Interactive SLURM Training Simulator
DGX Lab is a standalone, deterministic simulation environment for learning SLURM and shared GPU-computing workflows. It presents a generic DGX-scale cluster, simulated concurrent users, synthetic AI workloads, failure scenarios, guided labs, and local certification. It is deliberately incapable of connecting to a real scheduler.
Every module follows a consistent Learn → Practice → Assess rhythm. Learners receive one recommended action at a time, work entirely inside the simulated terminal, and advance only when the simulator observes the required evidence.
What to notice: Module 1 begins with one concrete next action. The lab path separates current work from later evidence, while the terminal and virtual cluster keep the command and its scheduler effect visible together.
What to notice: Practice removes the step-by-step script without removing orientation. Learners use a predict, observe, explain loop, edit a virtual batch file, and compare the resulting queue and GPU state with terminal evidence.
What to notice: Module 9 turns an out-of-memory event into a diagnosis-and-recovery exercise. Warning colors are reserved for the affected GPU and failed job, and the next action asks learners to inspect accounting before changing the workload.
What to notice: The same journey reflows for narrow screens. Controls remain touch-friendly, the recommended action stays prominent, and the capstone opens with an explicit evidence-backed objective rather than a compressed desktop layout.
This repository accompanies DGX Lab PRD v1.0 and provides:
- a Tauri 2 desktop shell and static GitHub Pages web distribution over one Leptos/Rust/WASM application;
- a responsive Learn, Practice, and Assess interface;
- a deterministic simulation core with constrained virtual scheduler, shell, and filesystem;
- state-backed practical grading and an offline knowledge assessment;
- generic DGX-H200-8, contended, and degraded scenario sources;
- twelve guided course modules, each with a validated lab definition and learner guide;
- a certification blueprint and question bank;
- schemas, content validators, CI/release workflows, ADRs, authoring guides, and security tests;
- the approved PRD, four design-direction mockups, and verified runtime screenshots.
| Area | Status in this pack |
|---|---|
| Product and architecture documentation | Substantial |
| Functional static prototype | Runnable reference implementation |
| Pure Rust domain model | Implemented and tested |
| FIFO/resource scheduler | Deterministic and tested |
| Virtual filesystem and shell | Constrained and tested |
| Synthetic workload planner | Implemented and tested |
| Practical grading and knowledge scoring | State-backed and tested |
| Scenario compiler and report renderer | Implemented |
| Leptos CSR UI | Responsive release build produced |
| GitHub Pages distribution | Path-aware build and deployment workflow implemented |
| WASM worker API | Native/WASM boundary verified |
| Tauri 2 shell | Minimal, no real-system commands |
| Native/WASM compilation evidence | Verified in the current workspace |
| Cargo lockfile | Resolved and committed |
| Signed installers | Deferred to a signing/notarization release lane |
The current release has native Rust test evidence, a verified WebAssembly build, strict linting, browser QA at desktop and phone widths, validated course content, 241 requirement links, and a checksum-verified course pack. The browser build runs entirely client-side and never executes learner commands on the host.
The canonical web edition is built from crates/web-ui and deployed from main by .github/workflows/pages.yml. GitHub Pages receives a generated static artifact rather than the checked-in dist/ reference snapshot. The workflow derives the repository project path, rebuilds the Leptos/WASM application, and rejects broken or root-hosted asset references before deployment.
Reproduce the project-site build locally with:
make web-pages PAGES_BASE=/DGX_Lab/A repository administrator must select GitHub Actions as the Pages source before the first production deployment. See docs/runbooks/GITHUB_PAGES.md for setup, trust boundaries, custom-domain migration, and rollback.
Serve the checked-in reference release build locally:
cd crates/web-ui/dist
python3 -m http.server 1421 --bind 127.0.0.1Then open http://127.0.0.1:1421. The release includes all twelve labs, free practice, visual cluster evidence, failure recovery, and the capstone-gated readiness assessment. It never executes entered commands.
The no-build implementation in prototype/ remains available as an early reference surface.
Install current stable Rust, the WASM target, Trunk, and Tauri CLI, then:
rustup target add wasm32-unknown-unknown
cargo install trunk --version 0.21.14 --locked
cargo install tauri-cli --version 2.11.4 --locked
python3 scripts/validate_all.py
cargo clippy --workspace --exclude web-ui --exclude sim-worker-wasm --exclude dgx-lab-desktop --all-targets -- -A clippy::manual_checked_ops -D warnings
cargo test --workspace --exclude web-ui --exclude sim-worker-wasm --exclude dgx-lab-desktop
cargo build -p sim-worker-wasm --target wasm32-unknown-unknown
cargo check -p web-ui --target wasm32-unknown-unknown
cargo tauri devSee docs/runbooks/LOCAL_DEVELOPMENT.md for OS prerequisites and the build order.
GitHub Pages static host or Tauri 2 desktop shell
│
▼
Leptos CSR interface (WASM)
│
▼
Rust simulation worker (WASM)
│
├── virtual scheduler
├── virtual users
├── virtual shell/filesystem
├── synthetic workloads
├── grading and assessment
└── deterministic event replay
NO SSH · NO SHELL · NO REAL SLURM · NO RUNTIME EXTERNAL NETWORK
The default cluster profile generalizes an eight-H200, 224-logical-CPU, cgroup-isolated Slurm environment into non-institutional names and paths. Production hostnames, IP addresses, credentials, and operational paths are intentionally absent.
crates/ Rust workspace crates
src-tauri/ Minimal Tauri 2 desktop shell
prototype/ No-build functional browser prototype
scenario-src/ Human-authored cluster/scenario YAML
course-src/ Course and lab Markdown/YAML
question-src/ Knowledge bank and certification blueprint
schemas/ JSON Schemas for portable content
scripts/ Validation and security tooling
docs/ Architecture, ADRs, authoring, runbooks, handoff
assets/mockups/ Approved high-fidelity UI directions
assets/screenshots/ Verified runtime screenshots used in this README
Produce signed and notarized desktop artifacts, add automated visual-regression coverage, and run structured usability sessions with learners while preserving the no-network, no-host-shell boundary. See docs/MILESTONES.md and docs/handoff/NEXT_ACTIONS.md.
- Code: Apache License 2.0
- Built-in original course content: CC BY 4.0
- Product name and marks: see
TRADEMARKS.md - Third-party dependencies: to be generated from the resolved
Cargo.lockbefore release