One for the chemists among us - AiiDAlab on NS8 (testing)

App
,
1

Hi all,

I’ve built an NS8 module for AiiDAlab — the web platform for computational chemistry and materials science built on the AiiDA workflow engine.

It’s in my community repository. To try it, add the repo once:

api-cli run add-repository --data '{"name":"tebbiworld","url":"https://raw.githubusercontent.com/tebbiworld/ns8-repo/main/ns8/updates/","status":true,"testing":false}'

then install AiiDAlab from the Software Center. (Or straight from the image: add-module Package aiidalab · GitHub 1.)

What it does:

  • Runs the official aiidalab/full-stack image (Jupyter, the AiiDA daemon, PostgreSQL and RabbitMQ, all in one rootless container) behind Traefik with Let’s Encrypt and password login
  • AiiDA workflows with full provenance — every input, code and result stays traceable and reproducible
  • Jupyter apps such as the Quantum ESPRESSO app (aiidalab-qe), installable at start, with more from the App Store inside AiiDAlab
  • Remote computer: register an SSH-reachable machine — e.g. a GPU node on your Proxmox host or a cluster — as an AiiDA computer from the settings; calculations run there while AiiDAlab keeps the bookkeeping. The instance’s SSH public key is shown on the settings page and a connection test is one click.
  • NS8 backup of the whole workspace (AiiDA profile, PostgreSQL data, file repository, apps, notebooks, SSH key) plus a consistent pg_dumpall

A few things to know:

  • One instance is one workspace for one user — that’s how the AiiDAlab image is built. Several instances per cluster are fine.
  • The login password is required; it’s hashed and handed to Jupyter, and token login is off.
  • For the remote computer: AiiDA computers are immutable once used, so changing the host or work dir means a new label; and a target on the NethServer node itself must be entered as host.containers.internal, since a rootless container can’t reach the node’s own IP.
  • Reckon on roughly 1.5 GB RAM for the daemon stack (image is about 2.7 GB); the heavy computation happens on the remote computer, or on the node’s CPUs through the built-in localhost computer.

Still testing, so I’d genuinely welcome any feedback — especially from anyone wiring up a remote GPU/compute node over SSH.

Docs: NethServer wiki (tebbiworld repository) · Source: github.com/tebbiworld/ns8-aiidalab

Thanks!

Category: App · Tags: ns8