Command Line Interface

sd CLI reference.

The sd command line interface allows you to access Seldon Deploy programatically. This allows you to trigger deployments and promotions directly from your training or CI pipelines.

overview

Getting Started

To use it, you can run the pre-built seldonio/sd image. Note that you can use this image directly as a container on your CI pipeline.

docker run -it --rm seldonio/sd:0.2.3 sd

Authentication

The sd CLI supports authentication by providing your Seldon Deploy user and password on each call. This means that it can be used from outside the Kubernetes cluster, pointing the tool the external Seldon Deploy domain.

You can see an example below:

sd list \
  --server https://ml.example.org/seldon-deploy \
  --user jdoe \
  --password 232323 \
  --namespace staging

Dex

The sd CLI supports authentication when Dex is configured at the ingress gateway or app level.

At the moment, it will only work when Dex only has a single authentication provider configured. For example, if you configure access to an LDAP server in Dex, you must disable other sources of credentials like the static passwords added by default.

OIDC

The sd CLI supports authentication using an standard OIDC auth flow. At the moment, it will only work when using Keycloak as an OIDC provider.

To use it, you will need to specify the --sso flag as well as the following parameters:

  • --oidc-server: OIDC provider endpoint.
  • --oidc-client: OIDC client ID.

Note that these parameters are specific to your Seldon Deploy and OIDC provider configuration.

You can see an example below:

sd list \
  --server https://ml.example.org/seldon-deploy \
  --user jdoe \
  --password 232323 \
  --sso \
  --oidc-server https://ml.example.org/auth/realms/ml-realm \
  --oidc-client seldon-deploy \
  --namespace staging

Usage

Some of the most common workflows you can perform using sd are deploying a new version of your model and requesting a promotion between environments. These workflows can be automated as part of your CI pipeline, regardless of your CI provider.

Deploying a Seldon model

To deploy a new version of your model you can use the sd deploy command. This command will take as input the filename of your SeldonDeployment resource and will deploy it on a given namespace. If GitOps is enabled, this will also create (or update) the relevant files in your GitOps repository.

For example, to deploy a model named my-model (defined in the file my-model-seldondeployment.yaml) in the staging namespace you can run:

sd deploy my-model \
  --server https://ml.example.org/seldon-deploy \
  --user jdoe \
  --password 232323 \
  --namespace staging \
  --filename ./my-model-seldondeployment.yaml

You can see all the supported flags in the sd deploy reference page.

Helm

Note that this approach is compatible with using Helm to manage complex templates. This templates can encode any complex patterns relevant to your organisation and take as input an arbitrary number of variables.

usage-with-helm

As an starting point, you can consider the seldon-single-model Helm chart available in Seldon Core. Using this template, you can generate and deploy a SeldonDeployment resource from an image name simply as:

helm template my-model \
  seldonio/seldon-single-model \
  --namespace staging \
  --set 'model.image=seldonio/my-model:2.0' \
  > my-model-seldondeployment.yaml
sd deploy my-model \
  --server https://ml.example.org/seldon-deploy \
  --user jdoe \
  --password 232323 \
  --namespace staging \
  --filename ./my-model-seldondeployment.yaml

Note that these templates can be extended to fit any custom requirement (e.g. complex graphs, service accounts, HPA, etc.).

Requesting a promotion between environments

To request a promotion of your Seldon deployment between different environments you can use the sd promote command. This will fetch the latest version of your deployment from the source environment and will create a pull request with the relevant changes on the target environment.

Note that this workflow is only supported if GitOps is enabled.

For example, to request a promotion for a deployment named my-model from the staging environment to the production environment you can run:

sd promote my-model \
  --server https://ml.example.org/seldon-deploy \
  --user jdoe \
  --password 232323 \
  --git-user git-ci \
  --git-password 242424 \
  --from staging \
  --to production \
  --promotion-author 'Jane Doe' \
  --promotion-email jdoe@example.org

You can see more details in the sd promote reference page.

Reference

Last modified August 19, 2020: Rephrase (252f350)