Skip to content

Before you start

Two things have to be in place before Part 1, and one of them can take several days. Read this page first even if you plan to skip ahead.

1. A Jetstream2 allocation

Jetstream2 is a national cloud resource funded by the U.S. National Science Foundation and operated by Indiana University. Access is free to U.S.-based researchers and educators, but it is allocated rather than open — you request a share of it through ACCESS, the NSF's allocation system.

If you don't have one yet:

  1. Register for an ACCESS ID at access-ci.org if you don't already have one.
  2. Request an allocation. For a dashboard, an Explore allocation (the smallest tier) is normally plenty and is the quickest to be granted.
  3. Wait. Explore requests are typically reviewed within a week.

The Jetstream2 documentation walks through this in detail: Getting started → Allocations.

Already in a lab that uses Jetstream2?

Ask your PI to add you to their existing allocation instead — that takes minutes rather than days. You'll need to give them your ACCESS ID.

What an allocation actually buys you

Allocations are measured in service units (SUs), which are roughly CPU-cores × hours. A dashboard server is small and cheap to run:

Instance size vCPUs RAM SUs per hour Roughly
m3.small 2 6 GB 2 1,500 SU/month
m3.medium 8 30 GB 8 5,800 SU/month

An Explore allocation of 100,000 SUs runs an m3.small continuously for over five years' worth of hours, or an m3.medium for about 14 months. Storage volumes are charged separately and much more cheaply.

The practical consequence: shut down instances you aren't using, but don't agonise about leaving a dashboard running. That's what it's for.

2. A working dashboard

The tooling supports four frameworks. It works out which one you're using by reading your code, so you don't have to tell it.

  • R Shiny

    An .R file that calls shinyApp() — app.R or any other name — or a ui.R + server.R pair, or an R Markdown document with runtime: shiny in its front matter.

  • Plotly Dash

    An app.py that does import dash and exposes server = app.server.

  • Python Shiny

    An app.py with from shiny import App and a top-level app = App(...).

  • Streamlit

    A streamlit_app.py or app.py that does import streamlit.

"Working" means: you can start it on your own computer and use it in a browser without errors. If it doesn't work on your laptop, publishing it will not fix it — and debugging is far harder once it's inside a container on a remote server. Get it working locally first.

Using something else?

Flask, FastAPI, Panel, Voilà, Quarto dashboards and Observable are not supported by this tooling. You can still run them on a Jetstream2 instance, but you'd be doing the Docker and web-server work yourself — start from the Jetstream2 documentation instead of this guide.

What you do not need

You do not need any of the following, and this guide will not ask you to learn them:

  • Docker. The application builds and runs the container for you.
  • A web server. nginx is already installed and configured on the prepared image.
  • A domain name. Your dashboard gets a numeric IP address, which works fine. A domain name is optional and only needed for https:// — see Share and update your dashboard.
  • SSH keys, unless you specifically want to copy files with rsync. The web desktop needs only a browser.
  • A terminal. Everything in Part 3 is done by clicking.

A note on how long this takes

Be realistic about the first time through:
Getting an allocation Hours to a week (do this first)
Part 1 — server and volume 20 minutes
Part 2 — preparing your project 30 minutes for a simple app; an afternoon for an old or complicated one
Part 3 — publishing 15 minutes of clicking, plus the build
The build itself Streamlit/Dash/Python Shiny: 5–15 minutes. R Shiny: 20–45 minutes the first time, and longer for geospatial projects

Subsequent publishes of the same project are much faster, because the slow parts are cached.

Ready? → Part 1 · Set up your server