SLURM install guide

Debian

• install munge and libmunge2
• install slurm-wlm and slurm-wlm-basic-plugins (the package slurm is not the workload manager)
• create a basic slurm.conf file by opening /usr/share/doc/slurmctld/slurm-wlm-configurator.html in a web-browser
• save the created file to /etc/slurm-llnl/slurm.conf
• create the user slurm on the system
• change ownership of slurmctld to slurm: sudo chown slurm /var/lib/slurm-llnl/slurmctld
• make sure the log-file location in the slurm.conf file are writable by user slurm (eg. chown the current locations, or change the location to /home/slurm/whatever_the_file_name_is_not_this_literal_text for the files)
• start slurmctld with sudo /etc/init.d/slurmctld start
• start slurmd with sudo /etc/init.d/slurmd start
• now sinfo should give you a non-error, and display the column headers for SLURM jobs
• new jobs can be added to the queue with srun ..., squeue will show the job queue, scontrol suspend id will pause the job with the id specified, scancel id will cancel the job with the given id.

CentOS

• SLURM is not in the package manager, so download the latest stable version
• you must install munge and munge-devel before installing SLURM
• follow the configure make sudo make install installation guide
• same as for Debian, create a slurm.conf file, but place it: /usr/local/etc/slurm.conf
• create user slurm, and change ownership of /usr/local/sbin/slurmctld
• make sure the logfile can be created by user slurm
• problems were encountered with munge and the default location of SLURM, so the logfile location of munge was changed to syslog with the following:
• sudo systemctl edit --system --full munge
• the line was changed to: ExecStart=/usr/sbin/munged --syslog
• start munge with sudo systemctl enable munge then sudo systemctl start munge
• start sudo /usr/local/sbin/slurmctld start and sudo /usr/local/sbin/slurmd start
• now sinfo, srun etc. should work