HPC Dashboard

A drop-in replacement for the Open OnDemand dashboard that adds cluster-aware monitoring: allocation balances, storage quotas, live Slurm partition and node status, searchable job history with per-job efficiency metrics, and site announcements — on top of everything the stock dashboard already does.

Important

This is for HPC system administrators deploying a portal, not for end users. Installing it replaces the dashboard your users see — read Known Limitations first. Deploying as a sandbox app needs no root; replacing the system dashboard does.

Try it first in demo mode

Before thinking about deploying this dashboard to your system, you can try all features in an interactive demo.

No Slurm, no Open OnDemand, no root required — stub scheduler tools serve a consistent fake cluster (7 partitions, 82 nodes, 250 jobs, 4 allocations) so every page and widget is populated. All jobs data are invented.

We've built a website to host the demo at: https://tinyurl.com/ood-dashboard-demo

If you want to play with the demo and build your own demo, please follow the options below.

(Recommended) Option1: Container - Apptainer

Apptainer needs no root and is the path of least resistance on HPC.

apptainer build --ignore-fakeroot-command dashboard-demo.sif demo/apptainer.def
apptainer run --cleanenv dashboard-demo.sif
# open http://localhost:3000 or check with `demo/smoke.sh` (see below)

--cleanenv matters: without it the host's LD_PRELOAD leaks in and the container will not start if anything pollutes it (e.g. XALT).

(Recommended) Option2: Container - Docker

A demo/Dockerfile mirrors it for Docker sites.

docker build -f demo/Dockerfile -t ood-dashboard:demo .
docker run --rm -p 3000:3000 ood-dashboard:demo
# open http://localhost:3000 or check with `demo/smoke.sh` (see below)

Option3: Install all the required applications and gems

You can install all required applications and gems under your own space (root is not required), then try the demo using pre-installed packages with the commands below

git clone https://github.com/PurdueRCAC/OOD-Dashboard.git
cd OOD-Dashboard && bundle install && yarn install
OOD_DEMO_MODE=true bundle exec rails server -b 0.0.0.0 -p 3000
# open http://localhost:3000 or check with `demo/smoke.sh` (see below)

Needs Ruby 3.1–3.3, Node 16+ and Yarn 1.x on your PATH. Demo mode talks to no PUN, so there is no Ruby ABI to match — any version the app runs under is fine (verified on 2.7.8, 3.1.2 and 3.3.10). If you manage Ruby with rbenv, pin a version you actually have (rbenv local 3.3.8): rbenv local on a version that is not installed fails without writing anything, which silently stops the rest of the && chain.

or, on the Open OnDemand web node of a configured instance, you can just run

# Install all required packages, gems etc.
./install.sh
OOD_DEMO_MODE=true bundle exec rails server -b 0.0.0.0 -p 3000
# open http://localhost:3000 or check with `demo/smoke.sh` (see below)

Two things to know about this second path, because neither fails loudly:

• install.sh bundles for the PUN's Ruby, so run the demo on that same host. On a login node carrying a different Ruby, bundle exec reports bundler: command not found: rails — the gems are there, just under a different ABI directory.
• install.sh also creates .env.local from the example, and anything set there wins over the demo's own defaults. If your site config points the news feed at a real API, the Announcements widget comes back empty. Comment out the OOD_NEWS_* lines and restart — .env.local is only read at boot.

Checking it works

You can open the assigned URL with port (defaults to http://localhost:3000) in the browser or use demo/smoke.sh to check in terminal.

See more details about using this demo in DEMO

Features

Everything below is in addition to the stock OOD dashboard's app launcher, interactive sessions, Files, Projects, and Active Jobs.

Allocations and storage — balances, service units, quotas

• Balance and account widgets — per-allocation service-unit balances and usage, with CSV/XLSX export. Reads Slurm accounting directly via sacctmgr and scontrol show assoc, not OOD's balance JSON. Which TRES an allocation is denominated in is configurable.
• Storage widget — quota and file-count usage per filesystem, with warnings as users approach limits and deep links into the Files app. Driven by a site-local quota command you name, not OOD's quota JSON.

Cluster monitoring — partitions, nodes, reservations

• System Status widget — live Slurm partition state: node and core allocation, plus GPU allocation derived from AllocTRES/CfgTRES.
• Cluster Status page — every compute node with its state, load, and reservations, and a per-node detail page listing the jobs running on it.

Jobs and efficiency — history, detail pages, utilization metrics

• My Jobs — searchable, column-filterable job history over an arbitrary date range, with charts for job state and GPU-hour distribution, and CSV/XLSX export.
• Job detail pages — full scontrol/sacct detail for a single job, linked from interactive sessions and My Jobs, with a Cancel action.
• Per-job efficiency metrics — CPU utilization, GPU utilization, and memory efficiency, via jobstats (optional).
• GPU-hour accounting — used and reserved GPU hours per job, with configurable charged partitions, discounted QoS, and policy start date.
• Performance Metrics page — aggregate job efficiency trends for the user.

Communication — announcements, help, support tickets

• Announcements widget — outage, maintenance, and announcement articles from a site news API, colour-coded by type, with active outages sorted first.
• Dashboard Guide — an in-portal help page explaining each widget.
• Support ticket integration — email or Request Tracker (RT) backends, using OOD's standard support_ticket configuration.

Install it

Requires Open OnDemand (tested 2.0.32–4.2.2), the PUN's Ruby (its system or OOD-bundled Ruby — 2.7 to 3.3 across tested portals), Node 16+, and Slurm client commands reachable from the PUN host. Verified on four portals spanning OnDemand 2.0.32–4.2.2 (Rocky Linux 8–9, Ruby 2.7–3.3, Node 18.20.8) — see Known-good versions for the full list. Deploy as a sandbox app first — nothing your users see changes:

git clone https://github.com/PurdueRCAC/OOD-Dashboard.git \
  "$HOME/ondemand/dev/dashboard"
cd "$HOME/ondemand/dev/dashboard"
./install.sh

Run install.sh on the OOD web node — it detects and matches the PUN's Ruby there (off-host, pass PUN_RUBY_ABI; see Installation). Then launch it from Develop → My Sandbox Apps (Development).

Every site-specific value is configuration — you should not need to edit any view or controller. Any feature whose configuration is missing hides itself or reports N/A rather than erroring, so this boots and behaves like the stock dashboard before you configure anything. Two values are worth setting right away:

OOD_SITE_NAME="Your Cluster"
OOD_DASHBOARD_DOCS_URL="https://docs.example.edu/"

➡️ Quick Start — the full evaluation path, about an hour  ·  Installation  ·  Configuration

Documentation

Document	What's in it
Quick Start	Evaluation path for a Slurm site, touching nothing your users see
Installation	Requirements, deployment modes, build, verification checklist
Configuration	Every OOD_* key, grouped by feature, with defaults and formats
Troubleshooting	Symptom-first fixes for empty widgets, missing metrics, stale assets
Known Limitations	What this fork does not do, and what needs site-local work
Development	Working on the dashboard itself
Architecture	Which files implement each widget and page
Demo mode	Running the whole dashboard with an invented cluster behind it

Before you deploy

Four things worth knowing up front

• Slurm only. The monitoring features parse Slurm command output. Other schedulers keep app launching but lose those pages.
• You are adopting a fork. This replaces the OOD dashboard rather than installing alongside it, so upstream updates must be merged in, not applied by upgrading OOD.
• Two features need site-local work. The news feed assumes Purdue RCAC's API shape, and the Storage widget needs a quota command you supply.
• There is no automated test suite. Upstream's tests were not carried into this fork. Changes are verified manually against a sandbox deployment.

Full detail in Known Limitations.

Contributing

Fork, branch, and open a pull request describing your changes. Please keep site-specific values out of views and controllers — add a key to site_string_configs in config/configuration_singleton.rb and document it in .env.local.example instead. See Development.

For bugs or feature requests, open an issue.

This app is part of the OOD Appverse.

References

• Open OnDemand — the portal framework
• OSC/ondemand — upstream dashboard this is forked from (apps/dashboard)
• jobstats — per-job efficiency metrics
• Slurm_tools showpartitions — the partition status widget's Slurm queries are derived from this
• OOD app development guide

The balance and service-unit widgets originate in the Anvil OOD dashboard.

License

MIT License — © Ohio Supercomputer Center (upstream dashboard) and Purdue University Research Computing.

Built by Purdue University Research Computing (RCAC) on top of Open OnDemand, which is supported by the National Science Foundation under Grant No. 2005632.