The Helm Grain

The HELM grain is Torque's native support for HELM v3 charts. Torque allows designers to use HELM specific features to easily orchestrate self-developer and community charts in a standard way and share them with others as building blocks. For a full blueprint yaml example, see Example 1: Helm Application with MySQL and S3 Deployed by Terraform.

Deploy in target-namespace

Torque will install the helm release in the namespace referred to in target-namespace. The target namespace must exist in the cluster prior to the deployment. It must not be equal to the namespaces used by Torque for agent deployments. Make sure the service account has enough permissions to create/read/delete everything in the helm chart and also create/read/delete secrets and volumes.

In your blueprint YAML, configure the following configuration for Helm grains

helloHelm:
  kind: helm
  spec:
    source:
      store: my-repo
      path: folder/my-app
    target-namespace: '{{ .inputs.namespace }}'  # This is the namespace which is passed along to the helm installation.
    agent:
      name: '{{.inputs.agent}}'
      service-account: '{{ .inputs.service_account }}'

It's also possible to use parameters for the service-account name. Note that when service account is not provided in the grain, the default service account provided on the agent will be used.

Tools and technologies

The following tools and technologies are installed out of the box on our agents in the Kubernetes and Helm runners and can be used when writing grain scripts (pre/post, etc.):

• dotnet
• curl
• tar
• unzip
• kubectl
• kustomize
• helm
• awscli

source

Please see the grain source for more details.

agent

Please see the grain agent for more details.

inputs

Similar to blueprint inputs and Terraform inputs, the HELM grain inputs allow you to reuse the same HELM chart in different ways using different values overrides. Inputs provided to the HELM grain are used when launching the HELM chart. We recommend using Torque's auto-discovery capability to quickly model your HELM chart within Torque including all defined inputs.

grains:
  nginx:
    kind: helm
    spec:
      source:
        store: my-repo 
        path: my-asset     
      agent:
      ...
      inputs:
        - replicaCount: '{{ .inputs.replicaCount }}'
        - service.image: '{{ .inputs.image }}'

info

Note that in the above example, blueprint input is used as the value of the HELM grain input, so the environment's owner is able to choose the replicaCount required for his need. The information provided by the user will be passed to HELM chart as values and affect the deployment process.

values.yaml as inputs to Helm grain

In Helm, the values.yaml file is a critical component that allows you to configure and customize the behavior of your Kubernetes application deployment.

The values.yaml file contains the default configuration values for the templates used in the Helm chart. These values can be overridden at the time of deployment or during an upgrade by providing a different set of values.

Torque supports referencing values.yaml files as inputs to the Helm grain, with the following syntax:

grains:
  nginx:
    kind: helm
    spec:
      source:
        store: my-repo 
        path: my-asset     
      agent: ...
      inputs:
        - replicaCount: '{{ .inputs.replicaCount }}'
        - service.image: '{{ .inputs.image }}'
      values-files:
      - source:
          store: <> 
          path: <>
      - source:
          store: <>
          path: <>

outputs

Helm does not natively expose outputs. However, this can be done using a post-helm-install script that assigns environment variables after the grain's deployment. The script must be referenced, along with the names of the environment variables in the grain's scripts. To expose the outputs to the environment, the names of the environment variables must be referenced in the blueprint's outputs section.

tip

For illustration purposes, here's a Helm chart that declares a URL environment variable and references the post-install script that publishes that environment variable as an output.

For example:

spec_version: 2
description: 
  ...

inputs:
  ...

outputs:
  output1:
    kind: regular
    value: '{{.grains.<grain name>.scripts.post-helm-install.outputs.<output name>}}'
  output2:
    kind: regular
    value: '{{.grains.<grain name>.scripts.post-helm-install.outputs.<output name>}}'

commands

The commands section allows to execute CLI code prior to the HELM chart deployment to make sure all dependencies are met to ensure a successful deployment. Common use for commands is to execute HELM dependencies update to collect all the sub-charts required for the deployment.

grains:
  nginx:
    kind: helm
    spec:
      source:
        store: my-repo 
        path: my-asset     
      agent:
        name: aws-demo
      inputs:
        - replicaCount: '{{ .inputs.replicaCount }}'
      commands:
        - dep up bitnami/nginx

scripts

Torque provides the ability to execute custom code after Torque executes the Helm chart. This is useful for generating and using outputs, as explained in helm outputs.

For example - grain with a post-install script and outputs "test1" and "test2":

grains:
  nginx:
    kind: helm
    spec:
      source:
        path: ...
      agent:
        name: ...
        service-account: ...
      inputs:
        - replicaCount: ...
      commands:
        - dep up bitnami/nginx
      scripts:
        post-helm-install:
          source:
            store: my-repo 
            path: my-asset     
          arguments: "{{.inputs.ACCOUNT_ID}},{{.inputs.ID2}},3"
          outputs:
            - test1
            - test2

Extracting outputs using scripts

This Blueprint is designed to deploy the MLflow platform, an open-source platform for managing the end-to-end machine learning lifecycle, using a Helm chart.

spec_version: 2
description: |
  MLflow provides an open-source platform to manage the end-to-end machine learning lifecycle. Key capabilities include experiment tracking, model management, model deployment, generative AI support, model evaluation, and integration with popular libraries. It enables data scientists to streamline ML development, maintain lineage, and operationalize models consistently across environments.

inputs:
  namespace:
    type: string
  admin-password: 
    type: string
    default: 'admin'
    sensitive: true
  agent:
    type: agent
    default: 'ml-agent'

outputs:
  URL: 
    value: 'http://{{.grains.mlflow.scripts.post-helm-install.outputs.url}}'
    kind: link
    quick: true
  credentials:
    value: 'user / {{ .inputs.admin-password }}'
    quick: true

grains:
  mlflow:
    kind: helm
    spec:
      source:
        store: ml-repo
        path: helm/mlflow
      target-namespace: '{{ .inputs.namespace }}'
      agent:
        name: '{{ .inputs.agent }}'
      inputs:
      - nameOverride: '{{ envid | downcase }}'
      - tracking.auth.password: '{{ .inputs.admin-password }}'
      scripts:
        post-helm-install:
          source:
            store: ml-repo
            path:  scripts/get-url.sh
          arguments: '{{ envid | downcase }} {{ .inputs.namespace }}'
          outputs:
            - url
      commands:
      - dep up helm/mlflow

get-url.sh

echo "Getting url service address"

export RELEASE_NAME=$1
export NAMESPACE=$2
export url=$(kubectl get service -n $NAMESPACE | grep $RELEASE_NAME | grep LoadBalancer | awk '{print $1}' | xargs kubectl get service -n $NAMESPACE --no-headers | awk '{print $4}')

echo url=$url

Let's break down the different sections of the Blueprint:

1. Inputs:

   • namespace: A string input that specifies the Kubernetes namespace where the MLflow deployment will be created.
   • admin-password: A sensitive string input that sets the password for the admin user in the MLflow deployment.The default value is 'admin'.
   • agent: An agent input that specifies the name of the Kubernetes agent to be used for the deployment. The default value is 'ml-agent'.

2. Outputs:

   • URL: A link output that provides the URL to access the deployed MLflow instance. It uses the post-helm-install script's output (url) to generate the URL.
   • credentials: A quick output that displays the credentials (user/password) for accessing the MLflow instance.

3. Grains:

   • mlflow: A Helm grain that defines the specifications for deploying the MLflow Helm chart.
     • source: Specifies the location of the Helm chart repository (ml-repo) and the path to the MLflow Helm chart (helm/mlflow).
     • target-namespace: The Kubernetes namespace where the deployment will be created, using the namespace input value.
     • agent: The name of the Kubernetes agent to be used, using the agent input value.
     • inputs: A list of input values to be passed to the Helm chart during installation.
       • nameOverride: Overrides the name of the Helm release with the environment ID (in lowercase).
       • tracking.auth.password: Sets the password for the admin user, using the admin-password input value.
     • scripts:
       • post-helm-install: A script that runs after the Helm chart installation.
         • source: Specifies the location of the script (ml-repo repository and scripts/get-url.sh path).
         • arguments: Passes the environment ID (in lowercase) and the namespace as arguments to the script.
         • outputs: Captures the url output from the script, which is used in the URL output.
     • commands:
       • dep up helm/mlflow: A command that runs the dep dependency management tool to update the dependencies for the MLflow Helm chart.

In summary, this Blueprint automates the deployment of the MLflow platform using a Helm chart. It allows you to specify the namespace, admin password, and the Kubernetes agent to be used. After the deployment, it provides the URL and credentials for accessing the MLflow instance. Additionally, it includes a post-helm-install script that retrieves the URL for the deployed MLflow instance and outputs it as a link.

command-arguments

The command-arguments field allows you to specify a set of option flags that will be passed to the helm upgrade command when deploying the Helm chart. This lets you customize the upgrade command in the same way you would when running it manually.

For example:

grains:
  nginx-simple:
    kind: helm
    spec:
      source:
        store: helm-charts-repo
        path: helm/nginx
      command-arguments: '--version 1.2.3 --verify --wait'

This will run helm upgrade with the --version 1.2.3 --verify --wait flags appended.

Here's a complete example showing usage of command-arguments along with other Helm grain features:

spec_version: 2

description: |
  Example Helm deployment showing:
  - Deploying from Tag  
  - Values files
  - Command arguments
  - Environment variables
  - Scripts 
  - Outputs

inputs:
  agent:
    type: agent
  release: 
    type: string
    default: "v1.0.0"

outputs:
  public_ip:
    value: 'https://{{.grains.nginx-simple.scripts.post-helm-install.outputs.public_ip}}'
  dns:
    value: '{{.grains.nginx-simple.scripts.post-helm-install.outputs.dns}}'
    kind: link

grains:
  nginx-simple:
    kind: helm
    spec:
      agent:
        name: '{{.inputs.agent}}'
      source:
        store: helm-charts-repo
        path: helm/nginx
        tag: '{{ .inputs.release }}'
      target-namespace: '{{ .inputs.namespace }}'  # This is the namespace which is passed along to the helm installation.
      commands:
        - dep up helm/nginx
      values-files:
        - source:
            store: application-config-repo
            path: helm/nginx/values.yaml
            tag: '{{ .inputs.release }}' 
      inputs: # will override values passed from the values.yaml
        - replicaSet: 1
        - applicationName: 'test'
      command-arguments: '--version 3.0.0'
      env-vars:
        - debug: true
      scripts:
        post-helm-install:
          source:
            store: scripts-repo
            path: scripts/post-install-script.sh
          outputs:
            - public_ip
            - dns

In this example, the command-arguments: '--version 3.0.0' line specifies that when deploying the nginx-simple Helm chart, the helm upgrade command should include the --version 3.0.0 flag.