Providers¶

NOTE:

The source code for the examples below can be found on GitHub.

Each workflow in the YAML file has to specify the provider property. This property defines the behaviour of the workflow and what parameters can be provided in the YAML file.

NOTE:

While each provider is unique, common functionalities such as Artifacts, Deps, Resources, Environment variables, Secrets, and Args are supported by most providers.

bash¶

The bash provider is the most common. It allows to run arbitrary bash commands.

workflows:
  - name: hello
    provider: bash
    commands:
      - echo "Hello, world!"

If you run it, you'll see real-time output:

$ dstack run hello

RUN      WORKFLOW  SUBMITTED  STATUS     TAG  BACKENDS
shady-1  hello     now        Submitted       local

Provisioning... It may take up to a minute. ✓

To interrupt, press Ctrl+C.

Hello, world!

$

NOTE:

The bash provider allows to use pip, conda, and python executables from the commands. See Python and Conda for more details.

Also, the bash provider allows to expose ports. See Apps for more details.

For more details, check the reference.

docker¶

The docker provider allows to run a user Docker image.

workflows:
  - name: hello-docker
    provider: docker
    image: ubuntu
    commands:
      - echo "Hello, world!"

If you run it, you'll see real-time output:

$ dstack run hello

RUN      WORKFLOW      SUBMITTED  STATUS     TAG  BACKENDS
shady-2  hello-docker  now        Submitted       local

Provisioning... It may take up to a minute. ✓

To interrupt, press Ctrl+C.

Hello, world!

$

For more details, check the reference.

code¶

The code provider runs a VS Code application.

workflows:
  - name: ide-code
    provider: code

If you run it, you'll see the URL:

$ dstack run ide-code

RUN         WORKFLOW  SUBMITTED  STATUS     TAG  BACKEND
lionfish-1  ide-code  now        Submitted       local

Provisioning... It may take up to a minute. ✓

To interrupt, press Ctrl+C.

Web UI available at http://127.0.0.1:51303/?tkn=f2de121b04054f1b85bb7c62b98f2de1

If you click the URL, it will open the VS Code application in the browser:

For more details, check the reference.

This is a great option if you'd like to run code interactively using a code editor and a terminal.

NOTE:

The code provider allows to use pip, conda, and python executables from the commands. See Python and Conda for more details.

lab¶

Similar to code, labs allows to run a JupyterLabs application.

For more details, check the reference.

notebook¶

Similar to code, labs allows to run a Jupyter notebook.

For more details, check the reference.