Cluster Configuration

The controller writes the Slurm configuration. The workers download it from the controller at start (configless mode), so they have no configuration files of their own.

Files on the Controller

FileContent
/etc/slurm/slurm.confCluster, scheduler, plugins and partition
/etc/slurm/gres.confAutoDetect=nvidia, to find NVIDIA GPUs
/etc/slurm/cgroup.confCgroupAutomount=yes and ConstrainDevices=yes

Default slurm.conf

SettingValueMeaning
ClusterNameoneName of the Slurm Cluster
SlurmctldHostslurm-one-controllerHostname of the controller
AuthTypeauth/mungeMunge authentication
SchedulerTypesched/backfillBackfill scheduling
SelectTypeselect/cons_tresAllocates CPUs, memory and GPUs per job
ProctrackTypeproctrack/cgroupTracks job processes with cgroups
TaskPlugintask/cgroup,task/affinityLimits tasks with cgroups and pins them to CPUs
GresTypesgpuGPUs as generic resources
SlurmctldParametersenable_configlessWorkers download the configuration
MaxNodeCount100Maximum number of dynamic nodes
Nodesetone Feature=oneGroups all the workers
PartitionNameall Nodes=ALL Default=yesOne partition with all the nodes

With InfiniBand enabled, the controller also adds MpiDefault=pmix and PropagateResourceLimitsExcept=MEMLOCK. Refer to the InfiniBand Documentation.

Changes to the Configuration

To change the Cluster, use the service inputs when you create it. The Configuration Parameters page lists them all.

You wantHow
Shared usersLDAP inputs, see Identity Management
Shared /home and /scratchNFS inputs, see Shared Storage
MPI over InfiniBandInfiniBand inputs, see InfiniBand
More CPU, memory or GPUs per workerWorker VM template, see Worker Profiles and Accelerators

The appliance does not configure accounting (slurmdbd), QoS or more partitions.