« back

Deploy Helm Charts on a Kubernetes cluster in Go

What is Helm #

The basic idea of Helm is to facilitate the deployment and manage complex application workloads in Kubernetes.

Helm is kinda like a package manager for Kubernetes, and the packages are so-called Helm Charts. A Helm Chart is essentially the combination of a template file, that describes the Kubernetes resources to deploy, and a values file used to valorize the former. The Helm Chart templating system enables the reusability of the packages.

The Helm Go SDK #

Helm comes with a handy CLI, a powerful command-line client for end-users, responsible for managing all the lifecycle phases of a Chart.
But, to programmatically install Helm Charts on a K8S cluster the CLI is not the solution if the intention is to build a robust and predictable application, luckily, the Helm developers designed the CLI as an interface for the Go SDK that I’ll reuse for my purpose.

Sync the Helm repository #

Helm supports Charts storage in remote repositories, so the first thing to do is to gather from the repository the information about the available charts.

I skip logging, error checking, and methods definitions to keep this blog post brief and concise.

Let’s start by downloading the repository index file.

import (
  "helm.sh/helm/v3/pkg/repo"
)

// repoEntry contains the name and the url of the repository
// settings contains the path to the directory where to store the repository index file
r, err := repo.NewChartRepository(repoEntry, getter.All(settings))

_, err = r.DownloadIndexFile()

Then I fetch the information about the charts stored in the repository and cache the results into a dedicated folder.

import (
  "helm.sh/helm/v3/pkg/repo"
)

var f repo.File
f.Update(repoEntry)
err := f.WriteFile(CacheFolderPath, 0644)

The before pieces of code replicate the Helm CLI commands:

Authenticate to the K8S API server #

The Helm Go SDK allows building an authenticated client instance by using the path of the kubeconfig file on the filesystem.

In the following code, the kube.GetConfig method returns an implementation of the RESTClientGetter interface, used to obtain a REST client configuration based on the provided kubeconfig path.
The REST client configuration contains everything is needed to dialog with the Kubernetes server API.

restClientGetter, err := kube.GetConfig(KUBECONFIG_PATH_ON_FS)
actionConfig := new(action.Configuration)

err = actionConfig.Init(restClientGetter, ...)

However, I can’t leverage that method because I don’t want to persist on the filesystem the Kubernetes credentials I got from the outside.

After having done some research and opened an issue on Github (I had to first search for similar issues, my bad), I finally ended up writing my implementation of the RESTClientGetter interface, restclient.go.

Then I build a RESTClientGetter by calling the NewRESTClientGetter method passing to it the kubeconfig content directly.

import (
  "helm.sh/helm/v3/pkg/action"
)


restClientGetter, err := NewRESTClientGetter(kubeconfigContent, namespace)
actionConfig := new(action.Configuration)

err = actionConfig.Init(restClientGetter, ...)

Due to an issue with the Helm SDK, I override the namespace value in my rest client builder. The install command, which you’ll see later, wrongly uses the default namespace, so I fixed it by forcing the namespace to the value passed in the NewRESTClientGetter method.

The install intent #

After that, I instantiate a new Install action and configure it by populating some of its attributes: the namespace where to create the resources and the release name.

import (
  "helm.sh/helm/v3/pkg/action"
)

client := action.NewInstall(actionConfig)

client.Namespace = namespace
client.ReleaseName = releaseName

Locate and load the chart #

The next step is to locate from the repository synced before the chart to deploy and load it.

import (
  "helm.sh/helm/v3/pkg/chart/loader"
)

cp, err := client.ChartPathOptions.LocateChart(chart, settings)
chartReq, err := loader.Load(cp)

Install the chart #

Once the chart is loaded, I can execute the deployment by calling the Run method and passing to it the requested chart.

err := client.Run(chartReq, nil)

All the above steps are required to imitate the Helm CLI command helm install name --namespace namespace chart.

Conclusions #

I hope I’ve given you an idea of ​​how to use the Helm Go SDK to install Helm Charts on a Kubernetes cluster.
More can be done, if the deployed Chart exposes a web service, you can use the K8S Go SDK to create a load balancer Service type to reach the application from the internet.

kubernetes, k8s, helm, golang, go, helm-chart