Troubleshooting and FAQ#

Step-by-step fixes for specific symptoms are in Troubleshooting. This page answers the questions people ask first.

Runtimes#

Why does my runtime stay `Pending`?#

No machine of the workspace can take it yet: the GPUs are held, no machine has the GPU model or memory asked, the pool has no machine, a machine has no data location for its drive, or the workspace's quota is full. Press Why? beside its state (in the notebook: Why is it waiting?): it lists each machine and what it lacks. Nothing you ask for is refused because no machine has it now; the runtime waits until one is free, or one that has it joins the workspace.

Why did my runtime stop?#

Its reason says. Most often Stopped after 30 minutes idle (60 for an environment): nobody used it for its idle time. Activity is what you do through it — running cells, output being saved while the notebook is open, an open SSH session, an open IDE window, requests to its apps. A cell that runs on with the tab closed, or prints nothing for longer than the idle time, is not activity. Raise the idle time (up to 1440 minutes), keep the tab open, or train in a run. See Idle stop.

What is kept when a runtime stops?#

Everything on its drive at /content: notebooks, your files, packages installed with %pip/pip install --user (onto the drive), an environment's home (/content/home/<user>: dotfiles, shell history, editors' servers, IDE settings and extensions) and SSH host key. Lost: the kernel's memory (variables: run the cells again), and anything written or installed elsewhere in the container (apt install by hand, files in /tmp or /root of a notebook's runtime). Packages a setup lists are installed again at each start from a cache on the drive.

I opened my environment on another machine and my files are gone#

They are on the first machine. An environment's home drive, like the notebooks drive, is kept on each machine's data location: one copy per machine, not synchronised. Pin the environment to one machine (or a pool of one) to always find it, or give it a shared drive as its home (--home-drive). See Drives and home.

Can I change an environment's GPUs or image?#

Not in place: a runtime's specification is fixed once made, except who it is shared with and its named apps. Stop and delete it, then make it again with the same name: its home drive <name>-home is still there, with your files. A notebook's runtime is changed with Change runtime…, which starts a new one.

Can another editor stop my environment?#

Yes. Any editor or admin of the workspace may stop, start or delete any runtime, as for runs. Only its owner, the people it is shared with and the workspace's admins may connect to it — SSH, its IDE, its apps.

People and access#

Who can see my environment?#

Everyone in the workspace sees it in Hesperus → Environments and Runtimes: its name, state, machine, resources and owner. Viewers reach nothing in any runtime. Editors and admins open any notebook of the workspace (Jupyter is every editor's). SSH, an IDE and apps are its owner's, its members' and the workspace's admins' only; others are refused (SSH_FORBIDDEN).

Someone left the team. Is their access gone?#

Take them off the environment's people: their next connection is refused; one already open runs until it closes. Removing them from the workspace refuses them everywhere at once. Their certificates are valid 8 hours but reach nothing without being allowed again at each connection.

Why is an environment shared with me missing from `astra ssh config`?#

astra ssh config writes hosts for environments you own. Connect to one shared with you with astra ssh <name>.

Your computer#

Does it work on a Mac?#

astra runs on macOS (Apple silicon), so astra ssh, VS Code, Cursor and JetBrains Gateway work from a Mac. A Mac as the machine running notebooks or environments does not yet: use a Linux machine.

Does it work on Windows?#

astra runs on Linux and macOS only, so astra ssh and astra ssh config do not run on Windows. From Windows, use the console's notebooks and an IDE environment's editor in the browser.

Which browsers and editors?#

Notebooks run in the console and the IDE in a tab of its own, in any current browser. Over SSH: VS Code and Cursor (Remote-SSH), JetBrains Gateway, and anything that uses OpenSSH's client (ssh, scp, rsync, sftp).

Cost and capacity#

What does a runtime cost?#

A runtime is a run on your machines: it waits for room, counts against the workspace's quota, and its hours count in usage like any worker, while it holds a machine (Pending, Starting, Ready). A stopped runtime holds nothing. See Usage, cost and budgets.

Can one person take every GPU?#

Workspace admins can cap it: environments a person may have running at once, and GPUs a person's runtimes may hold together (Set per-person limits). The workspace's quota caps the whole team.

No. A runtime gets whole GPUs, as a run does. CPU cores may burst beyond what it asks when the machine's cores are idle; memory is a hard limit.

Security and data#

What leaves my machines?#

The notebook's content and your files stay on the machine's drive: the console reads and saves the notebook through the runtime and Astralyx does not store it. Astralyx keeps a notebook's title, drive and path, a runtime's specification and state, and SSH sessions' who, from what client, when, how long and how many bytes — never what was typed. Jupyter's token is made on the machine and never leaves it; credentials are resolved on the machine. No port is opened on the machine: it connects out.

Is SSH into the container or the machine?#

The container, always. The SSH server is the machine's own build, mounted read-only into the container; it accepts only certificates for that one runtime, from its loopback. Nobody gets a login on the machine. See SSH access and certificates.

What can someone with a shell do?#

Everything a program in that container can: its files, its GPUs, the variables its credentials were given. That is why only its owner, its members and admins get one. Log masking applies to runs' logs, not to what a terminal shows.

What Hesperus does not do yet#

No astra commands for notebooks, for the workspace's limits per person, or for credentials of a runtime: use the console or the API. Credentials and gVisor isolation of a runtime are API-only.
An environment's resources, image and drives cannot be changed in place.
The console's notebook editor is the front end: Jupyter's own web interface (JupyterLab) is not opened.
The IDE's extensions come from Open VSX; Microsoft's Marketplace (and its Pylance, Remote and C# Dev Kit) is not available.
Machines running macOS do not run runtimes.