REST API Quick Start Guide

Slurm provides a REST API through the slurmrestd daemon, using JSON Web Tokens for authentication. This page provides a brief tutorial for setting up these components.

See also:

Contents

Prerequisites

The following development libraries are required at compile time in order for slurmrestd to be compiled:

Quick Start

This may be done on a dedicated REST API machine or your existing 'slurmctld' machine, depending on demand.

  1. Install components for slurmrestd
    • DEB: slurm-smd slurm-smd-slurmrestd
    • RPM: slurm slurm-slurmrestd (requires --with slurmrestd at build time)
  2. Set up JSON Web Tokens for authentication
  3. Ensure /etc/slurm/slurm.conf is present and correct for your cluster (see Quick Start Admin Guide and slurm.conf man page)
  4. Run slurmrestd (see below for systemd instructions) on your preferred [HOST]:PORT combination (':6820' is the default for production) 
    export SLURM_JWT=daemon
export SLURMRESTD_DEBUG=debug
slurmrestd <host>:<port>
    Adjust SLURMRESTD_DEBUG to the desired level of output (as described on the man page)

Running with systemd

Slurm ships with a slurmrestd service for systemd, however, it might require some additional setup to run properly. This section assumes you have either installed slurmrestd using DEB/RPM packages or built it manually such that the files are in the same places.
  1. Select a local user account to use to run the slurmrestd (not root or SlurmUser to prevent privilege escalation)
  2. Configure slurmrestd service to use this user and associated group. This can be accomplished in either of two ways:
    • Edit /etc/default/slurmrestd or /etc/sysconfig/slurmrestd
      Add -u USERNAME and -g GROUPNAME to SLURMRESTD_OPTIONS
    • Run systemctl edit slurmrestd to edit overrides for the service.
      Add content like this to the prescribed location in the overrides file: 
      [Service]
User=USERNAME
Group=GROUPNAME
  3. Configure the socket for slurmrestd. This may be accomplished by creating/changing permissions on the parent directory and/or changing the path to the socket in the service file.
    • Permissions: The user running the service must have write+execute permissions on directory that will contain the socket
    • Changing socket: Currently the way to change or disable the socket is to modify the 'ExecStart' line of the service (this is known to be an inelegant procedure and will be changed in the next major release)
      1. Run systemctl edit slurmrestd
      2. Add the following contents to the [Service] section: 
        ExecStart=
ExecStart=/usr/sbin/slurmrestd $SLURMRESTD_OPTIONS
Environment=SLURMRESTD_LISTEN=:6820
      3. Adjust the assignment of SLURMRESTD_LISTEN to contain the socket(s) you want the daemon to listen on.

      NOTE: Currently the way to change or disable the socket is to modify the 'ExecStart' line of the service. This is known to be an inelegant procedure and will be changed in the next major release (after 23.11).

Customizing slurmrestd.service

The next major release (after 23.11) changes the operation of the default service file and may break existing overrides. If you have overridden ExecStart= to contain any TCP/UNIX sockets directly, it will cause the service to fail if it duplicates any sockets contained in SLURMRESTD_LISTEN. These overrides will need to be changed after upgrading.

The default slurmrestd.service file has two intended ways of customizing its operation:

  1. Environment files: The service will read environment variables from two files: /etc/default/slurmrestd and /etc/sysconfig/slurmrestd. You may set any environment variables recognized by slurmrestd, but the following are particularly relevant:
    • SLURMRESTD_OPTIONS: CLI options to add to the slurmrestd command (see slurmrestd)
    • SLURMRESTD_LISTEN: Comma-delimited list of host:port pairs or unix sockets to listen on
      NOTE: if this duplicates what is already set in the 'ExecStart' line in the service file, it will fail
  2. Service editing: Systemd has a built in way to edit services by running systemctl edit slurmrestd.
    • This will create an override file in '/etc/systemd/' containing directives that will add to or replace directives in the default unit in '/lib/systemd/'.
    • Note that the override file must have the appropriate section declaration(s) for the directives you use (e.g., [Service]).
    • Changes may be reverted by running systemctl revert slurmrestd

Basic Usage

  1. Find the latest supported API version 
    slurmrestd -d list
  2. Get an authentication token for JWT 
    unset SLURM_JWT; export $(scontrol token)
    • This ensures an old token doesn't prevent a new one from being issued
    • By default, tokens will expire after 1800 seconds (30 minutes). Add lifespan=SECONDS to the 'scontrol' command to change this.
  3. Run a basic curl command to hit the API when listening on a TCP host:port 
    curl -s -o "/tmp/curl.log" -k -vvvv \
-H X-SLURM-USER-TOKEN:$SLURM_JWT \
-X GET 'http://<server>:<port>/slurm/v0.0.<api-version>/diag'
    • Replace the server, port, and api-version with the appropriate values.
    • Examine the output to ensure the response was 200 OK, and examine /tmp/curl.log for a valid JSON response.
    • Try other endpoints described in the API Reference . Change GET to the correct method for the endpoint.
  4. Alternate command to use the UNIX socket instead 
    curl -s -o "/tmp/curl.log" -k -vvvv \
-H X-SLURM-USER-TOKEN:$SLURM_JWT \
--unix-socket /path/to/slurmrestd.socket \
'http://<server>/slurm/v0.0.<api-version>/diag'
    • Replace the path, server, and api-version with the appropriate values.
    • Examine the output to ensure the response was 200 OK, and examine /tmp/curl.log for a valid JSON response.

Token management

This guide provides a simple overview using scontrol to obtain tokens. This is a basic introductory approach that in many cases should be disabled in favor of more sophisticated token management. Refer to the JWT page for more details.

Common Issues

In general, look out for these things:

  1. Validity of authentication token in SLURM_JWT
  2. Hostname and port number
  3. API version and endpoint
  4. Log output of slurmrestd

Unable to bind socket

This may be due to a permissions issue while attempting to set up the socket. Check the log output from slurmrestd for the path to the socket. Ensure that the user running the slurmrestd service has permissions to the parent directory of the configured socket path, or change/remove the socket path as described above.

If it says "Address already in use", check the command being run and the contents of "SLURMRESTD_LISTEN" for duplicates of the same TCP or UNIX socket.

Connection refused

Verify that slurmrestd is running and listening on the port you are attempting to connect to.

Protocol authentication error (HTTP 500)

One common authentication problem is an expired token. Request a new one: 

unset SLURM_JWT; export $(scontrol token)

This solution also applies to an HTTP 401 error caused by no authentication token being sent at all. This may appear in the slurmrestd logs as "Authentication does not apply to request."

Otherwise, consult the logs on the slurmctld and slurmdbd.

Unable to find requested URL (HTTP 404)

Check the API Reference page to ensure you're using a valid URL and the correct method for it. Pay attention to the path as there are different endpoints for slurm and slurmdbd.

Rejecting thread config token (HTTP 401)

Check that slurmrestd has loaded the auth/jwt plugin. You should see a debug message like this: 

slurmrestd: debug:  auth/jwt: init: JWT authentication plugin loaded
If it didn't load jwt, run this in the terminal you're using for slurmrestd: 
export SLURM_JWT=daemon

Unexpected URL character (HTTP 400)

Check the request URL and slurmrestd logs for characters that may be causing the URL to be parsed incorrectly. Use the appropriate URL encoding sequence in place of the problematic character (e.g., / = %2F). 

... -X GET "localhost:8080/slurmdb/v0.0.40/jobs?submit_time=02/28/24"
### 400 BAD REQUEST
... -X GET "localhost:8080/slurmdb/v0.0.40/jobs?submit_time=02%2F28%2F24"
### 200 OK

Other slurm commands not working

If SLURM_JWT is set, other slurm commands will attempt to use JWT authentication, causing failures. This can be fixed by clearing the variable: 

unset SLURM_JWT

Last modified 29 February 2024