Repositories in a Deployment

This page provides a reference guide to getting repos into a Gradient Deployments.

note

The best way to bring applications into a deployment is to include everything in a single Docker container. This improves repeatability and consistency between replicas and deployment specs.

If you don't want to package everything into a Docker container, you can clone any one or more Git repositories into the deployment spec. Supply the optional repositories configuration in your deployment spec:

• dataset: the dataset reference you want to use to sync the repositories into. This reference can be the dataset ID or dataset name.
• mountPath: the directory where you want to mount the repositories
• url: the URL for the Git repositories that you want to clone
• name: the name to give the repository directory in your container

Note that if you are using a private Git repository, you can configure authenticated access by also supplying the optional username and password parameters. The password field will need to reference a secret and should not contain the plaintext password itself.

Notes:

• GitHub no longer successfully authenticates with username and password and therefore will require username and access token. Details on how to create a GitHub access token can be found here.
• You must have already created a dataset to use this feature as the dataset reference is required to mount a Git repository. This is because, Gradient will clone your repositories into the specified dataset and mount that dataset to your deployment at the specified path.

For example, let's say that I want to clone two repositories, one that is public, and one that is private and requires a username and password, that be accomplish like below:

enabled: true
image: ubuntu:20.04
port: 8501
resources:
  replicas: 1
  instanceType: C4
repositories:
  dataset: dstflhstfexl6w3
  mountPath: /opt/repositories
  repositories:
    - url: https://github.com/tensorflow/serving.git
      name: tf-serving
    - url: https://github.com/paperspace/ml-private.git
      name: paperspace-ml
      username: paperspace-user
      password: secret:paperspace-github-token

Once this completes, I will have access to the TensorFlow Serving repository at /opt/repos/tf-serving, and to a repo of private ML code at /opt/repos/paperspace-ml.

Note: Certain paths do not work for repository mounting locations. Repositories must be mounted to absolute paths outside of root. Therefore mount paths like / or . will cause an error in the Deployment.

Container Registry integration

Private Docker images can be used by referencing your container registry which contains credentials to access that image.

To configure a Container Registry, follow this short guide.

Once configured, you can specify a containerRegistry value in your deployment spec with the name of the container registry you synced to your Workspace above.

enabled: true
image: paperspace/ml_server
containerRegistry: dockerhub-prod
port: 8501
models:
  - id: model-id
    path: /opt/models
env:
  - name: MODEL_BASE_PATH
    value: /opt/models
  - name: MODEL_NAME
    value: my-tf-model
resources:
  replicas: 1
  instanceType: C4