Getting Started

The JobServer is the service that runs the jobs you submit from the SEAMM GUI or the web interface. seamm-manager install installs it and seamm-manager services start runs it as a service of the machine; you never start it by hand. seamm-manager services status shows it running beside the web interface:

seamm-manager services status

Out of the box it runs each job as a subprocess on its own machine, as many at once as it is allowed. One file turns it into a dispatcher that can run jobs under the machine’s own queueing system or send them to a cluster over ssh: <root>/<name>.ini, where <root> is the installation (~/SEAMM by default) and <name> the JobServer’s name, which defaults to the hostname. A minimal file that sends jobs to a SLURM cluster reached over passwordless ssh, with the job directories staged there and back:

[DEFAULT]
default = cluster

[cluster]
transport = ssh
host = cluster
export = NONE
remote_root = /scratch/me/seamm_jobs
remote_run_from_jobserver = /path/to/SEAMM/venv/bin/run_from_jobserver
account = myaccount
partition = normal
ntasks = 4
mem_per_cpu = 2G
time = 04:00:00
max_concurrent_jobs = 20

[cluster.limits]
overridable = ntasks, mem_per_cpu, time
ntasks.max = 64
time.max = 2-00:00:00

The JobServer reads the file when it starts, so restart it after editing:

seamm-manager services restart

The User Guide describes every key, several queues in one file, PBS, what happens to a job on a remote cluster, per-job resource overrides and where a job’s calculations run. The main SEAMM documentation has a step-by-step how-to, “Configure the JobServer’s Queues”, with real examples.