slurm-centos7

Installation for the Slurm job scheduler on a CentOS 7 cluster.

Prerequisites

• btools
• SYS_UID_MIN in /etc/login.defs is < 980
  

Conventions

• The head node is identified as node01
• The compute nodes are identified as node02 and node03
  

Installation

1. On each node, run the installation script (this takes 20-30 minutes to complete)
   
   $ ./install-slurm
   
   

2. On the head node, create a pseudorandom secret key for MUNGE to use on all of the compute nodes
   
   $ ./create-munge-key
   
   

3. Copy the MUNGE secret key to all of the compute nodes
   
   $ bpush /etc/munge/munge.key /etc/munge/munge.key
   
   

4. Change the owner of /etc/munge/munge.key to the munge user on all of the nodes
   
   $ chown munge: /etc/munge/munge.key
$ bexec chown munge: /etc/munge/munge.key
   
   

5. Enable and start the MUNGE service on all of the nodes
   
   $ systemctl enable munge
$ systemctl start munge
$ bexec systemctl enable munge
$ bexec systemctl start munge
   
   

6. From any computer, complete the Slurm configuration file generator; edit the fields according to the values below (fields not addressed below should be left as their default value or empty if there is no default value)
   

   • ControlMachine: node01
   • NodeNames: node[02-03]
   • CPUs, Sockets, CoresPerSocket, and ThreadsPerCore: Values can be found by listing the CPU information on your machine with the lscpu command
     
   • StateSaveLocation: /var/spool/slurm
   • SlurmctldLogFile: /var/log/slurm/slurmctld.log
   • SlurmdLogFile: /var/log/slurm/slurmd.log
     
     

7. Click submit at the bottom of the page to generate the configuration file
   
   

8. Copy the configuration file to the head node and save the file to /etc/slurm/slurm.conf
   
   

9. Copy the configuration file to all of the compute nodes
   
   $ bpush /etc/slurm/slurm.conf /etc/slurm/slurm.conf
   
   

10. Move the cgroup configuration file to /etc/slurm/cgroup.conf (overwrite the existing file created with the install script)
    
    $ mv files/cgroup.conf /etc/slurm/cgroup.conf
    
    

11. Copy the cgroup configuration file to all of the compute nodes
    
    $ bpush /etc/slurm/cgroup.conf /etc/slurm/cgroup.conf
    
    

12. Disable and stop the firewalld service on all of the compute nodes
    
    $ bexec systemctl disable firewalld
$ bexec systemctl stop firewalld
    
    

13. On the head node, open port 6817 for Slurm
    
    $ firewall-cmd --permanent --zone=public --add-port=6817/tcp
$ firewall-cmd --reload
    
    

14. On the head node, enable and start the slurmctld service
    
    $ systemctl enable slurmctld
$ systemctl start slurmctld
    
    

15. On all of the compute nodes, enable and start the slurmd service
    
    $ systemctl enable slurmd
$ systemctl start slurmd
    
    

Testing

Check the Slurm configuration on the head node

$ slurmd -C

Check the Slurm configuration on all of the compute nodes

$ slurmd -C

Confirm that all of the compute nodes are reporting to the head node

$ scontrol show nodes

Run an interactive job

$ srun --pty bash

Links

• https://sysadm.mielnet.pl/building-and-installing-rpm-slurm-on-centos-7/
• https://github.com/mknoxnv/ubuntu-slurm