Introduction
What is sind?
sind (Slurm-IN-Docker) creates and manages containerized Slurm clusters for development, testing, and CI/CD workflows. Each node runs as a separate Docker container with systemd as init, providing a realistic multi-node Slurm environment without requiring bare-metal infrastructure.
Inspired by kind (Kubernetes in Docker), sind offers a familiar CLI experience for quickly spinning up and tearing down complete Slurm clusters in seconds.
Why sind?
Setting up a multi-node Slurm cluster for development or testing traditionally requires dedicated machines, VMs, or complex provisioning. Tools like Vagrant and Docker Compose can get you there, but they demand significant configuration effort and maintenance. sind complements these approaches by trading some flexibility for ease of use and speed — a single command creates a fully working cluster.
- Instant clusters — go from zero to a working Slurm cluster in seconds, not hours
- Reproducible environments — every team member gets the same setup, every CI run starts clean
- No infrastructure needed — just Docker Engine on a Linux laptop, workstation or CI runner
Features
Multi-node, multi-cluster & multi-realm
Each cluster consists of individual containers for controller, database, REST API, submitter, and worker nodes. Run multiple clusters simultaneously with shared networking, and organize them into isolated realms for parallel CI jobs and multi-tenant testing.
System containers
Unlike typical Docker containers that run a single application process, sind nodes are full system containers running systemd as init — closely emulating bare-metal machines. Services like munge, sshd, and slurmctld start and interact exactly as they would on real nodes, with proper service dependencies, process supervision, and signal handling. This means you can apply the same configuration management tools you use on bare metal — Ansible, Chef, Puppet, Salt — directly to sind nodes. An unmanaged cluster leaves Slurm itself to them, for end-to-end tests of your Slurm provisioning.
Designed for CI/CD
sind runs on standard GitHub Actions runners with the runner’s Docker daemon — no privileged containers, no custom runner images. The sind-action GitHub Action installs sind and creates clusters in a single workflow step. Use realms to isolate parallel matrix jobs on the same runner.
Multiple Slurm versions
Official node images cover the supported Slurm release lines, 26.05 and 25.11, on linux/amd64 and linux/arm64, each with a full MPI stack. Pin a release line, test against several, or bring your own image.
Worker lifecycle
Dynamically add and remove worker nodes from running clusters. Test how your workloads react to nodes joining and leaving — without touching the controller.
Job accounting, users and identity
A db node runs MariaDB and slurmdbd, so job accounting, sacct and sacctmgr work as on a production cluster. sind also creates Linux users and groups, Slurm accounts and associations with their limits, and lets you choose where the nodes resolve the users: in local accounts on every node, through nss_slurm on the workers, or from auth/slurm tokens with sackd on the login node.
REST API
An api node runs slurmrestd, Slurm’s REST API daemon, with JWT authentication (Slurm 26.05), so portals, workflow engines and tests can drive the cluster over HTTP.
CVMFS
With storage.cvmfs, every node mounts /cvmfs read-only, from the Docker host or a CVMFS Docker volume plugin, and repositories mount on demand as on bare metal.
Monitoring
With monitoring: true, or sind create cluster --monitoring, a monitoring stack of Prometheus, Loki and Grafana collects the cluster’s Slurm metrics, its nodes’ resource use and their journals, with dashboards for clusters, nodes, services, users and logs (Slurm 26.05). The stack is shared by the realm’s clusters.
Web UI
sind ui serves a web UI on 127.0.0.1 that shows the realms, clusters, nodes, networks and the runs of sind commands, and updates live. It says what needs a look and which fix to try first, such as a restart of a service that failed. Its pages and dialogs create clusters, add workers, power nodes, restart services and delete, and show the exact sind command first, so the UI also teaches the CLI. sind ui --demo shows made-up clusters without Docker.
Connect outside containers
sind connect container attaches containers that sind did not create, such as the nodes of a kind cluster, to a realm’s mesh and every cluster network, and sind get coredns-config prints the CoreDNS server block that forwards the realm’s zone to sind’s DNS server.
Power cycle simulation
Shutdown, reboot, freeze, and power-cycle individual nodes to simulate real-world failure scenarios. Validate your fault-tolerance strategies before they matter.
Minimal dependencies
sind needs nothing but Docker Engine on a Linux host and a container image. Install a single binary and you’re ready to go. sind is also usable as a Go library for embedding cluster management into your own tooling.
AI-ready via MCP
sind includes a built-in Model Context Protocol (MCP) server that exposes its CLI commands as tools for AI assistants. Run sind mcp start to let tools like Claude, VS Code Copilot, or Cursor create clusters, check status, and manage nodes on your behalf. Register sind with your editor in one command — sind mcp claude enable, sind mcp vscode enable, or sind mcp cursor enable.
Next steps
Ready to try it? Head to the Getting Started section to install sind and create your first cluster.