Skip to content

Use a lent machine#

Someone can lend you some of their machines — for yourself, for one of your workspaces, or for your organisation — on their terms. Once you accept, the machines appear in your Machines list, marked lent by , and your work uses them like your own.

Accept an offer#

You are e-mailed the offer (and notified in the console when you have an account with that address, verified).

  1. Open Machines offered to you (the link in the e-mail, or /lending-offers). Each offer says the machines and the terms in words.
  2. For an offer to a workspace or an organisation, choose which one (you need the permission lendings:accept there — a workspace's or organisation's admins have it).
  3. Click Accept (or Decline: the lender is told).
$ astra astraeus lendings offers
0199c2f1-…  Ana's lab offers ["gpu-home-1"] (person) until 2026-10-22T12:00:00Z
$ astra astraeus lendings accept 0199c2f1-…
accepted: the machines are in your Machines list (lent by Ana's lab)

For a workspace: --workspace <org>/<workspace>; for an organisation: --org <org>.

$ curl -sS "https://api.astralyx.cloud/v1/me/lending-offers" -H "Authorization: Bearer $ASTRALYX_TOKEN"
$ curl -sS -X POST "https://api.astralyx.cloud/v1/me/lending-offers/<id>/accept" \
    -H "Authorization: Bearer $ASTRALYX_TOKEN" -H 'content-type: application/json' -d '{}'

On Astraeus Cloud, the workspace that borrows is connected to it when it is not yet. An offer to yourself goes to your personal workspace and covers your work only.

Use them#

Submit runs, deployments, environments, notebooks and functions as usual: the scheduler places them on lent machines within the terms, besides your own. To aim at a lent machine, name it (machine_selection.names) like any machine.

What the terms change:

Term What you notice
Only when the owner's GPUs are idle Your work there runs at the lowest priority: the owner's work stops it within seconds (5 s grace) and it goes back to the queue, its restarts unspent. Keep checkpoints on a drive.
Hours Outside them nothing of yours starts there, and what runs is stopped within seconds.
GPUs at most Your work there holds at most that many GPUs of each machine at once.
Sandboxed work only Your work runs under gVisor. GPU work cannot run sandboxed here: it waits for a machine that allows plain containers.
Until The lending ends then; your work there is stopped.

A run that starts all its workers together on several machines (a gang across machines) does not use lent machines yet: a lent machine is on its owner's network, apart from your machines and their names. Runs of one machine, arrays, deployments, environments and functions use them.

Work that waits says why, for each lent machine:

  • lent machine gpu-home-1: outside the lender's hours (22:00–07:00 America/Sao_Paulo)
  • lent machine gpu-home-1: only when idle: the owner is using it
  • lent machine gpu-home-1: its lender lends 1 GPU(s) at most
  • lent machines are not used by a run across several machines yet
  • lent machine gpu-home-1: the lender allows only sandboxed work on this machine, and GPU work cannot run sandboxed …

Work stopped by the lending says so too: Stopped on lent machine gpu-home-1: the lender took the machine back (or the lending expired, outside the lender's hours, the owner's work needs it).

On a lent machine you see the machine, its GPUs and its hardware's metrics; the owner's work there only as taken (its GPUs, cores and memory, never whose or what). You cannot change anything of the machine.

What you used#

Organisation → Lendings → the lending shows the usage report of your work there — GPU-, CPU- and memory-hours, network, energy, per person and kind of work — and the list of what ran. The lender sees the same numbers, and the same list (names, images, GPUs and their use), never your logs or data.

Give it back#

Give back on the lending's page (POST /v1/organizations/<org>/lendings/<id>/return, astra astraeus lendings return <id>) ends it: your work there is stopped, and the lender is told.

Before you put data there#

The machine's owner has physical and administrator access to it: they can read what is on its disks and in its memory. Do not put sensitive data, credentials you cannot rotate, or other people's personal data on a borrowed machine. See Lending: trust and privacy.