Slurm on CentOS 7 Docker Image

This is an all-in-one Slurm installation. This container runs the following processes:

• slurmd (The compute node daemon for Slurm)
• slurmctld (The central management daemon of Slurm)
• slurmdbd (Slurm database daemon)
• slurmrestd (Slurm REST API daemon)
• munged (Authentication service for creating and validating credentials)
• mariadb (MySQL compatible database)
• supervisord (A process control system)

It also has the following Python versions installed using pyenv:

• 3.6.15
• 3.7.12
• 3.8.12
• 3.9.9
• 3.10.0

Usage

There are multiple tags available. To use the latest available image, run:

docker pull giovtorres/docker-centos7-slurm:latest
docker run -it -h slurmctl --cap-add sys_admin giovtorres/docker-centos7-slurm:latest

The above command will drop you into a bash shell inside the container. Tini is responsible for init and supervisord is the process control system . To view the status of all the processes, run:

[root@slurmctl /]# supervisorctl status
munged                           RUNNING   pid 23, uptime 0:02:35
mysqld                           RUNNING   pid 24, uptime 0:02:35
slurmctld                        RUNNING   pid 25, uptime 0:02:35
slurmd                           RUNNING   pid 22, uptime 0:02:35
slurmdbd                         RUNNING   pid 26, uptime 0:02:35
slurmrestd                       RUNNING   pid 456, uptime 0:02:00

In slurm.conf, the ControlMachine hostname is set to slurmctl. Since this is an all-in-one installation, the hostname must match ControlMachine. Therefore, you must pass the -h slurmctl to docker at run time so that the hostnames match.

You can run the usual Slurm commands:

[root@slurmctl /]# sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
normal*      up 5-00:00:00      5   idle c[1-5]
debug        up 5-00:00:00      5   idle c[6-10]

[root@slurmctl /]# scontrol show partition normal
PartitionName=normal
   AllowGroups=ALL AllowAccounts=ALL AllowQos=ALL
   AllocNodes=ALL Default=YES QoS=N/A
   DefaultTime=5-00:00:00 DisableRootJobs=NO ExclusiveUser=NO GraceTime=0 Hidden=NO
   MaxNodes=1 MaxTime=5-00:00:00 MinNodes=1 LLN=NO MaxCPUsPerNode=UNLIMITED
   Nodes=c[1-5]
   PriorityJobFactor=50 PriorityTier=50 RootOnly=NO ReqResv=NO OverSubscribe=NO PreemptMode=OFF
   State=UP TotalCPUs=5 TotalNodes=5 SelectTypeParameters=NONE
   DefMemPerCPU=500 MaxMemPerNode=UNLIMITED

Building

Using Existing Tags

There are multiple versions of Slurm available, each with its own tag. To build a specific version of Slurm, checkout the tag that matches that version and build the Dockerfile:

git clone https://github.com/giovtorres/docker-centos7-slurm
git checkout <tag>
docker build -t docker-centos7-slurm .

Using Build Args

You can use docker's --build-arg option to customize the version of Slurm and the version(s) of Python at build time.

To specify the version of Slurm, assign a valid Slurm tag to the SLURM_TAG build argument:

docker build --build-arg SLURM_TAG="slurm-19-05-1-2" -t docker-centos7-slurm:19.05.1-2

To specify the version(s) of Python to include in the container, specify a space-delimited string of Python versions using the PYTHON_VERSIONS build argument:

docker build --build-arg PYTHON_VERSIONS="3.6 3.7" -t docker-centos7-slurm:py3

Using docker-compose

The included docker-compose file will run the cluster container in the background. The docker-compose file uses data volumes to store the slurm state between container runs. To start the cluster container, run:

docker-compose up -d

To execute commands in the container, use docker exec:

docker exec dockercentos7slurm_slurm_1 sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
normal*      up 5-00:00:00      5   idle c[1-5]
debug        up 5-00:00:00      5   idle c[6-10]

docker exec dockercentos7slurm_slurm_1 sbatch --wrap="sleep 10"
Submitted batch job 27

docker exec dockercentos7slurm_slurm_1 squeue
            JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
            27    normal     wrap     root  R       0:07      1 c1

To attach to the bash shell inside the running container, run:

docker attach dockercentos7slurm_slurm_1

Press Ctrl-p,Ctrl-q to detach from the container without killing the bash process and stopping the container.

To stop the cluster container, run:

docker-compose down

Testing Locally

Testinfra is used to build and run a Docker container test fixture. Run the tests with pytest:

pytest -v