Slurm

From DISI
Jump to navigation Jump to search

Slurm user-guide

Submit Jobs with Slurm

SBATCH-MR (beta)

It is a slurm-version of qsub-mr for submitting job on Slurm queueing system. Note: this is have not been extensively tested yet. Please contact me if the script is not working out. We are hoping to fully migrate to Slurm from the out-dated SGE system. Any error report would be helpful - Khanh

New slurm scripts are located in /nfs/soft/tools/utils/sbatch-slice

Just simply replace /nfs/soft/tools/utils/qsub-slice/qsub-mr with /nfs/soft/tools/utils/sbatch-slice/sbatch-mr in your script

To check the status of your job:

By username
 $ squeue -u <username>
By jobid
 $ squeue -j <job_id>

Submit load2d Jobs

$ cd <catalog_shortname>
$ source /nfs/exa/work/khtang/ZINC21_load2d/loadenv_zinc21.sh
(development) $ sh /nfs/exa/work/khtang/submit_scripts/sbatch_slice/batch_zinc21.slurm <catalog_shortname>.ism

Submit DOCK Jobs

  • ANACONDA Installation (Python 2.7)

Each user is welcome to download anaconda and install into his/her own folder
https://www.anaconda.com/distribution/
wget https://repo.anaconda.com/archive/Anaconda2-2019.10-Linux-x86_64.sh
NB: It is also available for Python3, which is our nearest future

simple installation via /bin/sh Anaconda2-2019.10-Linux-x86_64.sh

After the installation is completed, you need to install a few packages:

conda install -c free bsddb
conda install -c rdkit rdkit
conda install numpy


Running DOCK-3.7 with Slurm

Here is a “guinea pig project”, which has been done with DOCK-3.7 locally.
GPR40 example: /mnt/nfs/home/dudenko/TEST_DOCKING_PROJECT
ChEMBL ligands: /mnt/nfs/home/dudenko/CHEMBL4422_active_ligands
This test calculation should run smoothly, if not, then there is a problem.
Ultimately, you may need to compare your results with the reference run:

  • CHEMBL4422_active_ligands.mol2 - TOP500 scoring poses
  • extract_all.sort.uniq.txt - a print-out of scoring details


Slurm queue manager is installed locally at gimel, use it to run this test (and all your future jobs) in parallel.
Do not forget to set DOCKBASE variable: export DOCKBASE=/nfs/soft/dock/versions/dock37/DOCK-3.7.3rc1/

Useful DOCKING commands to remind:

$DOCKBASE/docking/setup/setup_db2_zinc15_file_number.py ./ CHEMBL4422_active_ligands_ CHEMBL4422_active_ligands.sdi 100 count
$DOCKBASE/analysis/extract_all.py -s -20
$DOCKBASE/analysis/getposes.py -l 500 -o CHEMBL4422_active_ligands.mol2


Useful slurm commands (see https://slurm.schedmd.com/quickstart.html):

to see what machine resources are offered by the cluster, do sinfo -lNe
to submit a DOCK-3.7 job, run $DOCKBASE/docking/submit/submit_slurm_array.csh
to see what is happening in the queue, run squeue
to see a detailed info for a specific job: scontrol show jobid=_JOBID_
to delete a job from queue, run scancel _JOBID_

Should your slurm run correctly, type squeue and you should see something like this:

            JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
  4187_[637-2091]     gimel array_jo  dudenko PD       0:00      1 (Resources)
         4187_629     gimel array_jo  dudenko  R       0:00      1 n-1-20
         4187_630     gimel array_jo  dudenko  R       0:00      1 n-5-34
         4187_631     gimel array_jo  dudenko  R       0:00      1 n-1-21
         4187_632     gimel array_jo  dudenko  R       0:00      1 n-5-34
         4187_633     gimel array_jo  dudenko  R       0:00      1 n-1-21
         4187_634     gimel array_jo  dudenko  R       0:00      1 n-5-34
         4187_635     gimel array_jo  dudenko  R       0:00      1 n-5-35
         4187_636     gimel array_jo  dudenko  R       0:00      1 n-5-34
         4187_622     gimel array_jo  dudenko  R       0:01      1 n-5-34
         4187_623     gimel array_jo  dudenko  R       0:01      1 n-5-34
         4187_624     gimel array_jo  dudenko  R       0:01      1 n-5-35
         4187_625     gimel array_jo  dudenko  R       0:01      1 n-1-17


As root at gimel, it is possible to modify a particular job, e.g., scontrol update jobid=635 TimeLimit=7-00:00:00

Slurm Installation Guide

Detailed step-by-step installation instruction (for sysadmins only)

Setup Node master

TBA

Setup Compute Nodes

Useful links:

https://slurm.schedmd.com/quickstart_admin.html
https://wiki.fysik.dtu.dk/niflheim/Slurm_installation

node n-1-17 (Installation of the latest slurm version (20.02.04)? see below in "Migrating to gimel5" section)

  • make sure you have there Centos 7: cat /etc/redhat-release
  • wget https://download.schedmd.com/slurm/slurm-17.02.11.tar.bz2
  • yum install readline-devel perl-ExtUtils-MakeMaker.noarch munge-devel pam-devel openssl-devel
  • export VER=17.02.11; rpmbuild -ta slurm-$VER.tar.bz2 --without mysql; mv /root/rpmbuild .

installing built packages from rpmbuild:

  • yum install rpmbuild/RPMS/x86_64/slurm-plugins-17.02.11-1.el7.x86_64.rpm
  • yum install rpmbuild/RPMS/x86_64/slurm-17.02.11-1.el7.x86_64.rpm
  • yum install rpmbuild/RPMS/x86_64/slurm-munge-17.02.11-1.el7.x86_64.rpm

setting up munge: copy over /etc/munge/munge.key from gimel and put locally to /etc/munge. The key should be identical allover the nodes.
Munge is a daemon responsible for secure data exchange between nodes.
Set permissions accordingly: chown munge:munge /etc/munge/munge.key; chmod 400 /etc/munge/munge.key

starting munge: systemctl enable munge; systemctl start munge

setting up slurm:

  • create a user slurm: adduser slurm.
  • all UID/GUIDs of slurm user should be identical allover the nodes.
 Otherwise, one needs to specify a mapping scheme for translating each UID/GUIDs between nodes.
To edit slurm UID/GUID, do vipw and replace "slurm line" with slurm:x:XXXXX:YYYYY::/nonexistent:/bin/false
XXXXX and YYYYY for slurm user can be found at gimel in /etc/passwd
NB: don't forget to edit /etc/group as well.
  • copy /etc/slurm/slurm.conf from gimel and put locally to /etc/slurm.
  • figure out what CPU/Memory resources you have at n-1-17 (see /proc/cpuinfo) and append the following line:
 NodeName=n-1-17 NodeAddr=10.20.1.17 CPUs=24 State=UNKNOWN
  • append n-1-17 to the partition: PartitionName=gimel Nodes=gimel,n-5-34,n-5-35,n-1-17 Default=YES MaxTime=INFINITE State=UP
  • create the following folders:
 mkdir -p /var/spool/slurm-llnl /var/run/slurm-llnl /var/log/slurm-llnl
 chown -R slurm:slurm /var/spool/slurm-llnl /var/run/slurm-llnl /var/log/slurm-llnl
  • restarting slurm master node at gimel (Centos 6): /etc/init.d/slurm restart
  • enabling and starting slurm computing nodes (Centos 7): systemctl enable slurmd; systemctl start slurmd

And last but not least, asking the firewall to allow communication between master node and computing node n-1-17:

  • firewall-cmd --permanent --zone=public --add-port=6817/tcp #slurmctld
  • firewall-cmd --permanent --zone=public --add-port=6818/tcp #slurmd
  • firewall-cmd --reload


To see the current situation of the queue, so sinfo -lNe and you will see:

Wed May 27 09:49:54 2020
NODELIST   NODES PARTITION       STATE CPUS    S:C:T MEMORY TMP_DISK WEIGHT AVAIL_FE REASON              
gimel          1    gimel*     drained   24    4:6:1      1        0      1   (null) none                
n-1-17         1    gimel*        idle   24   24:1:1      1        0      1   (null) none                
n-5-34         1    gimel*        idle   80   80:1:1      1        0      1   (null) none                
n-5-35         1    gimel*        idle   80   80:1:1      1        0      1   (null) none


To disable a specific node, do scontrol update NodeName=n-1-17 State=DRAIN Reason=DRAINED
To return back to service, do scontrol update NodeName=n-1-17 State=RESUME


p.s. Some users/scripts may require csh/tcsh.
sudo yum install csh tcsh

Node down after reboot On gimel (master node)

sudo scontrol update NodeName=<node_name> State=RESUME

On GPUs

sudo nvidia-smi -c 3 (to wake up the GPUs and set them in exclusive mode)
sudo scontrol update NodeName=<node_name> State=RESUME


Useful links:

https://support.ceci-hpc.be/doc/_contents/QuickStart/SubmittingJobs/SlurmTutorial.html


Migrating to gimel5

yum install rpm-build gcc python3 openssl openssl-devel pam-devel numactl numactl-devel hwloc hwloc-devel lua lua-devel readline-devel rrdtool-devel ncurses-devel gtk2-devel libibmad libibumad perl-Switch perl-ExtUtils-MakeMaker xorg-x11-xauth http-parser-devel json-c-devel mysql-devel libssh2-devel man2html munge munge-devel munge-libs -y

wget https://download.schedmd.com/slurm/slurm-22.05.5.tar.bz2

export VER=22.05.5; rpmbuild -ta slurm-$VER.tar.bz2 --with mysql --with slurmrestd

cd /root/rpmbuild/RPMS/x86_64

yum install slurm-$VER*rpm slurm-devel-$VER*rpm slurm-perlapi-$VER*rpm slurm-torque-$VER*rpm slurm-example-configs-$VER*rpm slurm-slurmd-$VER*rpm slurm-libpmi-$VER*rpm slurm-slurmrestd-$VER*rpm -y

systemctl enable slurmd
systemctl start slurmd
systemctl status slurmd

wget https://download.schedmd.com/slurm/slurm-20.02.7.tar.bz2

  • yum install rpm-build gcc openssl openssl-devel libssh2-devel pam-devel numactl numactl-devel hwloc hwloc-devel lua lua-devel readline-devel rrdtool-devel ncurses-devel gtk2-devel libssh2-devel libibmad libibumad perl-Switch perl-ExtUtils-MakeMaker mysql-devel munge-devel munge-libs python3 python3-devel -y
  • export VER=20.02.7; rpmbuild -ta slurm-$VER.tar.bz2 --with mysql; mv /root/rpmbuild .

Installation on a Compute Node

  • yum install rpmbuild/RPMS/x86_64/slurm-20.02.7-1.el7.x86_64.rpm -y
  • yum install rpmbuild/RPMS/x86_64/slurm-slurmd-20.02.7-1.el7.x86_64.rpm -y
  • systemctl enable slurmd
  • systemctl start slurmd

Installation for a Backup Controller

Currently (04/05/2021): gimel4

  • In gimel5's /etc/slurm/slurm.conf, find "BackupController="
    • Set the value to gimel4
    • Copy the conf file to gimel4
  • yum install rpmbuild/RPMS/x86_64/slurm-20.02.4-1.el7.x86_64.rpm
  • yum install rpmbuild/RPMS/x86_64/slurm-slurmd-20.02.4-1.el7.x86_64.rpm
  • yum install rpmbuild/RPMS/x86_64/slurm-slurmctld-20.02.4-1.el7.x86_64.rpm
  • systemctl enable slurmctld.service
  • systemctl start slurmctld.service

GPUs specification

    - 32-core:
               + n-9-34 (GTX 1080 Ti)
               + n-9-36 (GTX 1080 Ti)
               + n-1-126 (GTX 980)
               + n-1-141 (GTX 980)
   - 40-core:
               + n-1-28 (RTX 2080 Super)
               + n-1-38 (RTX 2080 Super)
               + n-1-101 (RTX 2080 Super)
               + n-1-105 (RTX 2080 Super)
               + n-1-124 (RTX 2080 Super)

Log Rotattion

If you haven't had logrotate install:

$ yum install logrotate

Here is a sample logrotate configuration. Make appropriate site modifications and save as /etc/logrotate.d/slurm on all nodes.

      ##
      # Slurm Logrotate Configuration
      ##
      /var/log/slurm-llnl/*log {
          compress
          missingok
          nocopytruncate
          nocreate
          nodelaycompress
          nomail
          notifempty
          noolddir
          rotate 5
          sharedscripts
          size=500M
          create 640 slurm root
          postrotate
           /etc/init.d/slurm reconfig
          endscript
      }

Slurm Admin Notes

Add user on gimel2

sudo sacctmgr add user jji account=bks

Troubleshooting

Zero Bytes were transmitted or received

Error

slurm_load_partitions: Zero Bytes were transmitted or received

This could mean that the clock on worker node is out-of-sync from master

 timedatectl set-time [HH:MM:ss]


Back to DOCK_3.7