Slurm Installation Guide

From DISI
Jump to navigation Jump to search

This page will show you how to setup and configure a Slurm queueing system. Useful link: https://wiki.fysik.dtu.dk/Niflheim_system/Slurm_installation/

Pre-installation

Create global user account

Slurm and MUNGE users need to have a consistent UID/GID across all nodes in the cluster. Creating global user accounts must be done before installing the RPMs. It can be done via LDAPAdmin or any services that you use to manage users. If you don't have access to those services, please contact your system administrators.

Install the latest epel-release

[Centos/RHEL]

CentOS8: dnf install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm
CentOS7: yum install https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
RHEL7:   yum install https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm

[Ubuntu]

apt-get update

Install MUNGE

MUNGE is authentication service that Slurm uses validating users' credentials.

sudo yum install munge munge-libs munge-devel

(master node only) Create secret key

dd if=/dev/urandom bs=1 count=1024 > /etc/munge/munge.key
chown munge:munge /etc/munge/munge.key
chmod 400 /etc/munge/munge.key

For worker nodes, scp the munge.key from master node and set the correct ownership and permission

scp -p /etc/munge/munge.key hostXXX:/etc/munge/munge.key

Set ownership and permission to following directories

chown -R munge: /etc/munge/ /var/log/munge/ /var/lib/munge /run/munge
chmod 0700 /etc/munge/ /var/log/munge/

Start and enable MUNGE daemon at boot time

systemctl enable munge
systemctl start  munge

Increase number of MUNGE threads on master node (Optional by recommended on busy server)

cp /usr/lib/systemd/system/munge.service /etc/systemd/system/munge.service
vim /etc/systemd/system/munge.service
Edit this line >> ExecStart=/usr/sbin/munged --num-threads 10
Reload daemon and restart munge
 systemctl daemon-reload
 systemctl restart munge

Install Slurm

Although slurm is available on epel. It is better to build from RPMs to ensure we have the latest update.

This still shows you how to set up slurm with accounting (slurmdbd using MySQL as database). Accounting is optional and can be skipped, but it is useful for keeping records of job and managing resources.

Install prerequisite packages

$ 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

If you are seeing this error, this means httpd 2.2 is installed in the server

Error: httpd24u-filesystem conflicts with httpd-filesystem-2.4.35-5.el7.noarch
Error: httpd24u-tools conflicts with httpd-tools-2.4.35-5.el7.x86_64

Solution is to uninstall this version and run the yum command above again, it will install the correct package

yum remove httpd
yum remove httpd-tools
yum remove httpd-filesystem

If you are setting up slurmdbd, you will also need

yum install mariadb-server mariadb-devel

[Ubuntu] Download packages

apt-get update
apt-get install slurm slurmd slurm-client slurmrestd slurmdbd slurmcltd

[Centos/RHEL] Build RPMS

Check for the latest version in https://download.schedmd.com/slurm/

The current version at the time of this tutorial is 22.05.5

cd /root
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
     # Includes accounting support with the slurm-slurmdbd package
cd /root/rpmbuild/RPMS/x86_64

For Master Node

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

For Worker Nodes

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

Slurmctld

Slurm.conf
cd /etc/slurm
cp slurm.conf.example slurm.conf
Edit slurm.conf file, see gimel2's slurm.conf for example
If there is no slurm.conf.example, you can get one here https://github.com/SchedMD/slurm/blob/master/etc/slurm.conf.example
Other directories
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 /etc/slurm
Open Firewall
firewall-cmd --permanent --zone=public --add-port=6817/tcp #slurmctld
firewall-cmd --permanent --zone=public --add-port=6818/tcp #slurmd
firewall-cmd --reload
Start daemon
systemctl enable slurmctld
systemctl start slurmctld
systemctl enable slurmdbd
systemctl start slurmdbd 
Verify Setup
sinfo -lNe # list nodes and partitions

Slurmdbd

This will show you how to set up slurmdb's accounting storage in Mariadb. Slurmdbd doesn't need to be installed in the master node.

Make sure that mariadb packages are installed before built Slurm RPMs

rpm -q mariadb-server mariadb-devel
rpm -ql slurm-slurmdbd | grep accounting_storage_mysql.so     # Must show location of this file

Start daemon
systemctl start mariadb
systemctl enable mariadb
Configure db

1. Set up db's root password

/usr/bin/mysql_secure_installation

2. Create db

mysql -p
> grant all on slurm_acct_db.* TO 'slurm'@'localhost' identified by 'some_pass' with grant option;  ### WARNING: change the "some_pass" to the db password you just set.
> SHOW VARIABLES LIKE 'have_innodb';
> create database slurm_acct_db;
> quit;
To verify db grant for slurm user
mysql -p -u slurm
> show grants;
> quit;
Slurmdbd.conf
LogFile=/var/log/slurm/slurmdbd.log
DbdHost=XXXX    # Replace by the slurmdbd server hostname (for example, 
slurmdbd.my.domain)
DbdPort=6819    # The default value
SlurmUser=slurm
StorageHost=localhost
StoragePass=some_pass    # The above defined database password
StorageLoc=slurm_acct_db
.
.
LogFile=/var/log/slurm-llnl/slurmdbd.log
PidFile=/var/run/slurm-llnl/slurmdbd.pid
# Add these variables below so that the db doesn't get too big
PurgeEventAfter=3months
PurgeJobAfter=3months
PurgeResvAfter=3months
PurgeStepAfter=2months
PurgeSuspendAfter=1month
PurgeTXNAfter=3months
PurgeUsageAfter=3months

Change ownership and permission

chown slurm: /etc/slurm/slurmdbd.conf
chmod 600 /etc/slurm/slurmdbd.conf
systemctl restart slurmdbd
Firewall
firewall-cmd --permanent --zone=public --add-port=6819/tcp #slurmdbd
firewall-cmd --reload
Verify Setup
scontrol show config | grep AccountingStorageHost
sacct -a

Slurmd

Follow the Build RPMS above for Worker Nodes

Slurm.conf

You will need to edit the slurm.conf in master node and copy it to the worker node. Figure out what CPU/Memory resources you have at this worker node

cat /proc/cpuinfo
or
slurmd -C

append the following line:

NodeName=[Node_Name] NodeAddr=XX.XX.XX.XXX CPUs=[no_cpu] State=UNKNOWN
You can choose to add the new node into partition or having this line which will automatically add any nodes 
PartitionName=gimel2.cpu Nodes=ALL Default=YES MaxTime=INFINITE State=UP

Setup directories

mkdir -p /var/spool/slurm-llnl /var/run/slurm-llnl /var/log/slurm-llnl
chown -R slurm:slurm /etc/slurm /var/spool/slurm-llnl /var/run/slurm-llnl /var/log/slurm-llnl

Cgroup

cd /etc/slurm
cp cgroup.conf.example cgroup.conf

Firewall

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

Restart daemon

systemctl enable slurmd
systemctl start slurmd