The Helm package manager is one of the most commonly used tools for managing Kubernetes manifests. Garden supports using your own Helm charts, alongside your container modules, via the kubernetes and local-kubernetes providers. This guide shows you how to configure and use 3rd-party (or otherwise external) Helm charts, as well as your own charts in your Garden project. We also go through how to set up tests, tasks and hot-reloading for your charts.

In this guide we'll be using the vote-helm project. If you prefer to just check out a complete example, the project itself is also a good resource.

You may also want to check out the full helm module reference.

Basics

First off, a couple of things to note on how the Helm support is implemented, with respect to Garden primitives:

1) One helm module maps to a single Garden service (not to be confused with Kubernetes Service resources), with the same name as the module. 2) Because a Helm chart only contains manifests and not actual code (i.e. your containers/images), you'll often need to make two Garden modules for a single deployed service, e.g. one container module for your image, and then the helm module that references it.

Referencing external charts

Using external charts, where the chart sources are not located in your own project, can be quite straightforward. At a minimum, you just need to point to the chart, and perhaps provide some values as inputs. Here is the redis module from our example project, for example:

kind: Module
description: Redis service for queueing votes before they are aggregated
type: helm
name: redis
chart: stable/redis
values:
  usePassword: false

For a simple setup, this may be all you need for a chart to be deployed with the rest of your stack. You can also list redis as a dependency of one of your other services, and this Helm chart is automatically deployed ahead of it, in dependency order.

You may also add a repo field, to reference a specific chart repository. This may be useful if you run your own chart repository for your organization, or are referencing a module that isn't contained in the default Helm repo.

Local charts

Instead of fetching the chart sources from another repository, you'll often want to include your chart sources in your Garden project. To do this, you can simply add a garden.yml in your chart directory (next to your Chart.yaml) and start by giving it a name:

kind: Module
description: Helm chart for my module
type: helm
name: my-module

You can also use Garden's external repository support, to reference chart sources in another repo:

kind: Module
description: Helm chart for my module
type: helm
name: my-module
repositoryUrl: https://github.com/my-org/my-helm-chart#v0.1

Tasks and tests

You may also want to define tests and/or tasks that execute in one of the containers defined in the chart. An example of this is how we define tasks in the vote-helm/postgres module:

kind: Module
description: Postgres database for storing voting results
type: helm
name: postgres
chart: stable/postgresql
version: 3.9.2    # the chart version to fetch
serviceResource:
  kind: StatefulSet
  name: postgres
tasks:
  - name: db-init
    args: [ psql, -w, -U, postgres, ..., -c, "'CREATE TABLE IF NOT EXISTS votes ..." ]
    env:
      PGPASSWORD: postgres
    dependencies:
      - postgres
  - name: db-clear
    args: [ psql, -w, -U, postgres, ..., -c, "'TRUNCATE votes'" ]
    env:
      PGPASSWORD: postgres
    dependencies:
      - postgres

Note first the serviceResource field. This tells Garden which Kubernetes Deployment, DaemonSet or StatefulSet to regard as the primary resource of the chart. In this case, it is simply the postgres application itself. When running the db-init and db-clear tasks, Garden will find the appropriate container spec in the chart based on the serviceResource spec, and then execute that container with the task's args and (optionally) the specified env variables.

The same applies to any tests that you specify. Take for example the vote module:

kind: Module
description: Helm chart for the voting UI
type: helm
name: vote
serviceResource:
  kind: Deployment
...
tests:
  - name: integ
    args: [npm, run, test:integ]
    dependencies:
      - api

Instead of the top-level serviceResource you can also add a resource field with the same schema to any individual task or test specification. This can be useful if you have different containers in the chart that you want to use for different scenarios.

Linking container modules and Helm modules

When your project also contains one or more container modules that build the images used by a helm module, you want to make sure the containers are built ahead of deploying the Helm chart, and that the correct image tag is used when deploying. The vote-helm/worker module and the corresponding worker-image module provide a simple example:

kind: Module
description: Helm chart for the worker container
type: helm
name: worker
...
build:
  dependencies: [worker-image]
values:
  image:
    tag: ${modules.worker-image.version}

kind: Module
description: The worker that collects votes and stores results in a postgres table
type: container
name: worker-image

Here the worker module specifies the image as a build dependency, and additionally injects the worker-image version into the Helm chart via the values field. Note that the shape of the chart's values.yml file will dictate how exactly you provide the image version/tag to the chart (this example is based on the default template generated by helm create), so be sure to consult the reference for the chart in question.

Notice that this can also work if you have multiple containers in a single chart. You just add them all as build dependencies, and the appropriate reference under values.

Hot reloading

When your project contains the container module referenced by a helm module, you can even use Garden's hot-reloading feature for a Helm chart. Going back to the vote module example:

kind: Module
description: Helm chart for the voting UI
type: helm
name: vote
serviceResource:
  kind: Deployment
  containerModule: vote-image       # The name of your container module.
  hotReloadArgs: [npm, run, serve]  # Arguments to override the default arguments in the resource's container.
...

For hot-reloading to work you must specify serviceResource.containerModule, so that Garden knows which module contains the sources to use for hot-reloading. You can then optionally add serviceResource.hotReloadArgs to, for example, start the container with automatic reloading or in development mode.

For the above example, you could then run garden deploy -w --hot-reload=vote or garden dev --hot-reload=vote to start the vote service in hot-reloading mode. When you then change the sources in the vote-image module, Garden syncs the changes to the running container from the Helm chart.

Re-using charts

Often you'll want to re-use the same Helm charts for multiple modules. For example, you might have a generic template for all your backend services that configures auto-scaling, secrets/keys, sidecars, routing and so forth, and you don't want to repeat those configurations all over the place.

You can achieve this by using the base field on the helm module type. Staying with our vote-helm example project, let's look at the base-chart and api modules:

# base-chart
kind: Module
description: Base Helm chart for services
type: helm
name: base-chart
serviceResource:
  kind: Deployment
skipDeploy: true

# api
kind: Module
description: The API backend for the voting UI
type: helm
name: api
base: base-chart
serviceResource:
  containerModule: api-image
dependencies:
  - redis
values:
  name: api
  image:
    repository: api-image
    tag: ${modules.api-image.version}
  ingress:
    enabled: true
    paths: [/]
    hosts: [api.local.app.garden]

Here, the base-chart module contains the actual Helm chart and templates. Note the skipDeploy flag, which we set because the module should only be used as a base chart in this case.

The api module only contains the garden.yml file, but configures the base chart using the values field, and also sets its own dependencies (those are not inherited) and specifies its serviceResource.containerModule.

In our base chart, we make certain values like name, image.repository and image.tag required (using the required helper function) in order to enforce correct usage. We recommend enforcing constraints like that, so that mistakes can be caught quickly.

The result module also uses the same base chart, but sets different values and metadata:

kind: Module
description: Helm chart for the results UI
type: helm
name: result
base: base-chart
serviceResource:
  containerModule: result-image
  hotReloadArgs: [nodemon, server.js]
dependencies:
  - db-init
values:
  name: result
  image:
    repository: result-image
    tag: ${modules.result-image.version}
  ingress:
    enabled: true
    paths: [/]
    hosts: [result-helm.local.app.garden]
tests:
  - name: integ
    args: [echo, ok]
    dependencies:
      - db-init

This pattern can be quite powerful, and can be used to share common templates across your organization. You could even have an organization-wide repository of base charts for different purposes, and link it in your project config with something like this:

kind: Project
sources:
  - name: base-charts
    repositoryUrl: https://github.com/my-org/helm-base-charts.git#v0.1.0
...

The base chart can also be any helm module (not just "base" charts specifically made for that purpose), so you have a lot of flexibility in how you organize your charts.