The Machines View, and whitewolf Joins

Abstract

orrery takes a fresh Ubuntu install to a fleet node from its window:

• probe it;
• apply its setup recipes, each behind a screen of details and a typed name, with root typed by Paul in the machine's Sessions tile;
• join it.

whitewolf (GTX 1080) was set up by the new one-run script and joined. This Mac is in the view by its hostname.

I. Introduction

A. Intent

The steps a new machine went through by hand become orrery's process, safe to repeat and hard to misapply.

B. The prompts

i did it. fixed. booted. i also created openssh server at ssh user@whitewolf . i already ssh-copy-id. please apply or help me apply the make cuda script and the orrery , and the cuda install, and the copal-tm with tmux . there is a script. andor add it to orrery. which gets it setup in a similear manner like columbia. This should also have a gpu. Please also add the motd with cpu and gpu snapshots.

these steps should be made into the process of orrery. like after the media. we need to spruce up the general media createion tool of copal and use orrery as the interface of media creation. in this. we applied an install, following the instructions of the Ubuntu 24.04 with OpenSSH server and user authentication. Which the Orrery then applies scripts and allows idempotent running of the config scripts. Make then difficult to press (with confirm with details), we dont wont to run the wrong scripts. And also, in general the Orrery, do we have the ability to schedule an arbitrary bash job to any of the computers. Would be nice to be able to universally apply, or apply by Apline / Ubuntu year.

whitewolf is back up, probe it and join it. also before that. launch the new version of orrery.

in the machines. it should implicitly have This Mac, or whatever it is by hostname.

Paul's answers:

• "Type it in a Sessions tile";
• "As the user; root only in a tile";
• "Machines page: probe, recipes, join".

II. Decision Log

III. Methods

1. orrery tools/fleet-node.sh:
   • build tools;
   • rustup for SUDO_USER;
   • treadcore over HTTPS;
   • cuda-node.sh or tmux and copal-tm;
   • fleet-login.sh --role;
   • fleet-poweroff-setup.sh;
   • the coordinator's two lines.
2. orrery tools/cuda-node.sh: modprobe, else a reboot note, before the check.
3. orrery tools/machines.py: the registry, PROBE and LOCAL_PROBE, RECIPES, applies and applied, stage with sums read back, and join.
4. orrery src/machines.rs: the view and its confirmation; sessions.rs run() (Ask::Window); nav.rs View::Machines; main.rs: the status thread, asks on threads, the hand-off, --frame-view machines.
5. Checks:
   • 340 tests;
   • review frames of the page and the confirmation;
   • every machine probed;
   • whitewolf's host key over its address compared with its key by

   name before trusting it.

IV. Results

Errors. whitewolf dropped off the network once during its setup run (the run had finished; it rebooted). Its two ports share one subnet, and traffic can leave by either.

V. Discussion

A new machine is now three presses in one window, each behind a confirmation that shows what it will do. The probe also turned up what the older machines never finished. The command line remains for anything the window does not do.