Skip to main content
Version: 1.11.0

Creating custom application Helm charts

Epinio deploys applications on Kubernetes as Helm charts.

By default, Epinio provides a standard Helm chart installation. However, operators may wish to create custom charts specific to their environment, and register them in Epinio, so their developers can use them.

Setting up a custom Helm chart

To add a custom Helm chart to the system, there are two mandatory steps:

  1. Create the chart,
  2. Make it known to the Epinio installation

The following two sections explain these steps in detail.

An advanced feature of custom application charts is the ability to expose user-settable configuration values to the user.

Creating the Helm chart

You can use the standard chart provided by Epinio as a template for the modifications.

You can download the chart from the Epinio repository into a directory <work-dir> of your choice, or, you can clone the Epinio Helm chart repository, again, into a <work-dir> of your choice.

git clone https://github.com/epinio/helm-charts.git

The chart is in the subdirectory helm-charts/chart/application.

In the following steps, the commands given, assume that the directory <work-dir> is the working directory.

You create a variant of the chart by adding an annotation to every application Deployment. The annotation enables the filtering of Epinio applications in fluentd.

Open the file helm-charts/chart/application/templates/deployment.yaml in your editor. This file is the template for the application's Deployment resource.

First, locate the section annotations:

[...]
template:
metadata:
annotations:
app.kubernetes.io/name: {{ .Values.epinio.appName }}
[...]

Add the following annotations:

fluentd-enable: "true"
fluentd-application-name: {{ .Values.epinio.appName }}

The result should look like:

[...]
template:
metadata:
annotations:
fluentd-enable: "true"
fluentd-application-name: {{ .Values.epinio.appName }}
app.kubernetes.io/name: {{ .Values.epinio.appName }}
[...]

The templating uses the .Values.epinio.appName field to insert the application name into the annotation.

An explanation of the values Epinio sets when deploying an application using the chart is in the Application Chart Reference. It's also available locally in the comments at the top of file helm-charts/chart/application/values.yaml.

Once you have modified the chart, use the following command to package the changed chart into a tarball:

helm package helm-charts/chart/application

The helm command places the tarball into <work-dir> and the file name should be application-VERSION.tgz where VERSION is the chart version.

Don't change the chart version. Versioned chart development is out of scope for this How-to.

Making the helm Chart known to Epinio

Once you have created the new chart, it's necessary to place the generated tarball on a web server. A few possible options are:

  • A server in the public cloud available to you
  • The company's host web server, if available, and permitted by company policies
  • A personal web server
  • A local web server, perhaps an nginx in a docker container

Given the range of possible options, this How-to assumes that the tarball is available at the example URL.

https://mydomain.org/epinio-application-fluentd.tgz

Copy the following YAML text, change the RELEASE_NAMESPACE value to the namespace where Epinio was installed, by default it's epinio, and add it to a file of your choice:

apiVersion: application.epinio.io/v1
kind: AppChart
metadata:
namespace: RELEASE_NAMESPACE
name: fluentd
spec:
shortDescription: Fluentd filterable standard deployment
description: Epinio standard support chart extended for fluentd filtering
helmChart: https://mydomain.org/epinio-application-fluentd.tgz

This How-to assumes that the name of the chosen file is fluentd-appchart.yaml.

info

This is a Kubernetes custom resource, which describes the new Helm chart to Epinio. In this example, it provides the information about the name, the source and descriptions.

The Epinio installation provides the necessary Kubernetes CRD.

Apply this resource to the Epinio cluster:

kubectl apply -f fluentd-appchart.yaml

Verify that the new chart is now registered in Epinio:

epinio app chart list

You can also see the details of the chart in Epinio:

epinio app chart show fluentd

User-settable configuration values

To expose the user-settable configuration value foo the created application chart has to look for this variable in the .Values.userConfig map. That is, it has to use the helm variable .Values.userConfig.foo.

caution

The chart should use a default value when this helm variable isn't set.

With the application chart exposing foo Epinio has to be aware of the configuration value by adding a specification to the AppChart resource describing it.

Do this by adding an entry foo to the spec.settings map of the AppChart resource. This entry is itself a map, with a mandatory type field, and type-dependent optional restrictions on the valid values of foo.

Example:

kind: AppChart
metadata:
[...]
spec:
[...]
settings:
foo:
type: integer
minimum: '42'
bar:
type: number
maximum: '122'
flag:
type: bool
mode:
type: string
enum:
- "sequential"
- "out of order"
- "randomized"

The valid types and their possible restrictions are:

TypeRestrictionSemantics
integerminimumMinimum valid integer value
maximumMaximum valid integer value
numberminimumfloating-point
maximumfloating-point
stringenumSequence of valid values
booln/a

Restrictions set on a type not supporting them, for example, a minimum on a string-type configuration value, are ignored.

caution

The restriction values have to be YAML strings, regardless the type of the configuration value. Note how in the example the maximum and minimum values are quoted to make them strings.

Troubleshooting

If the new Helm chart isn't correctly registered in Epinio, check that the RELEASE_NAMESPACE is set in fluentd-appchart.yaml.

Going further

The standard app chart, and the chart created here, use a Kubernetes Deployment as the main resource describing the active application.

Up to Epinio version 1.0, this was the only kind of resource supported. Epinio supports other kinds of controllers, for example, StatefulSet since versions greater than 1.0.

However, even when changing controllers, it's important to keep the Pod annotations and labels the same as for the standard chart. This lets Epinio's server locate the required resources and use them.

Here's a couple of examples:

  1. The value of epinio.io/stage-id is the id of the staging Pod which created the application's image.

  2. The value of epinio.io/app-container is the name of the application's main Container, used by epinio exec. This label has to match the actual name of the container in the pod spec.