Skip to main content
A SLURM cluster is one compute in Expanse. A single daemon runs on the controller (the slurmctld host, usually your login node) and is the only process that talks to Expanse. Workers run a lightweight sampler with no credential and no outbound network access; samples travel over your shared filesystem. Start in the console’s Add compute panel (or with expanse compute register): it shows the short-lived install token plus your deployment’s control-plane and data-plane URLs and the pinned daemon version, which this guide’s commands take. The token is prompted for during install; it never goes on a command line. Running AWS ParallelCluster? Use the ParallelCluster guide instead; it removes the shared-prefix step.

Prefer to delegate? Copy this into an AI assistant with shell access on the login node; it fetches this guide itself, so it always follows the current steps.

​
Prerequisites

  • Console access in your Expanse organisation (any member can register).
  • Root (or sudo) on the controller node.
  • A shared filesystem mounted at the same path on every node, for example /srv/expanse on NFS or Lustre. Single-node clusters can use local disk.
  • Outbound HTTPS from the controller only. Workers need none.

​
1. Register the compute

In the console, open Compute → Add compute, pick SLURM, and generate an install token. Keep the panel open: it polls registration state live and shows the endpoint URLs used below.

​
2. Install the CLI on the controller node

​
3. Install the controller daemon

On the controller node, using the control-plane and data-plane URLs and the daemon version from the Add compute panel:
--daemon-version downloads the daemon release and verifies its signed manifest, checksum, and cosign signature against Expanse’s pinned release signing key before anything reaches the shared prefix. Air-gapped deployment? Copy the release archive and its .sig from your bundle mirror and pass --daemon-binary <path> instead. The install preflights everything (paths, SLURM config discovery, hook plan, permissions) before exchanging the token, so a failed preflight never burns it. It places the binary and Prolog/Epilog hooks under the shared prefix, installs the systemd unit, starts the daemon, and waits for the first heartbeat. On failure it rolls back its own changes. --patch-slurm-conf=true lets the install apply the hook wiring it planned; without it, all SLURM-owned config is read-only and the plan is printed for your config management to apply.

​
4. Install the sampler on each worker

No token needed; the sampler holds no credential:
Single-node clusters skip this step.

​
5. Verify

The compute flips to active in the console when the token is exchanged, and shows its first heartbeat within a minute of the daemon starting. Submit any job with sbatch as normal; capture is automatic.

​
Optional: Python stack attribution

The install never downloads third-party binaries or changes ptrace policy. If you make a pinned py-spy 0.4.2 available on PATH (or EXPANSE_PYSPY) on every node that runs jobs, with ptrace permitted, Python stack attribution switches on; otherwise it is recorded as unavailable and everything else works.