Skip to content

sandboxes.rst

Latest commit

 

History

History
57 lines (42 loc) · 1.97 KB

sandboxes.rst

File metadata and controls

57 lines (42 loc) · 1.97 KB
Raw

Sandboxes SDK tutorial

A sandbox is a lightweight, programmable execution environment for commands, files, and short-lived runtime state. Use sandboxes for isolated code execution, tool calls, temporary build steps, and workloads that do not need a full persistent Studio.

Sandboxes use a sandbox API key. For teamspace-scoped operations, create a teamspace API key in Lightning AI and provide it through the environment or the SDK client configuration. Do not commit API keys to source control.

Prerequisites

Install the package and configure a sandbox API key:

$ pip install lightning-sdk -U
$ export LIGHTNING_SANDBOX_API_KEY="replace-at-runtime"

The SDK exposes a client and a sandbox instance handle. Create the sandbox, write a file, run commands, and stop it from Python:

.. literalinclude:: ../../../examples/sandboxes.py
   :language: python
   :start-after: # sdk-sandbox-create-start
   :end-before: # sdk-sandbox-create-end
   :dedent: 8

Use the client object when you need to list, fetch, resume, or delete sandboxes from reusable application code:

.. literalinclude:: ../../../examples/sandboxes.py
   :language: python
   :start-after: # sdk-sandbox-client-start
   :end-before: # sdk-sandbox-client-end
   :dedent: 8

Run the companion script directly when you want to execute the SDK example:

$ python python/examples/sandboxes.py --teamspace owner/teamspace create
$ python python/examples/sandboxes.py --teamspace owner/teamspace inspect --sandbox-id sbx_1234567890

Operational notes

  • Sandbox.create returns a running SandboxInstance.
  • persistent=True lets stop capture an automatic snapshot that can be resumed later.
  • delete destroys the sandbox and discards auto-snapshot state.
  • run_command accepts either a command string plus args or a RunCommandOpts object for cwd, environment, detached execution, and other command settings.