Slurmer

A job manager for Slurm, written in Go.

Installation

From sources

Requirements:

• Go >= 1.18
• GNU Make
• Slurm commands on the host that will run Slurmer and be connected to the cluster

First, you have to compile the code to get both slurmer and executor executables. A Makefile is available to do so:

make

There is few variables that you can override to specify custom options:

Variable name	Description	Default
MIN_UID, MAX_UID	the minimum and maximum uids that the executor can spoof	1000, 65565
SLURMER_UID	The user that is authorized to use the executor	1000
PREFIX	The path prefix to install slurmer to	/usr/local

Don't forget to make clean between variable changes to make sure that your changes have taken effect.

Example of custom variables

make MIN_UID=1100 MAX_UID=1200 SLURMER_UID=1099

Then you can install the executables to the destination directory (need root permissions):

make PREFIX=/opt/slurmer install

Now Slurmer should be installed on your system. Next we'll see how to use it.

Configuration

For this section, let's consider that slurmer is installed to prefix /usr/local/. Let's say that we have the configuration in /usr/local/etc/slurmer/config.yml (see sample configuration) and the Slurmer files have to be in /var/lib/slurmer. Make sure that only Slurmer account can edit config.yml.

First, create a directory /var/lib/slurmer and set the owner to the Slurmer user (we assume that slurmer is the Slurmer user):

mkdir /var/lib/slurmer
chown -R slurmer /var/lib/slurmer/ /usr/local/etc/slurmer/

# make sure that only slurmer can access and modify these files
chmod 0700 /var/lib/slurmer/ /usr/local/etc/slurmer/
chmod 0600 /usr/local/etc/slurmer/config.yml

Next, set the value working_dir in config.yml to /var/lib/slurmer.

Now you can run Slurmer with the following line to consider the given configuration (run with the Slurmer user):

slurmer -c /usr/local/etc/slurmer/config.yml

Sample configuration

A sample config file is also available in configs/config-example.yml.

# Useful when 'connector' is set to 'slurmrest'; not implemented yet
slurmrest:
  url: "http+unix:///default?socket=/tmp/slurmrestd.sock"

slurmer:
  ip: 127.0.0.1
  port: 8080
  connector: slurmcli
  working_dir: /var/lib/slurmer
  templates_dir: /etc/slurmer/templates
  executor_path: /usr/bin/executor
  applications:
    - name: app1
      token: averycomplicatedchallenge
      uuid: a9a5fc66-9bed-4a13-874a-d8d7d1756224 # a random app uuid
  logs:
    # available: text, json
    format: text

    stdout: false
    output: /var/logs/slurmer.log

    # available, from most to less verbose:
    # trace, debug, info, warning, error, fatal, panic
    level: debug

# To enable OIDC
oidc:
  enabled: false
  issuer: ""
  audience: ""

REST API summary

Each request must be authenticated with an application token and eventually a user for jobs related resources.

Header	Description
X-Auth-Token	Authenticates an application
User	The user that owns a job

Routes

Verb	Route	Description
GET	/apps	list all applications, need admin access token
POST	/apps	add a new application, need admin access token
GET	/apps/{appId}	show {appId} application, need admin access token
GET	/apps/{appId}/jobs	list all jobs of the app {appId}
POST	/apps/{appId}/jobs	add a new job to app {appId}
GET	/apps/{appId}/jobs/{jobId}	show job {jobId} details
DELETE	/apps/{appId}/jobs/{jobId}	delete a registered job
GET	/apps/{appId}/jobs/{jobId}/batch	read jobs batch file
GET	/apps/{appId}/jobs/{jobId}/out	read jobs standard output
GET	/apps/{appId}/jobs/{jobId}/files	download the job files as zip
POST	/apps/{appId}/jobs/{jobId}/files	upload a zip to job directory
PUT	/apps/{appId}/jobs/{jobId}/status	start or stop a job
PUT	/apps/{appId}/jobs/{jobId}/prune	clear all job data except batch file
PUT	/apps/{appId}/jobs/{jobId}/start	start a job
PUT	/apps/{appId}/jobs/{jobId}/stop	stop a job