --rebuild pmix-2.1.1-1.el8.src.rpm dnf install pmix-devel-2.1.1-1.el8.x86_64.rpm pmix-2.1.1-1.el8.x86_64.rpm Then build Slurm from slurm-20.11.9.tar.bz2 (just changing python3 to python38 in the spec file) rpmbuild -ta --with mysql --with hwloc --with pmix slurm-20.11.9.tar.bz2 And the try to install these package on the GPU node dnf install slurm-slurmd-20.11.9-1.el8.x86_64.rpm slurm-20.11.9-1.el8.x86_64.rpm slurm-devel-20.11.9-1.el8.x86_64.rpm slurm-libpmi-20.11.9-1.el8.x86_64.rpm But I get this strange error: Error: Problem: conflicting requests - nothing provides pmix = 20.11.9 needed by slurm-slurmd-20.11.9-1.el8.x86_64 (try to add '--skip-broken' to skip uninstallable packages or '--nobest' to use not only best candidate packages) Why this request on PMIX with the slurm version number ? Am I wrong somewhere ? Thanks for your help Patrick

2 1

Slurm management of dual-node server trays?
by Ole Holm Nielsen 07 Mar '24

07 Mar '24

We're in the process of installing some racks with Lenovo SD665 V3 [1] water-cooled servers. A Lenovo DW612S chassis contains 6 1U trays with 2 SD665 V3 servers mounted side-by-side in each tray. Lenovo delivers SD665 V3 servers including water-cooled NVIDIA InfiniBand "SharedIO" adapters [2] so that one node is the Primary including a PCIe adapter, and the other is Auxiliary with just a cable to the Primary's adapter. Obviously, servicing 2 "Siamese twin" Slurm nodes requires a bit of care and planning. What is worse is that when the Primary node is rebooted or powered down, the Auxiliary node will lose its Infiniband connection and may have a PCIe fault or an NMI as documented in [3]. And when nodes are powered up, the Primary must have completed POST before the Auxiliary gets started. I wonder how to best deal with power failures? It seems that when Slurm jobs are running on Auxiliary nodes, these jobs are going to crash when the possibly unrelated Primary node goes down. This looks like a pretty bad system design on the part of Lenovo :-( The goal was apparently to same some money on IB adapters and having fewer IB cables. Question: Do any Slurm sites out there already have experiences with Lenovo "Siamese twin" nodes with SharedIO IB? Have you developed some operational strategies, for example dealing with node pairs as a single entity for job scheduling? Thanks for sharing any ideas and insights! Ole [1] https://lenovopress.lenovo.com/lp1612-lenovo-thinksystem-sd665-v3-server [2] https://lenovopress.lenovo.com/lp1693-thinksystem-nvidia-connectx-7-ndr200-… [3] https://support.lenovo.com/us/en/solutions/ht510888-thinksystem-sd650-and-c… -- Ole Holm Nielsen PhD, Senior HPC Officer Department of Physics, Technical University of Denmark

3 3

Question about IB and Ethernet networks
by Dan Healy 04 Mar '24

04 Mar '24

Hi Fellow Slurm Users, This question is not slurm-specific, but it might develop into that. My question relates to understanding how *typical* HPCs are designed in terms of networking. To start, is it typical for there to be a high speed Ethernet *and* Infiniband networks (meaning separate switches, NICs)? I know you can easily setup IP over IB, but is IB usually fully reserved for MPI messages? I’m tempted to spec all new HPCs with only a high speed (200Gbps) IB network, and use IPoIB for all slurm comms with compute nodes. I plan on using BeeGFS for the file system with RDMA. Just looking for some feedback, please. Is this OK? Is there a better way? If yes, please share why it’s better. Thanks, Daniel Healy

6 6

How to get usage data for a QOS
by thomas.hartmann＠th-ht.de 01 Mar '24

01 Mar '24

Hi, so, I figured out that I can give some users priority access for a specific amount of TRES by creating a qos with the GrpTRESMins property and the DenyOnLimit,NoDecay flags. This works nicely. However, I would like to know, how much of this has already been consumed and I have not yet found a way to do this. Like: How can I get the amount of TRES/TRES minutes consumed for a certain QOS? Thanks a lot! Thomas

3 2

pty jobs are killed when another job on the same node terminates
by Thomas Hartmann 01 Mar '24

01 Mar '24

Hi, when I start an interactive job like this: srun --pty --mem=3G -c2 bash And then I schedule and run other jobs (can be interactive or non interactive) and one of these jobs that runs on the same node terminates, the interactive job gets killed with this message: srun: error: node01.abc.at: task 0: Killed I attached our slurm config. Does anybody have an idea what is going on here or where I could look to debug? I'm quite new to slurm, so I don't know all the places to look... Thanks a lot in advance! Thomas

4 3

GPU shards not exclusive
by Reed Dier 29 Feb '24

29 Feb '24

I seem to have run into an edge case where I’m able to oversubscribe a specific subset of GPUs on one host in particular. Slurm 22.05.8 Ubuntu 20.04 cgroups v1 (ProctrackType=proctrack/cgroup) It seems to be partly a corner case with a couple of caveats. This host has 2 different GPU types in the same host. All GPUs are configured with 6 shards/gpu. > NodeName=gpu03 CPUs=88 RealMemory=768000 Sockets=2 CoresPerSocket=22 ThreadsPerCore=2 State=UNKNOWN Feature=avx512 Gres=gpu:rtx:2,gpu:p40:6,shard:48 My gres.conf is auto-nvml, with nothing else in the file. > AutoDetect=nvml Schedule and select from slurm.conf > SchedulerType=sched/backfill > SelectType=select/cons_tres > SelectTypeParameters=CR_CPU_Memory > GresTypes=gpu,shard Here’s a copy of slurmd’s log starting up with debug logging: https://pastebin.com/fbriTAZD <https://pastebin.com/fbriTAZD> I checked there to make sure that it was allocating an even 6 shards per GPU as intended, and it did. The only thing odd about the rtx cards is that they have an NVLink bridge between them, but I wouldn’t expect that to have any impact with respect to shards. So far it only appears to manifest if anything is using the named gpu (--gres gpu:rtx:1), but not as a general gres (--gres gpu:1). It also doesn’t occur with any count of named p40 gres either, only with the rtx gres. The below was as “high” as I was able to capture it oversubscribed in very quick testing. You can see there are 2 rtx and 43 shards allocated, which should be an impossible number, if we correctly assume an rtx==6 shards, and with 48 total shards, that should reduce the available pool of shards to 36 shards. > NodeName=gpu03 Arch=x86_64 CoresPerSocket=22 > CPUAlloc=88 CPUEfctv=88 CPUTot=88 CPULoad=0.27 > AvailableFeatures=avx512 > ActiveFeatures=avx512 > Gres=gpu:p40:6(S:0),gpu:rtx:2(S:0),shard:p40:36(S:0),shard:rtx:12(S:0) > NodeAddr=gpu03 NodeHostName=gpu03 Version=22.05.8 > OS=Linux 5.4.0-164-generic #181-Ubuntu SMP Fri Sep 1 13:41:22 UTC 2023 > RealMemory=768000 AllocMem=45056 FreeMem=563310 Sockets=2 Boards=1 > State=ALLOCATED+RESERVED ThreadsPerCore=2 TmpDisk=0 Weight=1 Owner=N/A MCS_label=N/A > Partitions=gpu,gpu-prod > BootTime=2024-02-02T14:17:07 SlurmdStartTime=2024-02-14T16:28:17 > LastBusyTime=2024-02-14T16:38:03 > CfgTRES=cpu=88,mem=750G,billing=88,gres/gpu=8,gres/gpu:p40=6,gres/gpu:rtx=2,gres/shard=48 > AllocTRES=cpu=88,mem=44G,gres/gpu=2,gres/gpu:rtx=2,gres/shard=43 > CapWatts=n/a > CurrentWatts=0 AveWatts=0 > ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/s The above is 1 rtx:2 jobs, and 43 shard:1 jobs, which is as many as I can seem to push through with the given cpu cores on the system, I expect if there were 98-100 cores I could completely oversubscribe it. If I do (1) rtx:2 + (24) shard:2 it actually limits it correctly to 36 shards, which is odd because the case that caused us to discover this was when 2 rtx:1 jobs were running for multiple days, and somehow shards got scheduled on those gpus. > AllocTRES=cpu=40,mem=20G,gres/gpu=2,gres/gpu:rtx=2,gres/shard=36 If I flip it around to shards before rtx, then I’m unable to oversubscribe it. If I add any P40 jobs before the rtx and then shard jobs, also unable to oversubscribe. This sounded familiar to Bug 16484 <https://bugs.schedmd.com/show_bug.cgi?id=16484> although they claimed it was resolved by using auto-nvml which would preclude this node. Here’s my very simple bash script that just srun's a sleep 20 so that it runs long enough to observe it. > #!/bin/bash > #rtx 2 > for((i=1;i<=1;++i)) ; > do > srun --gres gpu:rtx:2 --reservation=shardtest -J "RTX $i" sleep 20 & > done > sleep 1 > #shard 48 > for((i=1;i<=43;++i)) ; > do > srun --gres shard:1 --reservation=shardtest -J "shard $i" sleep 20 & > done > sleep 5 ; scontrol show node gpu03 Curious if anyone has any ideas beyond try a newer release? Thanks, Reed

2 2

sbatch and cgroup v2
by Dietmar Rieder 29 Feb '24

29 Feb '24

Hi, I'm new to slrum, but maybe someone can help me: I'm trying to restrict the CPU usage to the actually requested/allocated resources using cgroup v2. For this I made the following settings in slurmd.conf: ProctrackType=proctrack/cgroup TaskPlugin=task/cgroup,task/affinity And in cgroup.conf CgroupPlugin=cgroup/v2 CgroupAutomount=yes ConstrainCores=yes ConstrainRAMSpace=yes ConstrainDevices=yes AllowedRAMSpace=98 cgroup v2 seems to be active on the compute node: # mount | grep cgroup cgroup2 on /sys/fs/cgroup type cgroup2 (rw,nosuid,nodev,noexec,relatime,nsdelegate,memory_recursiveprot) # cat /sys/fs/cgroup/cgroup.subtree_control cpuset cpu io memory pids # cat /sys/fs/cgroup/system.slice/cgroup.subtree_control cpuset cpu io memory pids Now, when I use sbatch to submit the following test script, the python script which is started from the batch script is utilizing all CPUs (96) at 100% on the allocated node, although I only ask for 4 cpus (--cpus-per-task=4). I'd expect that the task can not use more that these 4. #!/bin/bash #SBATCH --output=/local/users/appadmin/test-%j.log #SBATCH --job-name=test #SBATCH --chdir=/local/users/appadmin #SBATCH --cpus-per-task=4 #SBATCH --ntasks=1 #SBATCH --nodes=1 #SBATCH --mem=64gb #SBATCH --time=4:00:00 #SBATCH --partition=standard #SBATCH --gpus=0 #SBATCH --export #SBATCH --get-user-env=L export PATH=/usr/local/bioinf/jupyterhub/bin:/usr/local/bioinf/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/usr/local/bioinf/miniforge/condabin source .bashrc conda activate test python test.py The python code in test.py is the following using the cpu_load_generator package from [1]: #!/usr/bin/env python import sys from cpu_load_generator import load_single_core, load_all_cores, from_profile load_all_cores(duration_s=120, target_load=1) # generates load on all cores Interestingly, when I use srun to launch an interactive job, and run the python script manually, I see with top that only 4 cpus are running at 100%. And I also python errors thrown when the script tries to start the 5th process (which makes sense): File "/usr/local/bioinf/miniforge/envs/test/lib/python3.12/multiprocessing/process.py", line 314, in _bootstrap self.run() File "/usr/local/bioinf/miniforge/envs/test/lib/python3.12/multiprocessing/process.py", line 108, in run self._target(*self._args, **self._kwargs) File "/usr/local/bioinf/miniforge/envs/test/lib/python3.12/site-packages/cpu_load_generator/_interface.py", line 24, in load_single_core process.cpu_affinity([core_num]) File "/usr/local/bioinf/miniforge/envs/test/lib/python3.12/site-packages/psutil/__init__.py", line 867, in cpu_affinity self._proc.cpu_affinity_set(list(set(cpus))) File "/usr/local/bioinf/miniforge/envs/test/lib/python3.12/site-packages/psutil/_pslinux.py", line 1714, in wrapper return fun(self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/bioinf/miniforge/envs/test/lib/python3.12/site-packages/psutil/_pslinux.py", line 2213, in cpu_affinity_set cext.proc_cpu_affinity_set(self.pid, cpus) OSError: [Errno 22] Invalid argument What am I missing, why are the CPU resources not restricted when I use sbatch? Thanks for any input or hint Dietmar [1]: https://pypi.org/project/cpu-load-generator/

3 7

slurmdbd error - Symbol `slurm_conf' has different size in shared object
by Miriam Olmi 29 Feb '24

29 Feb '24

Hi all, I am having some issue with the new version of slurm 23.11.0-1. I had already installed and configured slurm 23.02.3-1 on my cluster and all the services were active and running properly. Following the instructions of the official SLURM webpage, for the moment I upgrated only the slurmdbd service. In principle the cluster should be able to work properly if the slurmdbd has a higher version with respect to slurmctld and slurmd. Unfortunately the slurmdbd service fails to start with the following status: slurmdbd.service - Slurm DBD accounting daemon Loaded: loaded (/usr/lib/systemd/system/slurmdbd.service; enabled; vendor preset: disabled) Active: failed (Result: core-dump) since Wed 2024-02-28 10:05:53 CET; 9min ago Process: 534938 ExecStart=/usr/sbin/slurmdbd -D -s $SLURMDBD_OPTIONS (code=dumped, signal=SEGV) Main PID: 534938 (code=dumped, signal=SEGV) Feb 28 10:05:53 slurm-db systemd[1]: Started Slurm DBD accounting daemon. Feb 28 10:05:53 slurm-db slurmdbd[534938]: /usr/sbin/slurmdbd: Symbol `slurm_conf' has different size in shared object, consider re-linking Feb 28 10:05:53 slurm-db systemd[1]: slurmdbd.service: Main process exited, code=dumped, status=11/SEGV Feb 28 10:05:53 slurm-db systemd[1]: slurmdbd.service: Failed with result 'core-dump'. Can anyone help me? Thanks in advance, Miriam

2 4

salloc+srun vs just srun
by wdennis＠nec-labs.com 29 Feb '24

29 Feb '24

Hi list, In our institution, our instructions to users who want to spawn an interactive job (for us, a bash shell) have always been to do "srun ..." from the login node, which has always been working well for us. But when we had a recent Slurm training, the SchedMD folks advised us to use "salloc" and then "srun" to do interactive jobs. I tried this today, "salloc" gave me a shell on a server, the same as srun does, but then when I tried to "srun [programname]" it hung there with no output. Of course when I tried "srun [programname] &" it spawned the background job, and gave me back a prompt. Either time I had to Ctrl-C the running srun job, and got no output other than the srun/slurmstepd termination output. I think I read somewhere that directly invoking srun creates an allocation; why then would I want to do an initial salloc, and then srun? (i the case that I want a foreground program, such as a bash shell) I have surveyed some other institution's Slurm interactive jobs documentation for users, I see both examples of advice to run srun directly, or salloc and then srun. Please help me to understand how this is intended to work, and if we are "doing it wrong" :) Thanks, Will

3 4

canonical way to run longer shell/bash interactive job (instead of srun inside of screen/tmux at front-end)?
by Josef Dvoracek 28 Feb '24

28 Feb '24

What is the recommended way to run longer interactive job at your systems? Our how-to includes starting screen at front-end node and running srun with bash/zsh inside, but that indeed brings dependency between login node (with screen) and the compute node job. On systems with multiple front-ends users need to remember the login node where they have their screen session.. Are you anybody using something more advanced and still understandable by casual user of HPC? (I know Open On Demand, but often the use of native console has certain benefits. ) cheers josef

7 9

Enforcing relative resource restrictions in submission script
by Matthew R. Baney 28 Feb '24

28 Feb '24

Hello Slurm users, I'm trying to write a check in our job_submit.lua script that enforces relative resource requirements such as disallowing more than 4 CPUs or 48GB of memory per GPU. The QOS itself has a MaxTRESPerJob of cpu=32,gres/gpu=8,mem=384G (roughly one full node), but we're looking to prevent jobs from "stranding" GPUs, e.g., a 32 CPU/384GB memory job with only 1 GPU. I might be missing something obvious, but the rabbit hole I'm going down at the moment is trying to check all of the different ways job arguments could be set in the job descriptor. i.e., the following should all be disallowed: srun --gres=gpu:1 --mem=49G ... (tres_per_node, mem_per_node set in the descriptor) srun --gpus=1 --mem-per-gpu=49G ... (tres_per_job, mem_per_tres) srun --gres=gpu:1 --ntasks-per-gpu=5 ... (tres_per_node, num_tasks, ntasks_per_tres) srun --gpus=1 --ntasks=2 --mem-per-cpu=25G ... (tres_per_job, num_tasks, mem_per_cpu) ... Essentially what I'm looking for is a way to access the ReqTRES string from the job record before it exists, and then run some logic against that i.e., if (CPU count / GPU count) > 4 or (mem count / GPU count) > 48G, error out. Is something like this possible? Thanks, Matthew -- Matthew Baney Assistant Director of Computational Systems mbaney(a)umd.edu | (301) 405-6756 University of Maryland Institute for Advanced Computer Studies 3154 Brendan Iribe Center 8125 Paint Branch Dr. College Park, MD 20742

2 1

User-facing documentation on shard use
by wdennis＠nec-labs.com 28 Feb '24

28 Feb '24

Hello list, We have just enabled "gres/shard" in order to enable sharing of GPUs on our cluster. I am now looking for examples of user-facing documentation on this feature. If anyone has something, and can send a URL or other example, I'd appreciate it. Thanks, Will

1 0

Partition, Qos Limits & Scheduling of large jobs
by Muck, Katrin 28 Feb '24

28 Feb '24

Hi everyone! I read the slurm documentation about qos, resource limits, scheduling and priority now multiple times and even looked into the slurm source but I'm still not sure if I got everything correctly, so this is why I decided to ask here ... The problem: we see the effect that sometimes larger jobs with e.g. 16 gpus in our (small) gpu queue get delayed and shifted back without any reason that is apparent to us. small jobs that only use e.g. 1 or 2 gpus get scheduled much quicker even though they have a runtime of 3 days ... What we want to do: - We have a number of nodes with 2x gpus that are usable by the users of our cluster - Some of these nodes belong to so called 'private projects'. Private projects have higher priority than other projects. Attached to that is a contingent of nodes & "guaranteed" nodes e.g. they could have a contingent of 4 nodes (8 gpus) and e.g. 2 "guaranteed" nodes (4 gpus) - Guaranteed nodes are nodes that should always be kept idle for the private project, so users of the private project can immediately schedule work on those nodes - The other nodes are shared with other projects in general if they are not "in use" How we are currently doing this (it has history): Lets assume we have 50 nodes and 100 gpus. - We have a single partition for all gpu nodes (e.g. 50 nodes) - Private projects have private queues with a very high priority and a gres limit of the number of gpus they reserved (e.g. 10 nodes -> 20 gpus) - Normal projects only have access to the public queue and schedule work there. - This public queue has an upper gres limit of "total number of gpus" - "guaranteed gpus of all private projects" (e.g. 50 - 10 nodes -> 40 nodes -> 80 gpus). Regarding the scheduler, we currently use the following settings: SchedulerType=sched/backfill SelectType=select/cons_tres SelectTypeParameters=CR_CORE_MEMORY,CR_CORE_DEFAULT_DIST_BLOCK,CR_ONE_TASK_PER_CORE SchedulerParameters=defer,max_sched_time=4,default_queue_depth=1000,partition_job_depth=500,enable_user_top,bf_max_job_user=20,bf_interval=120,bf_window=4320,bf_resolution=1800,bf_continue Partition/queue depth is deliberately set high at the moment to avoid problems with jobs not even being examined. The problem in more detail: One of the last jobs (16 gpus needed) we diagnosed had a approximate start time that was beyond all end times of running/scheduled jobs - like jobs on feb 22 would release more than enough gpus so the job could have been immediately scheduled afterwards, but the start time was still feb 23. Priority wise the job had the highest priority of all pending jobs for the partition. When we turned on scheduler debugging and increased log levels, we observed the following messages for this job: JobId=xxxx being held, if allowed the job request will exceed QOS xxxxx group max tres(gres/gpu) limit yy with already used yy + requested 16 followed by sched: JobId=2796696 delayed for accounting policy So to us this meant that the scheduler was always hitting the qos limits, which makes sense because the usage is always very high in the gpu queue and thus the job wasn't scheduled ... At first we were worried, that this meant that "held"/"delayed" jobs like this would never actually get scheduled when contention is high enough e.g. small jobs getting backfilled in and thus qos limits stay at max for a long time. But for some reason we could not determine the job eventually got scheduled at one point and then ran at the scheduled start time. Open Questions: - why it couldn't be scheduled in the first place. initially we thought (from the source code i looked into) the "delayed for accounting policy" prevents further scheduling in general, but since it was scheduled this assumption must be wrong? - why it was scheduled at some point. when it was scheduled, contention was still high and the qos limits definitely still applied - how we could modify the current setup so that the scheduling of larger jobs becomes "better" and more reproducible/explainable Apart from all of this I'm also asking myself if there is maybe a better way to setup a system that works the way we want? This got a bit long but I hope its clear enough :) Kind regards, Katrin

1 0

"Optimal" slurm configuration
by Max Grönke 26 Feb '24

26 Feb '24

Hello! In our current cluster the workflows are quite diverse (bunch of large, long (24-72h) jobs; medium size <4h job; and many small 1 node jobs). The current priority is fair share only (averaged on a ~few months timescale). For the new setup we would like to (1) discourage the 1 node jobs [especially the long ones] (2) have a fair use of the machine for the different cases. In particular, we found that the waiting time / runtime ratio is very large for the ~1-4h jobs. For this reason, I have been trying to read up on strategies in terms of partition / priority politics but could not find any good documentation in this regard (as https://slurm.schedmd.com/documentation.html is more focused on the technical aspects). Does anyone know any good references for this? In particular, we were thinking about (a) somehow have a long-term as well as shorter term fairshare count, i.e., lowering the priority if a user has something running already or (b) introduce a "small" partition for the <4h jobs with higher priority but we're unsure if this will block all the larger jobs to run... Any help or reading material is highly appreciated. Many thanks, Max

2 1

FAQ describing how to hold a job ignores scontrol subcommands specifically for that purpose
by urbanjost 25 Feb '24

25 Feb '24

There are scontrol subcommands uhold/hold/release/requeuehold that are ignored when describing how to place a job on hold in FAQ 21; and it is never explained why the method described therein is the best method, it just states it is. Does anyone know why the FAQ method is better than using the subcommands? Is it because the PRIORITY and/or NICE values are not altered (maybe)? The question is also about Running but the answer is just about Starting and not Suspending which is not quite as clear (I think "running" should be "starting" to make that clear; and/or how to suspend should be described as well). If the answer is not clear to anyone, I might turn this into a request for clarification in the Slurm bugzilla as a documentation change request but wanted to see if this was already clear to anyone and I am missing something. From FAQ: 21. How can I temporarily prevent a job from running (e.g. place it into a hold state)? The easiest way to do this is to change a job's earliest begin time (optionally set at job submit time using the --begin option). The example below places a job into hold state (preventing its initiation for 30 days) and later permitting it to start now. <METHOD I> $ scontrol update JobId=1234 StartTime=now+30days ... later ... $ scontrol update JobId=1234 StartTime=now Note: Empirically in METHOD I the JobId can be a <job_list> , which I initially thought required single JobIDs. No explanation is given on why METHOD I is best; and there are other methods that seem more intuitive. I wonder what is undesirable about the following method which I have been using -- using the scontrol(1) subcommands hold/uhold/release/requeuehold. <METHOD II> $ scontrol hold <job_list> # advantage to administrator as user cannot change $ scontrol uhold <job_list> $ scontrol release <job_list> Examples: $ scontrol uhold jobname=JOB_NAME $ scontrol uhold '[100-200],300,500' Using uhold the "Reason" changes to something easily identifying the job is being held, as "Reason=None" became "Reason=JobHeldUser which seems better that Method I in that regard. The downside might be PRIORITY changed to zero and then went to a very large value when released? Another method appears to be that setting PRIORITY to zero also places jobs in hold. <METHOD III> $ scontrol update jobid=373 Priority=0 $ scontrol release jobid=373 # sets to a very high value $ scontrol update jobid=373 Priority=11111 # put back to lower desired value Once lowered, does an optional setting prevent a user from raising PRIORITY(?) The manual says Only the Slurm administrator or root can increase job's priority. At least on my machine the "release" buts the priority to a very high value, and a regular user can lower the value back to the (probably) lower original value. I did not see it happening but there are some statements in the documentation that make me think not only PRIORITY but perhaps the NICE value might be changed by METHOD II and METHOD III, although I could not get the NICE value to be inadvertently changed. Sent with [Proton Mail](https://proton.me/) secure email.

1 1

slurm-23.11.3-1 with X11 and zram causing permission errors: error: _forkexec_slurmstepd: slurmstepd failed to send return code got 0: Resource temporarily unavailable; Requeue of JobId=530 returned an error: Only batch jobs are accepted or processed
by Robert Kudyba 25 Feb '24

25 Feb '24

We switched over from using systemctl for tmp.mount and change to zram, e.g., modprobe zram echo 20GB > /sys/block/zram0/disksize mkfs.xfs /dev/zram0 mount -o discard /dev/zram0 /tmp srun with --x11 was working before changing this. We're on RHEL 9. slurmctld logs show this whenever --x11 is used with srun: [2024-02-23T20:22:43.442] [529.extern] error: setup_x11_forward: failed to create temporary XAUTHORITY file: Permission denied [2024-02-23T20:22:43.442] [529.extern] error: x11 port forwarding setup failed [2024-02-23T20:22:43.442] error: _forkexec_slurmstepd: slurmstepd failed to send return code got 0: Resource temporarily unavailable [2024-02-23T20:22:43.443] Could not launch job 529 and not able to requeue it, cancelling job [2024-02-23T20:26:15.881] [530.extern] error: setup_x11_forward: failed to create temporary XAUTHORITY file: Permission denied [2024-02-23T20:26:15.881] [530.extern] error: x11 port forwarding setup failed [2024-02-23T20:26:15.882] error: _forkexec_slurmstepd: slurmstepd failed to send return code got 0: Resource temporarily unavailable [2024-02-23T20:26:15.883] Could not launch job 530 and not able to requeue it, cancelling job slurmd log entries from a node: [2024-02-23T20:26:15.859] sched: _slurm_rpc_allocate_resources JobId=530 NodeList=2402-node005 usec=1800 [2024-02-23T20:26:15.882] _slurm_rpc_requeue: Requeue of JobId=530 returned an error: Only batch jobs are accepted or processed [2024-02-23T20:26:15.883] _slurm_rpc_kill_job: REQUEST_KILL_JOB JobId=530 uid 0 [2024-02-23T20:26:15.962] _slurm_rpc_complete_job_allocation: JobId=530 error Job/step already completing or completed srun -v --pty -t 0-4:00 --x11 --mem=10g srun: defined options srun: -------------------- -------------------- srun: account : me srun: mem : 10G srun: nodelist : our-node srun: pty : srun: time : 04:00:00 srun: verbose : 1 srun: x11 : all srun: -------------------- -------------------- srun: end of defined options srun: Waiting for resource configuration srun: error: Nodes our-node are still not ready srun: error: Something is wrong with the boot of the nodes. slurm.conf has PrologFlags=x11 set. /usr/bin/xauth is installed on each compute node. Is this a known issue with zram or is that just a red herring and there's something else wrong?

3 5

Slurm version 23.11.4 is now available
by Tim McMullan 22 Feb '24

22 Feb '24

We are pleased to announce the availability of Slurm version 23.11.4. The 23.11.4 release includes a number of fixes to stability and various bug fixes. Some notable changes include that VSZ is no longer being reported when using cgroup/v2 (this is not provided by the kernel), a warning has been added if using select/linear and tolology/tree together as this will not be supported in the next major release, and a backwards compatibility issue that caused jobs using --gpus to be rejected when submitted from 23.02 or 22.05. Slurm can be downloaded from https://www.schedmd.com/downloads.php . -Tim > * Changes in Slurm 23.11.4 > ========================== > -- Fix a memory leak when updating partition nodes. > -- Don't leave a partition around if it fails to create with scontrol. > -- Fix segfault when creating partition with bad node list from scontrol. > -- Fix preserving partition nodes on bad node list update from scontrol. > -- Fix assertion in developer mode on a failed message unpack. > -- Fix repeat POWER_DOWN requests making the nodes available for ping. > -- Fix rebuilding job alias_list on restart when nodes are still powering up. > -- Fix INVALID nodes running health check. > -- Fix cloud/future nodes not setting addresses on invalid registration. > -- scrun - Remove the requirement to set the SCRUN_WORKING_DIR environment > variable. This was a regression in 23.11. > -- Add warning for using select/linear with topology/tree. > This combination will not be supported in the next major version. > -- Fix health check program not being run after first pass of all nodes when > using MaxNodeCount. > -- sacct - Set process exit code to one for all errors. > -- Add SlurmctldParameters=disable_triggers option. > -- Fix issue running steps when the allocation requested an exclusive > allocation shards along with shards. > -- Fix cleaning up the sleep process and the cgroup of the extern step if > slurm_spank_task_post_fork returns an error. > -- slurm_completion - Add missing --gres-flags= options > multiple-tasks-per-sharing and one-task-per-sharing. > -- scrun - Avoid race condition that could cause outbound network > communications to incorrectly rejected with an incomplete packet error. > -- scrun - Gracefully handle kernel giving invalid expected number of incoming > bytes for a connection causing incoming packet corruption resulting in > connection getting closed. > -- srun - return 1 when a step lauch fails > -- scrun - Avoid race condition that could cause deadlock during shutdown. > -- Fix scontrol listpids to work under dynamic node scenarios. > -- Add --tres-bind to --help and --usage output. > -- Add --gres-flags=allow-task-sharing to allow GPUs to still be accessible > among all tasks when binding GPUs to specific tasks. > -- Fix issue with CUDA_VISIBLE_DEVICES showing the same MIG device for all > tasks when using MIGs with --tres-per-task or --gpus-per-task. > -- slurmctld - Prevent a potential hang during shutdown/reconfigure if the > association cache thread was previously shut down. > -- scrun - Avoid race condition that could cause scrun to hang during > shutdown when connections have pending events. > -- scrun - Avoid excessive polling of connections during shutdown that could > needlessly cause 100% CPU usage on a thread. > -- sbcast - Use user identity from broadcast credential instead of looking it > up locally on the node. > -- scontrol - Remove "abort" option handling. > -- Fix an error message referring to the wrong RPC. > -- Fix memory leak on error when creating dynamic nodes. > -- Fix a slurmctld segfault when a cloud/dynamic node changes hostname on > registration. > -- Prevent a slurmctld deadlock if the gpu plugin fails to load when > creating a node. > -- Change a slurmctld fatal() to an error() when attempting to create a > dynamic node with a global autodetect set in gres.conf. > -- Fix leaving node records on error when creating nodes with scontrol. > -- scrun/sackd - Avoid race condition where shutdown could deadlock. > -- Fix a regression in 23.02.5 that caused pam_slurm_adopt to fail when > the user has multiple jobs on a node. > -- Add GLOB_SILENCE flag that silences the error message which will display if > an include directive attempts to use the "*" wildcard. > -- Fix jobs getting rejected when submitting with --gpus option from older > versions of job submission commands (23.02 and older). > -- cgroup/v2 - Return 0 for VSZ. Kernel cgroups do not provide this metric. > -- scrun - Avoid race condition where outbound RPCs could be corrupted. > -- scrun - Avoid race condition that could cause a crash while compiled in > debug mode. > -- gpu/rsmi - Disable gpu usage statistics when not using ROCM 6.0.0+ > -- Fix stuck processes and incorrect environment when using --get-user-env. > -- Avoid segfault in the slurmdbd when TrackWCKey=no but you are still using > use WCKeys. > -- Fix ctld segfault with TopologyParam=RoutePart and no partition defined. > -- slurmctld - Fix missing --deadline handling for jobs not evaluated by the > schedulers (i.e. non-runnable, skipped for other reasons, etc.). > -- Demote some eio related logs from error to verbose in user commands. These > are not generally actionable by the user and are easilly generated by port > scanning a machine running srun. > -- Make sprio correctly print array tasks that have not yet been split out. > -- topology/block - Restrict the number of last-level blocks in any allocation. -- Tim McMullan Release Management, Support, and Development SchedMD LLC - Commercial Slurm Development and Support

1 0

RHEL 8.9+SLURM-23.11.3+MLNX_OFED_LINUX-23.10-1.1.9.0+ OpenMPI-5.0.2
by S L 20 Feb '24

20 Feb '24

Hi All, We're currently in the process of setting up SLURM on a RHEL 8.9 based cluster. Here's a summary of the steps we've taken so far: Installed MLNX OFED ConnectX-5.2. Compiled and installed PMiX and UCX. Compiled and installed Slurm with PMiX_v4 and UCX support. Compiled OpenMPI with SLURM, PMIx, libevent, and hwloc support. All compute nodes are reachable via the IB network. *Problem:* While hello world MPI jobs are working fine on multiple nodes, the jobs are not utilizing Infiniband. srun --mpi=pmix -N2 -n2 --ntasks-per-node=2 ./hello > log.out 2>&1 Output from srun --mpi=list: MPI plugin types are... none cray_shasta pmi2 pmix specific pmix plugin versions available: pmix_v4 Could someone please point me in the right direction on how to troubleshoot this issue? Thank you for your assistance. Sudhakar

2 1

URL for how to do for SLURM accounting setup
by John Joseph 20 Feb '24

20 Feb '24

Dear ALL, Good morning we were able to setup a test SLURM based system, with 4 nodes , Ubuntu 22.04 LTS and we were able to run COMSOL using "comsol batch" command Now we plan to have accounting https://slurm.schedmd.com/accounting.html Like to reach out and get guidance on any tutorial or how to do documentation on setting up accounting Appreciate your support Thanks Joseph John

2 2

CPU utilisation using two commands scontrol association and sreport makes a huge difference
by prachikakade.lit＠gmail.com 20 Feb '24

20 Feb '24

Dear Team, I created a small cluster of 3 nodes on my VM ware to work on the CPU utilization concept. I created a user name= hpcuser01, and allocated GrpTresMin=cpu=59000040 -> CPU minutes and gpu=0 Now, when I checked his utilization using scontrol association cmd # scontrol show ass user=hpcuser01 | grep -n15 hpcuser01 "" Output: QOS=hpcuser01_test(11) UsageRaw=5789263754.153907 GrpJobs=N(0) GrpJobsAccrue=N(5) GrpSubmitJobs=N(5) GrpWall=N(1188290.27) GrpTRES=cpu=N(0),mem=N(0),energy=N(0),node=N(0),billing=N(0),fs/disk=N(0),vmem=N(0),pages=N(0),gres/gpu=N(0) GrpTRESMins=cpu=59000040(59001152),mem=N(293429427523),energy=N(0),node=N(1993778),billing=N(96073808),fs/disk=N(0),vmem=N(0),pages=N(0),gres/gpu=1(0) GrpTRESRunMins=cpu=N(0),mem=N(0),energy=N(0),node=46080(0),billing=N(0),fs/disk=N(0),vmem=N(0),pages=N(0)"" Here we can see that GrpTRESMins=cpu=59000040(59001152) after converting in Hours ( value /60) it will be: 983334(983352) CPU hours This means 983352 CPU hours have been used by hpcuser01 but, when I checked his utilization using sreport cmd #sreport -t hour --tres=cpu,gres/gpu,billing cluster AccountUtilizationByUser user=hpcuser01 start=2021-01-01T00:00:00 end=2024-02-20T23:59:59 -P | grep -iw -e billing -e cpu -e gres/gpu Output: cpu|979403 billing|1594902 gres/gpu|0 here I found that his utilisation is : 979403 I could like to understand why there is so much of a difference using two cmds to find out utilisation. Almost the difference is 979403 - 983352 = 3949 Request you to help me out with this. Regards, Prachi

1 0

Slurm Power Saving Guide: Why doesnt slurm mark as failed when resumeProgram returns =/= 0
by Xaver Stiensmeier 19 Feb '24

19 Feb '24

Dear slurm-user list, I had cases where our resumeProgram failed due to temporary cloud timeouts. In that case the resumeProgram returns a value =/= 0. Why does Slurm still wait until resumeTimeout instead of just accepting the startup as failed which then should lead to a rescheduling of the job. Is there some way to achieve the described effect i.e. tell Slurm: "You can stop waiting, the node won't come alive." or am I missing the correct way how this should be handled in Slurm? Best regards, Xaver

1 0

slurmdbd 17.02: "cluster not registered" (but things work)
by Matthias Leopold 19 Feb '24

19 Feb '24

Hi, I need to take care of a 17.02 Slurm cluster (I'm preparing it for upgrades). I see that slurmdbd logs various "cluster not registered" messages at startup (DBD_CLUSTER_TRES,DBD_JOB_START,DBD_STEP_START), but I don't see a real problem. Accounting works. Do I have to worry? Can this be related to upper/lower case issues with ClusterName? For some reason ClusterName is all upper case in slurm.conf. According to docs (and my experience with Slurm 21.08) though, this should be OK. Thx for help Matthias

1 0

Recover Batch Script Error
by Jason Simms 16 Feb '24

16 Feb '24

Hello all, I've used the "scontrol write batch_script" command to output the job submission script from completed jobs in the past, but for some reason, no matter which job I specify, it tells me it is invalid. Any way to troubleshoot this? Alternatively, is there another way - even if a manual database query - to recover the job script, assuming it exists in the database? sacct --jobs=38960 JobID JobName Partition Account AllocCPUS State ExitCode ------------ ---------- ---------- ---------- ---------- ---------- -------- 38960 amr_run_v+ tsmith2lab tsmith2lab 72 COMPLETED 0:0 38960.batch batch tsmith2lab 40 COMPLETED 0:0 38960.extern extern tsmith2lab 72 COMPLETED 0:0 38960.0 hydra_pmi+ tsmith2lab 72 COMPLETED 0:0 scontrol write batch_script 38960 job script retrieval failed: Invalid job id specified Warmest regards, Jason -- *Jason L. Simms, Ph.D., M.P.H.* Manager of Research Computing Swarthmore College Information Technology Services (610) 328-8102 Schedule a meeting: https://calendly.com/jlsimms

4 3

Need help managing licence
by Sylvain MARET 16 Feb '24

16 Feb '24

Hello everyone ! Recently our users bought a cplex dynamic license and want to use it on our slurm cluster. I've installed the paid version of cplex within modules so authorized user can load it with a simple module load cplex/2111 command but I don't know how to manage and ensure slurm doesn't launch a job if 20 people are already running code with this license. How do you guys manage paid licenses on your cluster ? Any advice would be appreciated ! Regards, Sylvain Maret

2 1

MPI/PMIx Issues after 23.11 Update
by Oliver Smith 16 Feb '24

16 Feb '24

Hi all, We’re running a small slurm dev cluster on Ubuntu and are facing issues with MPI/PMIx after upgrading slurm from 23.02.5 to 23.11.3. The first job step to use MPI within a job fails roughly 80% of the time but following attempts to use MPI within the same job work fine. For the failing job step we see this error after hitting the MPI timeout: slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_reset_if_to: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:741: 0x55a03f8d7a90: collective timeout seq=0 slurmstepd: error: mpi/pmix_v4: pmixp_coll_log: hpc-d-msh-01a02 [1]: pmixp_coll.c:286: Dumping collective state slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:759: 0x55a03f8d7a90: COLL_FENCE_RING state seq=0 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:762: my peerid: 1:hpc-d-msh-01a02 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:769: neighbor id: next 0:hpc-d-msh-01a01, prev 0:hpc-d-msh-01a01 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:778: Context ptr=0x55a03f8d7b08, #0, in-use=0 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:778: Context ptr=0x55a03f8d7b40, #1, in-use=0 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:778: Context ptr=0x55a03f8d7b78, #2, in-use=1 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:787: seq=0 contribs: loc=1/prev=1/fwd=0 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:791: neighbor contribs [2]: slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:824: done contrib: hpc-d-msh-01a01 slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:826: wait contrib: - slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:828: status=PMIXP_COLL_RING_FINILIZE slurmstepd: error: mpi/pmix_v4: pmixp_coll_ring_log: hpc-d-msh-01a02 [1]: pmixp_coll_ring.c:831: buf (offset/size): 0/33362 [hpc-d-msh-01a01.tds.hpc.barf1.com:47652] pml_ucx.c:178 Error: Failed to receive UCX worker address: Not found (-13) [hpc-d-msh-01a01.tds.hpc.barf1.com:47652] pml_ucx.c:477 Error: Failed to resolve UCX endpoint for rank 31 [hpc-d-msh-01a01:47652] *** An error occurred in MPI_Send [hpc-d-msh-01a01:47652] *** reported by process [683360612,0] [hpc-d-msh-01a01:47652] *** on communicator MPI_COMM_WORLD [hpc-d-msh-01a01:47652] *** MPI_ERR_OTHER: known error not in list [hpc-d-msh-01a01:47652] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, [hpc-d-msh-01a01:47652] *** and potentially your MPI job) slurmstepd: error: *** STEP 1890.0 ON hpc-d-msh-01a01 CANCELLED AT 2024-02-14T16:14:52 *** OpenMPI/PMIx versions have not changed and downgrading slurm to 23.02.5 seems to resolve the issue. We’d appreciate any pointers anyone might have. Thanks Oli This email comprises confidential information of Mercedes-Benz Grand Prix Limited ("MGP") unless it contains an explicit statement to the contrary made by an authorised representative of MGP. Contracts may only be concluded on behalf of MGP by its authorised signatories and not solely by email communication. No employee, agent, contractor, consultant and/or other representative of MGP is authorised to conclude any legally binding agreement on behalf of MGP by email alone without the express prior written confirmation of two authorised signatories of MGP. Mercedes-Benz Grand Prix Limited. Registered in England No. 787446. Registered Office at Mercedes-Benz Grand Prix Limited, Operations Centre, Brackley, Northants NN13 7BD. Note: The MGP Legal Department also acts on behalf of Mercedes-Benz Motorsport Limited ("MBM") and the above notice applies mutatis mutandis in respect of all email communications of MBM. MBM: Mercedes-Benz Motorsport Limited. Registered in England No. 13057973. Registered office at Mercedes-Benz Motorsport Limited, Lauda Drive, Brackley, Northants NN13 7BD. Please consider the environment before printing this email.

1 0

slurmctld: slurm_bufs_sendto(msg_type=SRUN_STEP_SIGNAL) failed: Connection reset by peer
by Rike-Benjamin Schuppner 16 Feb '24

16 Feb '24

Hi, I am getting the following error in the logs whenever I run a few srun jobs in a batch. Jan 25 11:24:03 slurmctl.XYZ slurmctld[272961]: slurmctld: debug: _send_timeout: Socket POLLERR: Connection reset by peer Jan 25 11:24:03 slurmctl.XYZ slurmctld[272961]: slurmctld: error: slurm_send_node_msg: [socket:[921897]] slurm_bufs_sendto(msg_type=SRUN_STEP_SIGNAL) failed: Connection reset by peer Jan 25 11:24:03 slurmctl.XYZ slurmctld[272961]: slurmctld: debug: laying out the 1 tasks on 1 hosts compute2 dist 1 The slurm version is 23.11.3 and an example sbatch file is: #!/bin/bash #SBATCH --job-name=slurm_test #SBATCH --mem=1gb #SBATCH --time=00:05:00 #SBATCH --output=slurm_test_%j.log pwd; hostname; date srun --exclusive -c1 -N1 -n1 bash -c "hostname ; sleep 10" & srun --exclusive -c1 -N1 -n1 bash -c "hostname ; sleep 10" & wait The more sruns I have in the script (&-backgrounded or not), the more the error shows up. Is there anything I could do to fix this error? Best /rike

2 1

Naive SLURM question: equivalent to LSF pre-exec
by Cutts, Tim 14 Feb '24

14 Feb '24

Hi, I apologise if I’ve failed to find this in the documentation (and am happy to be told to RTFM) but a recent issue for one of my users resulted in a question I couldn’t answer. LSF has a feature called a Pre-Exec where a script executes to check whether a node is ready to run a task. So, you can run arbitrary checks and go back to the queue if they fail. For example, if I have some automounted filesystems, and I want to be able to check for failure of the automounted, in an LSF world, I can do: bsub -E “test -f /nfs/someplace/file_I_know_exists” my_job.sh What’s the equivalent in SLURM? Thanks, Tim -- Tim Cutts Scientific Computing Platform Lead AstraZeneca Find out more about R&D IT Data, Analytics & AI and how we can support you by visiting our Service Catalogue<https://azcollaboration.sharepoint.com/sites/CMU993> | ________________________________ AstraZeneca UK Limited is a company incorporated in England and Wales with registered number:03674842 and its registered office at 1 Francis Crick Avenue, Cambridge Biomedical Campus, Cambridge, CB2 0AA. This e-mail and its attachments are intended for the above named recipient only and may contain confidential and privileged information. If they have come to you in error, you must not copy or show them to anyone; instead, please reply to this e-mail, highlighting the error to the sender and then immediately delete the message. For information about how AstraZeneca UK Limited and its affiliates may process information, personal data and monitor communications, please see our privacy notice at www.astrazeneca.com<https://www.astrazeneca.com>

3 2

job_submit.lua - uid in Docker cluster
by Loris Bennett 14 Feb '24

14 Feb '24

Hi, Having used https://github.com/giovtorres/slurm-docker-cluster successfully a couple of years ago to develop a job_submit.lua plugin, I am trying to do this again. However, the plugin which works on our current cluster (CentOS 7.9, Slurm 23.02.7) fails in the Docker cluster (Rocky 8.9, Slurm 23.02.7) with error: job_submit/lua: /etc/slurm/job_submit.lua: /usr/share/lua/5.3/posix/deprecated.lua:453: bad argument #1 to 'getpwuid' (int expected, got number) which occurs when submit_user = posix.getpasswd(submit_user.uid) The problems is that 'submit_user.uid' has the value 0.0 and is thus not an integer. The only user within the Docker cluster is 'root'. Has anyone come across this issue? Is it to do with the Docker environment or the difference in the OS versions (Lua 5.1.4 vs. 5.3.4, lua-posix 32 vs. 33.3.1)? Cheers, Loris -- Dr. Loris Bennett (Herr/Mr) FUB-IT (ex-ZEDAT), Freie Universität Berlin

1 0

Increasing SlurmdTimeout beyond 300 Seconds
by Andrew Baughan (ITCS - Staff) 12 Feb '24

12 Feb '24

Hi, We've been experiencing issues with network saturation on our older nodes caused by storage (GPFS) backups. This causes slurmctld to loose contact with slurmd on some compute nodes and for user jobs to be killed. While the longer term solution is to replace these and upgrade the network, I'm wondering if there are any ramifications, beyond nodes with genuine issues taking longer to get marked down, by increasing SlurmdTimeout. We've already applied a modest increase which has helped but not resolved the issue and wondering if we should push it further in the interim. Kind Regards Andy Baughan HPC Systems Developer

4 3

simple question, I guess… from a newbie sysadmin
by Richard Randriatoamanana 12 Feb '24

12 Feb '24

Hi, I am trying to help a sysadmin colleague (and to understand for myself) trying to configure a new slurm server and he struggles to understand if there is an alternative way to config slurm managing job policy submission per user without necessarily installing an accounting mariadb service. I said no but maybe I misunderstand slurm admin configuration. Thanks in advance for feedback. Bests, Richard — Sent from my rotary and brandy phone ! Apologies for typos and any spelling errors

2 1

Question on using a suffix with "squeue --Format=FORMAT"
by urbanjost 11 Feb '24

11 Feb '24

I do not see how to use the suffix for the --Format option of the squeue(1) command. The description does not indicate how to tell the end of the size and period components and the beginning of the suffix, but assuming the suffix must being with something other than a period or numeric character I found all suffix values produce a null separator. A wanted to use --Format somewhat like --format=%all, producing all fields. As the --Format did not have an equivalent descriptor, (ie. something like --Format=All ) it looked like it would be required to list all the allowed keywords with a suffix of "|", which from the description I assumed would be as simple as a longer version of something like thi: squeue --Format='STDOUT:|,STDERR:|,STDIN:|,WorkDir:|' Given the description: -O, --Format=<output_format> Specify the information to be displayed. Also see the -o, --format=<output_format> option described above (which supports greater flexibility in formatting, but does not support access to all fields because we ran out of letters). Requests a comma separated list of job information to be displayed. The format of each field is "type[:[.][size][suffix]]" size Minimum field size. If no size is specified, 20 char‐ acters will be allocated to print the information. . Indicates the output should be right justified and size must be specified. By default output is left justified. suffix Arbitrary string to append to the end of the field. Note that many of these type specifications are valid only for jobs while others are valid only for job steps. Valid type specifications include: Even something like this produces no error but the suffix is always a null string: squeue --Format='STDOUT:SUFFIX,STDERR:SUFFIXSTDIN:SUFFIXWorkDir:SUFFIX So if I want to generate output with squeue --Format=FORMAT with a vertical line character between the output fields what is the proper syntax? A test bash(1) script where I could not get the SUFFIX variable to work as expected ... #!/bin/bash #@(#) Build value for squeue --Format='FORMAT' that includes all available fields with a vertical bar separating each field GOODNAMES='' BADNAMES='' SUFFIX='SUFFIX' GOODNAMES='' BADNAMES='' SUFFIX='1000:|' GOODNAMES='' BADNAMES='' SUFFIX='|' for NAME in STDOUT Account AccrueTime admin_comment AllocNodes AllocSID ArrayJobID ArrayTaskID AssocID BatchFlag BatchHost BoardsPerNode BurstBuffer BurstBufferState Cluster ClusterFeature Command Comment Container ContainerID Contiguous Cores CoreSpec CPUFreq cpus-per-task cpus-per-tres Deadline DelayBal Licenses MaxCPUs MaxNodes MCSLabel mem-per-tres MinCpus MinMemory MinTime Mintition PendingTime PreemptTime Prefer Priority PriorityLong Profile QOS Reason Riable SiblingsViableRaw Sockets SPerBoard StartTime State StateCompact STDERR STres-per-node tres-per-socket tres-per-step tres-per-task UserID UserName Wait4Sw do if squeue --Format="$NAME" >/dev/null then GOODNAMES="$GOODNAMES${NAME}:${SUFFIX}," else BADNAMES="$BADNAMES${NAME}," fi done 2>&1 #FORMAT=$(tr --squeeze-repeats ' ' ' '<<<$GOODNAMES) FORMAT=${GOODNAMES%%,} # trim off trailing comma read r c < <(stty size) # get screen width for use with cut(1) # try it squeue --Format="$FORMAT"|column -t|cut -c 1-$c # echo it echo "$FORMAT" exit Sent with [Proton Mail](https://proton.me/) secure email.

1 0

question on sbatch --prefer
by Alan Stange 10 Feb '24

10 Feb '24

Hello all, I'm somewhat new to Slurm, but long time user of other batch systems. Assume we have a simple cluster of uniform racks of systems with no special resources, and our jobs are all single cpu tasks. Lets say I have a long running job in the cluster, which needs to spawn a helper process into the cluster. We have a strong preference for this helper to run on the same cluster node as the original job, but if that node is already scheduled full, then we want this new task to be scheduled on another systems without any delay. The problem I have is that the --nodelist doesn't solve this, and, as far as I can tell, there's no option with --prefer to specify a node name as a resource, without creating a gres for every hostname in the cluster. It seems like what I'm trying to do should be achievable, but having read though the documentation and searched the archives of this list, I'm not seeing a solution. I'm hoping someone here has some experience with this and can point me in the right direction. Sincerely, Alan

3 3

Compilation question
by Sylvain MARET 10 Feb '24

10 Feb '24

Hello everyone ! I'm trying to compile slurm 22.05.11 on Rocky linux 8.7 with freeipmi support I've seen the documentation so I've done the configure step : ./configure --with-pmix=$PMIXHOME --with-ucx=$UCXHOME --with-nvml=$NVMLHOME --prefix=$SLURMHOME --with-freeipmi=/usr but when I run make I end up with the following error : /bin/sh ../../../../../libtool --tag=CC --mode=link gcc -DNUMA_VERSION1_COMPATIBILITY -g -O2 -fno-omit-frame-pointer -pthread -ggdb3 -Wall -g -O1 -fno-strict-aliasing -export-dynamic -L/usr/lib64 -lhdf5_hl -lhdf5 -lsz -lz -ldl -lm -o sh5util sh5util.o -Wl,-rpath=/softs/batch/slurm/22.05.11/lib/slurm -L../../../../../src/api/.libs -lslurmfull -ldl ../libhdf5_api.la -lpthread -lm -lresolv libtool: link: gcc -DNUMA_VERSION1_COMPATIBILITY -g -O2 -fno-omit-frame-pointer -pthread -ggdb3 -Wall -g -O1 -fno-strict-aliasing -o .libs/sh5util sh5util.o -Wl,-rpath=/softs/batch/slurm/22.05.11/lib/slurm -Wl,--export-dynamic -L/usr/lib64 -L../../../../../src/api/.libs /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so ../.libs/libhdf5_api.a -lhdf5_hl -lhdf5 -lsz -lz -ldl -lpthread -lm -lresolv -pthread -Wl,-rpath -Wl,/softs/batch/slurm/22.05.11/lib/slurm sh5util.o:(.init_array+0x0): undefined reference to `__nv_init_env' sh5util.o:(.init_array+0x8): undefined reference to `__flushz' sh5util.o:(.init_array+0x10): undefined reference to `__daz' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_transfer_unique' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_sort_key_pairs' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_xstrchr' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_unsetenvp' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_sort' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_for_each' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__builtin__pgi_isnanld' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_get_extra_conf_path' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__blt_pgi_ctzll' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_running_in_slurmctld' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__c_mcopy1' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__blt_pgi_clzll' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_create' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_count' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__builtin_va_gparg1' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_destroy_config_key_pair' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_xfree_ptr' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_getenvp' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_free_buf' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_get_log_level' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__c_mset8' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_xstrdup_printf' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_delete_first' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_append' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_error' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_init_buf' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_log_var' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_s_p_parse_file' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_xstrdup' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_fatal' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `__blt_pgi_popcountll' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_xstrcmp' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_s_p_hashtbl_destroy' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_s_p_hashtbl_create' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_setenvpf' /softs/batch/slurm/slurm-22.05.11/src/api/.libs/libslurmfull.so: undefined reference to `slurm_list_destroy' collect2: error: ld returned 1 exit status What can I do to resolve these undefined reference error ? Regards, Sylvain Maret

3 2

Re: Selecting only a subset of GPU's from all available GPU's
by Minulakshmi S 10 Feb '24

10 Feb '24

Hi Loris, I have different kind of GPU's in same node , and I beleive feature applies to particular node , and cant be applied to only a few GPU's connected to a single node. On Mon, 18 Dec 2023, 12:14 Loris Bennett, <loris.bennett(a)fu-berlin.de> wrote: > Hi Minu, > > Minulakshmi S <minulakshmi.s(a)gmail.com> writes: > > > I'm submitting jobs to a cluster via SLURM scheduler, and let's say I > have access to 8 GPUs in my cluster in same node. They are GPUs of type > A,B,C,D,E,F,G,H. I > > would like to submit a job that requests the use of GPUs of type A or B > or C but NOT of type D/E/F/G/H. So I need some type of OR logic with the > --gres flag. > > > > Eg : When I request GPU of type A , I can do sbatch –gres=gpu:TypeA:1, > I need to input a subset of GPU’s and let slurm schedule job utilizing one > of the GPU > > from this allowed list. > > > > Regards > > Minu > > Assuming the GPUs within a node are all of the same type, could you define > a feature for each GPU type, assign the features to the appropriate > nodes, and then run the job with > > --contraint=gpu_TypeA --gres=gpu:1 > > ? > > This is obviously rather clunky and it would be much nicer if > multiple GPUs types passed to '--gres' were ORed. > > Cheers, > > Loris > > -- > Dr. Loris Bennett (Herr/Mr) > ZEDAT, Freie Universität Berlin >

1 0

Memory used per node
by Gerhard Strangar 09 Feb '24

09 Feb '24

Hello, I'm wondering if there's a way to tell how much memory my job is using per node. I'm doing #SBATCH -n 256 srun solver inputfile When I run sacct -o maxvmsize, the result apparently is the maxmimum VSZ of the largest solver process, not the maximum of the sum of them all (unlike when calling mpirun instead). When I sstat -o TresUsageInMax, I get the memory summed up over all nodes being used. Can I get the maximum VSZ per node? Gerhard

2 1

Could not find group with gid even when they exist
by Nic Lewis 09 Feb '24

09 Feb '24

After upgrading to version 23.11.3 we started to get slammed with the following log messages from slurmctld "error: validate_group: Could not find group with gid <id>" This spans a handful of groups and repeats constantly, drowning out just about everything else. Attempting to do a lookup on the group shows that they exist on the scheduler node, same for all the submission and compute nodes. As far as I can tell, slurm should be able to locate the group in question. Jobs submitted from users within those groups go through just fine. They get scheduled, run, and clean up no problem. I'm at a loss on where to look next.

1 1

Relative QOS limits in sacctmgr?
by Chip Seraphine 09 Feb '24

09 Feb '24

Hello, TL,DR: How does the relative QOS flag work? I have a QOS and I want it to be collectively restricted to 50% of the reachable cores in the cluster. I’ve been managing this by dividing my core count to 2 to get N, and doing ‘sacctmgr update qos foobar set MaxTRES=cpu=N’. That’s fine, except that N changes frequently as worker nodes come and go, so I have an hourly cronjob that enforces this. The problem with *that* is the values you back out with ‘sacctmgr show qos’ are often in different units than you input (e.g. gigs instead of megs of ram), which means I need to do some annoying parsing, will miss minor rounding errors, etc. (Using JSON solves that problem but introduces others that I do not want to deal with.) So while looking for a better way I noticed that QOSs support a flag called “relative”, which says “the QOS limits will be treated as percentages of a cluster/partition instead of absolutes”. Perfect! However, it’s not very clear to me what that means in practice. For example, with the “relative” flag set, the MaxTRES units still seem to work the same, expect metric suffices (“G”, ”M”, etc) and won’t accept periods. Is anyone actually using this flag, and if so can you tell me what attributes it affects, and in what way? -- Chip Seraphine Grid Operations For support please use help-grid in email or slack. This e-mail and any attachments may contain information that is confidential and proprietary and otherwise protected from disclosure. If you are not the intended recipient of this e-mail, do not read, duplicate or redistribute it by any means. Please immediately delete it and any attachments and notify the sender that you have received it by mistake. Unintended recipients are prohibited from taking action on the basis of information in this e-mail or any attachments. The DRW Companies make no representations that this e-mail or any attachments are free of computer viruses or other defects.

1 0

Is there a way to list allocated/unallocated resources defined in a QoS?
by Alastair Neil 09 Feb '24

09 Feb '24

Slurm version 23.02.07 If I have a QoS defined that has a set number of say GPU devices set in the GrpTRES. Is there an easy way to generate a list of how much of the defined quota is allocated or conversely un-allocated? e.g.: Name|Priority|GraceTime|Preempt|PreemptExemptTime|PreemptMode|Flags|UsageThres|UsageFactor|GrpTRES|GrpTRESMins|GrpTRESRunMins|GrpJobs|GrpSubmit|GrpWall|MaxTRES|MaxTRESPerNode|MaxTRESMins|MaxWall|MaxTRESPU|MaxJobsPU|MaxSubmitPU|MaxTRESPA|MaxJobsPA|MaxSubmitPA|MinTRES| normal|0|00:00:00|||cluster|||1.000000|||||||||||cpu=3000,gres/gpu=20||||||| dept1|1|00:00:00|||cluster|||1.000000|cpu=256,gres/gpu:1g.10gb=16,gres/gpu:2g.20gb=8,gres/gpu:3g.40gb=8,gres/gpu:a100.80gb=8||||||||||||||||| dept2|1|00:00:00|||cluster|||1.000000|cpu=256,gres/gpu:1g.10gb=0,gres/gpu:2g.20gb=0,gres/gpu:3g.40gb=0,gres/gpu:a100.80gb=16||||||||||||||||| So dept1 and dept2 qos are set on the same partition. How can a user with access to one or other see if there are available resources in the partition?

2 1

Errors upgrading to 23.11.0
by Xaver Stiensmeier 08 Feb '24

08 Feb '24

Dear slurm-user list, I got this error: Unable to start service slurmctld: Job for slurmctld.service failed because the control process exited with error code.\nSee \"systemctl status slurmctld.service\" and \"journalctl -xeu slurmctld.service\" for details. but in slurmctld.service I see nothing suspicious: slurmctld.service - Slurm controller daemon Loaded: loaded (/lib/systemd/system/slurmctld.service; enabled; vendor preset: enabled) Drop-In: /etc/systemd/system/slurmctld.service.d └─override.conf Active: active (running) since Wed 2024-02-07 15:50:56 UTC; 19min ago Main PID: 51552 (slurmctld) Tasks: 21 (limit: 9363) Memory: 10.4M CPU: 1min 16.088s CGroup: /system.slice/slurmctld.service ├─51552 /usr/sbin/slurmctld --systemd └─51553 "slurmctld: slurmscriptd" "" "" "" "" "" "" Feb 07 15:58:21 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: sched: _slurm_rpc_allocate_resources JobId=3 NodeList=(null) usec=959 Feb 07 15:58:23 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _job_complete: JobId=3 WTERMSIG 2 Feb 07 15:58:23 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _job_complete: JobId=3 cancelled by interactive user Feb 07 15:58:23 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _job_complete: JobId=3 done Feb 07 15:58:23 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _slurm_rpc_complete_job_allocation: JobId=3 error Job/step already completing or completed Feb 07 15:58:42 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: sched: _slurm_rpc_allocate_resources JobId=4 NodeList=cluster-master-2vt2bqh7ahec04c,cluster-worker-2vt2bqh7ahec04c-2 usec=512 Feb 07 16:06:04 cluster-master-2vt2bqh7ahec04c slurmctld[51553]: slurmctld: error: _run_script: JobId=0 resumeprog exit status 1:0 Feb 07 16:09:33 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _job_complete: JobId=4 WTERMSIG 2 Feb 07 16:09:33 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _job_complete: JobId=4 done Feb 07 16:09:33 cluster-master-2vt2bqh7ahec04c slurmctld[51552]: slurmctld: _slurm_rpc_complete_job_allocation: JobId=4 error Job/step already completing or completed I am unsure how to debug this further. It might be coming from a previous problem I tried to fix (basically a few deprecated keys in the configuration). I will try to restart the entire cluster with the added changes to rule out any follow up errors, but maybe it's something obvious a fellow list user can see. Best regards, Xaver

2 2

sinfo gresused shard count wrong/incomplete
by Reed Dier 07 Feb '24

07 Feb '24

I have a bash script that grabs current statistics from sinfo to ship into a time series database to use for Grafana dashboards. We recently began using shards with our gpus, and I’m seeing some unexpected behavior with the data reported from sinfo. > $ sinfo -h -O "NodeHost:5 ,GresUsed:100 ,Gres:100" | grep gpu03 > gpu03 gpu:p40:0(IDX:N/A),gpu:rtx:0(IDX:N/A),shard:p40:1(IDX:1),shard:rtx:0(IDX:N/A) gpu:p40:6(S:0),gpu:rtx:2(S:0),shard:p40:36(S:0),shard:rtx:12(S:0) > > $ scontrol show node gpu03 > NodeName=gpu03 Arch=x86_64 CoresPerSocket=22 > CPUAlloc=52 CPUEfctv=88 CPUTot=88 CPULoad=43.67 > AvailableFeatures=avx,avx512,largeMemory,matlab > ActiveFeatures=avx,avx512,largeMemory,matlab > Gres=gpu:p40:6(S:0),gpu:rtx:2(S:0),shard:p40:36(S:0),shard:rtx:12(S:0) > NodeAddr=gpu03 NodeHostName=gpu03 Version=22.05.8 > OS=Linux 5.4.0-164-generic #181-Ubuntu SMP Fri Sep 1 13:41:22 UTC 2023 > RealMemory=768000 AllocMem=425984 FreeMem=116850 Sockets=2 Boards=1 > State=MIXED ThreadsPerCore=2 TmpDisk=0 Weight=1 Owner=N/A MCS_label=N/A > Partitions=gpu > BootTime=2024-02-02T14:17:07 SlurmdStartTime=2024-02-05T12:33:50 > LastBusyTime=2024-02-07T09:48:50 > CfgTRES=cpu=88,mem=750G,billing=88,gres/gpu=8,gres/gpu:p40=6,gres/gpu:rtx=2,gres/shard=48 > AllocTRES=cpu=52,mem=416G,gres/shard=21 > CapWatts=n/a > CurrentWatts=0 AveWatts=0 > ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/s Specifically, we can see there are 48 total shards on this specific node, 36 of type p40, 12 of type rtx. From control, I can see that 21 shards are in use, and from nvidia-smi I can see that the breakdown of shards per gpu index is: [0]p40:4/6 [1]p40:6/6 [2]rtx:4/6 [3]rtx:3/6 [4]p40:4/6 [5]p40:0/6 [6]p40:0/6 [7]p40:0/6 So, I can tell that what “GresUsed” is reporting is actually an entire gpu’s worth of shards used, and the index of that specific entire GPU consumed. However, what it doesn’t show is the amount of specific gres/shard actually used, or more succinctly, I can’t seem to extrapolate that there are 21 shards in use. Right now my script says there is 1 shard out of 48 used, because I’m scraping for the num of shards in use, which is reporting the value of entire GPUs used by shards, and anything less than 100% is rounded to 0.. So, given that I’m on 22.05.8, maybe its better in 23.02.X which I’m hoping to move to within the next month. However I can’t seem to find anything in the release notes for 23.02 or 23.11 that would imply that sinfo reports (my) expected value of the actual count of shards used, Does anyone have any ideas for how I might be able to achieve what I’m looking for using sinfo, or should I instead try to use the sinfo json parser which has a “tres_used” field, that doesn’t appear to be accessible outside of the json output? > { > "architecture": "x86_64", > "burstbuffer_network_address": "", > "boards": 1, > "boot_time": 1706901428, > "comment": "", > "cores": 22, > "cpu_binding": 0, > [SNIP] > "tres": "cpu=88,mem=750G,billing=88,gres/gpu=8,gres/gpu:p40=6,gres/gpu:rtx=2,gres/shard=48", > "slurmd_version": "22.05.8", > "alloc_memory": 284380, > "alloc_cpus": 62, > "idle_cpus": 26, > "tres_used": "cpu=62,mem=284380M,gres/gpu=7,gres/gpu:p40=5,gres/gpu:rtx=2,gres/shard=17", > "tres_weighted": 62 > }, Any ideas appreciated. Thanks, Reed

1 0

Re: Restricting local disk storage of jobs
by Tim Schneider 07 Feb '24

07 Feb '24

Hey Jeffrey, thanks for this suggestion! This is probably the way to go if one can find a way to access GRES in the prolog. I read somewhere that people were calling scontrol to get this information, but this seems a bit unclean. Anyway, if I find some time I will try it out. Best, Tim On 2/6/24 16:30, Jeffrey T Frey wrote: > Most of my ideas have revolved around creating file systems on-the-fly > as part of the job prolog and destroying them in the epilog. The > issue with that mechanism is that formatting a file system (e.g. > mkfs.<type>) can be time-consuming. E.g. formatting your local > scratch SSD as an LVM PV+VG and allocating per-job volumes, you'd > still need to run a e.g. mkfs.xfs and mount the new file system. > > > ZFS file system creation is much quicker (basically combines the LVM + > mkfs steps above) but I don't know of any clusters using ZFS to manage > local file systems on the compute nodes :-) > > > One /could/ leverage XFS project quotas. E.g. for Slurm job 2147483647: > > > *[root@r00n00 /]# mkdir /tmp-alloc/slurm-2147483647* > *[root@r00n00 /]# xfs_quota -x -c 'project -s -p > /tmp-alloc/slurm-2147483647 2147483647' /tmp-alloc* > Setting up project 2147483647 (path /tmp-alloc/slurm-2147483647)... > Processed 1 (/etc/projects and cmdline) paths for project > 2147483647 with recursion depth infinite (-1). > *[root@r00n00 /]# xfs_quota -x -c 'limit -p bhard=1g 2147483647' > /tmp-alloc* > *[root@r00n00 /]# cd /tmp-alloc/slurm-2147483647* > *[root@r00n00 slurm-2147483647]# dd if=/dev/zero of=zeroes bs=5M > count=1000* > dd: error writing ‘zeroes’: No space left on device > 205+0 records in > 204+0 records out > 1073741824 bytes (1.1 GB) copied, 2.92232 s, 367 MB/s > > : > > [root@r00n00 /]# rm -rf /tmp-alloc/*slurm-2147483647* > [root@r00n00 /]# *xfs_quota -x -c 'limit -p bhard=0 2147483647' > /tmp-alloc* > > > Since Slurm jobids max out at 0x03FFFFFF (and 2147483647 = 0x7FFFFFFF) > we have an easy on-demand project id to use on the file system. Slurm > tmpfs plugins have to do a mkdir to create the per-job directory, > adding two xfs_quota commands (which run in more or less O(1) time) > won't extend the prolog by much. Likewise, Slurm tmpfs plugins have to > scrub the directory at job cleanup, so adding another xfs_quota > command will not do much to change their epilog execution times. The > main question is "where does the tmpfs plugin find the quota limit for > the job?" > > > > > >> On Feb 6, 2024, at 08:39, Tim Schneider via slurm-users >> <slurm-users(a)lists.schedmd.com> wrote: >> >> Hi, >> >> In our SLURM cluster, we are using the job_container/tmpfs plugin to >> ensure that each user can use /tmp and it gets cleaned up after them. >> Currently, we are mapping /tmp into the nodes RAM, which means that >> the cgroups make sure that users can only use a certain amount of >> storage inside /tmp. >> >> Now we would like to use of the node's local SSD instead of its RAM >> to hold the files in /tmp. I have seen people define local storage as >> GRES, but I am wondering how to make sure that users do not exceed >> the storage space they requested in a job. Does anyone have an idea >> how to configure local storage as a proper tracked resource? >> >> Thanks a lot in advance! >> >> Best, >> >> Tim >> >> >> -- >> slurm-users mailing list -- slurm-users(a)lists.schedmd.com >> To unsubscribe send an email to slurm-users-leave(a)lists.schedmd.com >

2 1

scheme for protected GPU jobs from preemption
by Paul Raines 06 Feb '24

06 Feb '24

After using just Fairshare for over a year on our GPU cluster, we have decided it is not working for us for what we really want to achieve among our groups. We have decided to look at preemption. What we want is for users to NOT have a #job/GPU maximum (if they are only person on the cluster they should be able to use it all), but if another user comes to the "full" cluster they should immediately be able to run some jobs. Thus preemption is needed. In our scheme we want * users to have N protected GPU jobs that cannot be preempted where N is the number of GPUs allocated. * N may not be the same for all users. Some priviledged users get more. * jobs pending in the queue will have lower priority dependent on the number of GPUs allocated to running jobs. Maybe doable somehow with PriorityWeightJobSize though not sure how. * Jobs over N are subject to preemption (and requeued if --requeue is given) with shortest running jobs of the user with most unprotected GPUs preempted first. * another complication is we have a variety of different GPUs and users may ask for specific ones which can limit what unprotected GPU jobs are available for preemption My first attempt to do this in SLURM was to just create two partitions, GPU and GPU-req, with different PriorityTier values and the later partition have PreemptMode=REQUEUE. But N would be set by a MaxTRES on the first partition and be the same for everyone and we need it to be INDEPENDENT for each user. Also users would have to "think" about which partition to submit jobs to. And users want their longest running "unprotected" job to be able to be PROMOTED to a "protected" jobs automatically when a "protected" job finishes. However slurm does not allow running jobs to move between partitions. I am trying to figure out QOS pre-emption which might solve the independent N per user issue but I don't think it will solve the promotion issue. Any ideas how this scheme might be possible in SLURM? Otherwise I might have to write a complicated cron job that tries to do it all "outside" of SLURM issuing scontrol commands. --------------------------------------------------------------- Paul Raines http://help.nmr.mgh.harvard.edu MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging 149 (2301) 13th Street Charlestown, MA 02129 USA The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Mass General Brigham Compliance HelpLine at https://www.massgeneralbrigham.org/complianceline <https://www.massgeneralbrigham.org/complianceline> . Please note that this e-mail is not secure (encrypted). If you do not wish to continue communication over unencrypted e-mail, please notify the sender of this message immediately. Continuing to send or respond to e-mail after receiving this message means you understand and accept this risk and wish to continue to communicate over unencrypted e-mail.

1 0

Restricting local disk storage of jobs
by Tim Schneider 06 Feb '24

06 Feb '24

Hi, In our SLURM cluster, we are using the job_container/tmpfs plugin to ensure that each user can use /tmp and it gets cleaned up after them. Currently, we are mapping /tmp into the nodes RAM, which means that the cgroups make sure that users can only use a certain amount of storage inside /tmp. Now we would like to use of the node's local SSD instead of its RAM to hold the files in /tmp. I have seen people define local storage as GRES, but I am wondering how to make sure that users do not exceed the storage space they requested in a job. Does anyone have an idea how to configure local storage as a proper tracked resource? Thanks a lot in advance! Best, Tim

3 2

Re: [ext] Restricting local disk storage of jobs
by Tim Schneider 06 Feb '24

06 Feb '24

Hi Magnus, I understand. Thanks a lot for your suggestion. Best, Tim On 06.02.24 15:34, Hagdorn, Magnus Karl Moritz wrote: > Hi Tim, > in the end the InitScript didn't contain anything useful because > > slurmd: error: _parse_next_key: Parsing error at unrecognized key: > InitScript > > At this stage I gave up. This was with SLURM 23.02. My plan was to > setup the local scratch directory with XFS and then get the script to > apply a project quota, ie quota attached to the directory. > > I would start by checking if slurm recognises the InitScript option. > > Regards > magnus > > On Tue, 2024-02-06 at 15:24 +0100, Tim Schneider wrote: >> Hi Magnus, >> >> thanks for your reply! If you can, would you mind sharing the >> InitScript >> of your attempt at getting it to work? >> >> Best, >> >> Tim >> >> On 06.02.24 15:19, Hagdorn, Magnus Karl Moritz wrote: >>> Hi Tim, >>> we are using the container/tmpfs plugin to map /tmp to a local NVMe >>> drive which works great. I did consider setting up directory >>> quotas. I >>> thought the InitScript [1] option should do the trick. Alas, I >>> didn't >>> get it to work. If I remember correctly, slurm complained about the >>> option being present. In the end we recommend our users to make >>> exclusive use a node if they are going to use a lot of local >>> scratch >>> space. I don't think this happens very often if at all. >>> Regards >>> magnus >>> >>> [1] >>> https://slurm.schedmd.com/job_container.conf.html#OPT_InitScript >>> >>> >>> On Tue, 2024-02-06 at 14:39 +0100, Tim Schneider via slurm-users >>> wrote: >>>> Hi, >>>> >>>> In our SLURM cluster, we are using the job_container/tmpfs plugin >>>> to >>>> ensure that each user can use /tmp and it gets cleaned up after >>>> them. >>>> Currently, we are mapping /tmp into the nodes RAM, which means >>>> that >>>> the >>>> cgroups make sure that users can only use a certain amount of >>>> storage >>>> inside /tmp. >>>> >>>> Now we would like to use of the node's local SSD instead of its >>>> RAM >>>> to >>>> hold the files in /tmp. I have seen people define local storage >>>> as >>>> GRES, >>>> but I am wondering how to make sure that users do not exceed the >>>> storage >>>> space they requested in a job. Does anyone have an idea how to >>>> configure >>>> local storage as a proper tracked resource? >>>> >>>> Thanks a lot in advance! >>>> >>>> Best, >>>> >>>> Tim >>>> >>>>

1 0

Re: [ext] Restricting local disk storage of jobs
by Tim Schneider 06 Feb '24

06 Feb '24

Hi Magnus, thanks for your reply! If you can, would you mind sharing the InitScript of your attempt at getting it to work? Best, Tim On 06.02.24 15:19, Hagdorn, Magnus Karl Moritz wrote: > Hi Tim, > we are using the container/tmpfs plugin to map /tmp to a local NVMe > drive which works great. I did consider setting up directory quotas. I > thought the InitScript [1] option should do the trick. Alas, I didn't > get it to work. If I remember correctly, slurm complained about the > option being present. In the end we recommend our users to make > exclusive use a node if they are going to use a lot of local scratch > space. I don't think this happens very often if at all. > Regards > magnus > > [1] > https://slurm.schedmd.com/job_container.conf.html#OPT_InitScript > > > On Tue, 2024-02-06 at 14:39 +0100, Tim Schneider via slurm-users wrote: >> Hi, >> >> In our SLURM cluster, we are using the job_container/tmpfs plugin to >> ensure that each user can use /tmp and it gets cleaned up after them. >> Currently, we are mapping /tmp into the nodes RAM, which means that >> the >> cgroups make sure that users can only use a certain amount of storage >> inside /tmp. >> >> Now we would like to use of the node's local SSD instead of its RAM >> to >> hold the files in /tmp. I have seen people define local storage as >> GRES, >> but I am wondering how to make sure that users do not exceed the >> storage >> space they requested in a job. Does anyone have an idea how to >> configure >> local storage as a proper tracked resource? >> >> Thanks a lot in advance! >> >> Best, >> >> Tim >> >>

2 1

Starting a job after a file is created in previous job (dependency looking for soluton)
by Amjad Syed 06 Feb '24

06 Feb '24

Hello I have the following scenario: I need to submit a sequence of up to 400 jobs where the even jobs depend on the preceeding odd job to finish and every odd job depends on the presence of a file generated by the preceding even job (availability of the file for the first of those 400 jobs is guaranteed). If I just submit all those jobs via a loop using dependencies, then I end up with a lot of pending jobs who might later not even run because no output file has been produced by the preceding jobs. Is there a way to pause the submission loop until the required file has been generated so that at most two jobs are submitted at the same time? Here is a sample submission script showing what I want to achieve. for i in {1..200}; do FILE=GHM_paramset_${i}.dat # How can I pause the submission loop until the FILE has been created???? #if test -f "$FILE"; then jobid4=$(sbatch --parsable --dependency=afterok:$jobid3 job4_sub $i) jobid3=$(sbatch --parsable --dependency=afterok:$jobid4 job3_sub $i) #fi done Any help will be appreciated Amjad

3 2

SLURM configuration for LDAP users
by Richard Chang 05 Feb '24

05 Feb '24

Hi, I am a little new to this, so please pardon my ignorance. I have configured slurm in my cluster and it works fine with local users. But I am not able to get it working with LDAP/SSSD authentication. User logins using ssh are working fine. An LDAP user can login to the login, slurmctld and compute nodes, but when they try to submit jobs, slurmctld logs an error about invalid account or partition for user. Someone said we need to add the user manually into the database using the sacctmgr command. But I am not sure we need to do this for each and every LDAP user. Yes, it does work if we add the LDAP user manually using sacctmgr. But I am not convinced this manual way is the way to do. The documentation is not very clear about using LDAP accounts. Saw somewhere in the list about using UsePAM=1 and copying or creating a softlink for slurm PAM module under /etc/pam.d . But it didn't work for me. Saw somewhere else that we need to specifying LaunchParameters=enable_nss_slurm in the slurm.conf file and put slurm keyword in passwd/group entry in the /etc/nsswitch.conf file. Did these, but didn't help either. I am bereft of ideas at present. If anyone has real world experience and can advise, I will be grateful. Thank you, Richard

4 4

timeout of get_user_env does not obey time limit
by urbanjost 03 Feb '24

03 Feb '24

If I use the sbatch(1) option --export=NONE or wipe the environment with "env -i /usr/bin/sbatch ..." or use --export=NIL then the environment is not properly constructed and I see the message in the /var/log/*slurm* files: [2024-02-03T11:50:33.052] _get_user_env: get env for user jsu here [2024-02-03T11:52:33.152] timeout waiting for /bin/su to complete [2024-02-03T11:52:34.152] error: Failed to load current user environment variables [2024-02-03T11:52:34.153] error: _get_user_env: Unable to get user's local environment, running only with passed environment This occurs at 120 seconds not matter if I add --get-user-env=3600 or adjust many slurm.conf time-related parameters. It is easy to reproduce by adding "sleep 100" into a .cshrc file and sbatch(1) the file #!/bin/csh #SBATCH --export=NONE --propagate=NONE --get-user-env=3600L printenv HOME printenv USER printenv PATH env I have adjust MANY time-related limits in the slurm.conf file to no avail. When the system is unresponsive or heavily loaded or users have prologues that set up complex environments via module commands (which can be notoriously slow) the jobs are failing or producing errors. If I configure Slurm so that jobs that timeout requeue instead of running then a user with a slow login setup can submit a large number of jobs and basically close down a cluster because this option not only requeues jobs that fail but puts the node it occurred on in a DRAIN state. We see this as very dangerous as by defaults jobs proceed to execute even when their environment is not properly constructed. I can see that "slurmrestd getenv" and the procedure get_user_env(3c) are involved, but a preliminary scan of the code looked like the --get-user-env=NNNN value was being parsed, and I did not see a reason the setup always times out at 120 seconds (at least on my system). Does anyone know how to get the time allowed to get the default user environment to use the value on the --get-user-env option when no environment is being exported to a job? This is showing up sporadically and causing intermittent failures that are very confusing and disturbing to the users it occurs with. Sent with [Proton Mail](https://proton.me/) secure email.

1 0

selecting job-specific log messages
by urbanjost 03 Feb '24

03 Feb '24

log files use many strings to identify job, including not having a jobID in the message NUMBER=$SLURM_JOBID egrep "\.\<$NUMBER\>\] |\<$NUMBER\>\.batch|jobid \<$NUMBER\>|JObId=\<$NUMBER\>|job id \<$NUMBER\>|job\.\<$NUMBER\>|job \<$NUMBER\>|jobid \[\<$NUMBER\>\]|task_p_slurmd_batch_request: \<$NUMBER\>" /var/log/slurm* Even that misses cruciall data that does not even contain the jobid [2024-02-03T11:50:33.052] _get_user_env: get env for user jsu here [2024-02-03T11:52:33.152] timeout waiting for /bin/su to complete [2024-02-03T11:52:34.152] error: Failed to load current user environment variables [2024-02-03T11:52:34.153] error: _get_user_env: Unable to get user's local environment, running only with passed environment It would be very useful if all messages related to a job had a consistent string in them for grepping the log files; even better might be a command like "scontrol show jobid=NNNN log_messages But I could not find what I wanted (an easy way to find all daemon log messages related to a specific job). I would find it particularly useful if there were a way to automatically append such information to the stdout of the job at job termination so users would automatically get information about job failures or warnings. Is there such a feature available I have missed? Sent with [Proton Mail](https://proton.me/) secure email.

1 0

after upgrade to 23.11.1 nodes stuck in completion state
by Paul Raines 01 Feb '24

01 Feb '24

I finally had downtime on our cluster running 20.11.3 and decided to upgrade SLURM. All daemons were stopped on nodes and master. Rocky 8 Linux OS was updated but not changed configuration-wise in anyway. On the master, when I first installed 23.11.1 and tried to run slurmdbd -D -vvv at the command line, it balked as the more than 2 major version jump. I then installed 22.05.6 and run slurmdbd -D -vvv which took a long time but completed fine. I reinstalled 23.11.1 and did slurmdbd -D --vvv which went quick. I stopped it then ran slurmdbd normally with systemctl. When I ran slurmctld it complained about incompatable state files. I didn't care about loosing state so ran with the -i option to ignore. Seemed happy. I then started up slurmd on my various GPU and non-GPU nodes. I ran a test interactive job just to see if I would get a shell with expected SLURM environment on the nodes and it worked. Except on the GPU nodes I got errors about not being able to get/set GPU frequences. I removed GpuFreqDef=medium from slurm.conf and that went away. I had my people start running jobs last night. This morning I come in and see a lot of nodes are stuck in completing state. All these are GPU nodes. For example on node 'sevilla' squeue reports JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON) 3679433 lcna100 sjob_123 sg1526 CG 8:29 1 sevilla 3679432 lcna100 sjob_123 sg1526 CG 8:35 1 sevilla 3679431 lcna100 sjob_123 sg1526 CG 8:42 1 sevilla In the slurmd.log on sevilla I see every minute the lines: [2024-01-28T09:37:52.003] debug2: Start processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug2: Processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug: _rpc_terminate_job: uid = 1150 JobId=3679432 [2024-01-28T09:37:52.004] debug2: Start processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug2: Processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug: _rpc_terminate_job: uid = 1150 JobId=3679431 [2024-01-28T09:37:52.004] debug2: Finish processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug2: Start processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug2: Processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug: _rpc_terminate_job: uid = 1150 JobId=3679433 [2024-01-28T09:37:52.004] debug2: Finish processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T09:37:52.004] debug2: Finish processing RPC: REQUEST_TERMINATE_JOB Back at the time 3679433 was cancelled by the user I see [2024-01-28T00:27:48.658] debug2: Start processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T00:27:48.658] debug2: Processing RPC: REQUEST_TERMINATE_JOB [2024-01-28T00:27:48.658] debug: _rpc_terminate_job: uid = 1150 JobId=3679433 [2024-01-28T00:27:48.658] debug: credential for job 3679433 revoked [2024-01-28T00:27:48.658] debug: _rpc_terminate_job: sent SUCCESS for 3679433, waiting for prolog to finish [2024-01-28T00:27:48.658] debug: Waiting for job 3679433's prolog to complete On sevilla itself there are no processes currently running for user sg1526 so no sign of these jobs. There ARE processes running for user mu40 though [root@sevilla ~]# pgrep -f slurm_script | xargs -n1 pstree -p | grep ^slurm slurm_script(30691)---starter-suid(30764)-+-python3(30784)-+-{python3}(30817) slurm_script(30861)---starter-suid(30873)-+-python3(30888)-+-{python3}(30915) slurm_script(30963)---starter-suid(30975)-+-python3(30992)-+-{python3}(31020) [root@sevilla ~]# strings /proc/30691/cmdline /bin/sh /var/slurm/spool/d/job3679220/slurm_script /usr/bin/sacct -p --jobs=3679220 --format=JobID,User,ReqTRES,NodeList,Start,End,Elapsed,CPUTime,State,ExitCode JobID|User|ReqTRES|NodeList|Start|End|Elapsed|CPUTime|State|ExitCode| 3679220|mu40|billing=6,cpu=3,gres/gpu=1,mem=32G,node=1|sevilla|2024-01-27T22:33:27|2024-01-27T23:26:39|00:53:12|02:39:36|NODE_FAIL|1:0| 3679220.batch|||sevilla|2024-01-27T22:33:27|2024-01-27T23:26:39|00:53:12|02:39:36|CANCELLED|| 3679220.extern|||sevilla|2024-01-27T22:33:27|2024-01-27T23:26:39|00:53:12|02:39:36|CANCELLED|| The slurmd.log has at the time it started lines like: [2024-01-27T22:33:28.679] [3679220.batch] debug2: gpu/nvml: _get_gpuutil: Couldn't find pid 30691, probably hasn't started yet or has already finished [2024-01-27T22:33:58.092] [3679220.extern] debug2: gpu/nvml: _get_gpuutil: Couldn't find pid 30650, probably hasn't started yet or has already finished [2024-01-27T22:33:58.166] [3679220.batch] debug2: profile signaling type Task [2024-01-27T22:33:58.168] [3679220.batch] debug2: gpu/nvml: _get_gpuutil: Couldn't find pid 30691, probably hasn't started yet or has already finished [2024-01-27T22:33:58.170] [3679220.batch] debug2: gpu/nvml: _get_gpuutil: Couldn't find pid 30764, probably hasn't started yet or has already finished [2024-01-27T22:33:58.172] [3679220.batch] debug2: gpu/nvml: _get_gpuutil: Couldn't find pid 30784, probably hasn't started yet or has already finished ... [2024-01-27T22:46:39.018] debug: _step_connect: connect() failed for /var/slurm/spool/d/sevilla_3679220.4294967292: Connection refused [2024-01-27T22:46:39.018] debug: Cleaned up stray socket /var/slurm/spool/d/sevilla_3679220.4294967292 [2024-01-27T22:46:39.018] debug: _step_connect: connect() failed for /var/slurm/spool/d/sevilla_3679220.4294967291: Connection refused [2024-01-27T22:46:39.018] debug: Cleaned up stray socket /var/slurm/spool/d/sevilla_3679220.4294967291 [2024-01-27T22:46:39.018] _handle_stray_script: Purging vestigial job script /var/slurm/spool/d/job3679220/slurm_script then later at the time of termination [2024-01-27T23:26:39.042] debug: _rpc_terminate_job: uid = 1150 JobId=3679220 [2024-01-27T23:26:39.042] debug: credential for job 3679220 revoked [2024-01-27T23:26:39.042] debug2: No steps in jobid 3679220 to send signal 998 [2024-01-27T23:26:39.043] debug2: No steps in jobid 3679220 to send signal 18 [2024-01-27T23:26:39.043] debug2: No steps in jobid 3679220 to send signal 15 [2024-01-27T23:26:39.044] debug2: set revoke expiration for jobid 3679220 to 1706416119 UTS [2024-01-27T23:26:39.044] debug: Waiting for job 3679220's prolog to complete [2024-01-27T23:26:39.044] debug: Finished wait for job 3679220's prolog to complete [2024-01-27T23:26:39.068] debug: completed epilog for jobid 3679220 [2024-01-27T23:26:39.071] debug: JobId=3679220: sent epilog complete msg: rc = 0 So I have a node stuck in 'comp' state that claims to have jobs completing that are definite NOT running on the box BUT there are jobs running on the box that SLURM thinks are done --------------------------------------------------------------- Paul Raines http://help.nmr.mgh.harvard.edu MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging 149 (2301) 13th Street Charlestown, MA 02129 USA Please note that this e-mail is not secure (encrypted). If you do not wish to continue communication over unencrypted e-mail, please notify the sender of this message immediately. Continuing to send or respond to e-mail after receiving this message means you understand and accept this risk and wish to continue to communicate over unencrypted e-mail.

4 7

2025

2024

slurm-users February 2024

2025

2024

slurm-users February 2024 ----- 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 February 2024