How to package a Kubernetes application with Helm
Helm is the "The package manager for Kubernetes", it allows the management of the life-cycle of a Kubernetes application (deploy, configure, upgrade, retire, ...). It is an Infrastructure as Code (IaC) tool, so it allows us to version control an application and follow its evolution along time, make identical copies (prod, preprod, dev, ...) and predictable upgrades, and of course share and publish the application. It is one of the main tools used upstream, if a tool has a "Kubernetes deployment" it is most likely going to be using Helm.
Introduction
Helm packages applications into Charts. A Helm chart is a collection of YAML
templates. In order to create a chart, one must first install the Helm command line tool and install the oc command line tool. Once done, to continue:
Make sure you are logged in:
this should return your Rahti 2 username. Then create an example chart:
the result will be:
$ find example
example
example/Chart.yaml
example/templates
example/templates/tests
example/templates/tests/test-connection.yaml
example/templates/deployment.yaml
example/templates/service.yaml
example/templates/ingress.yaml
example/templates/hpa.yaml
example/templates/serviceaccount.yaml
example/templates/_helpers.tpl
example/templates/NOTES.txt
example/values.yaml
example/charts
example/.helmignore
It creates a mostly self-explanatory skeleton of a Chart. The structure is:
- The
Chart.yaml
file contains basic description values:name
,description
,version
, ... - The
values.yaml
file contains default values for the Chart and shows which parameters can be configured. - The
.helmignore
contains the patterns to ignore, it is similar togitignore
. WE will not change this file. - The
charts
folder contains the other charts that this one depends on. We will not use this feature. - Finally the
templates
folder contains the different API Kubernetes objects to be deployed. The templates engine syntax allows for a great deal of configurability. It supports Built-in Objects that for example show the current cluster capabilities, it supports external values files where each deployment of the application can have its own separate value file, it has an extensive list of template functions, flow control, and more.
Package a deployed application
Before we can start the process, we need to "clean up" the current Helm example chart.
-
Delete (or move anywhere else) all files inside the
templates
folder. -
Empty the
values.yaml
file. -
Edit
Chart.yaml
and fill up the values as needed.
helm lint
The helm tool provides a lint
command that will report any syntaxt issue with the current template.
Now we can create the YAML
files that contain the different parts of the application. As an example we will use a simple webserver with a volume attached to it. We will use an iterative process to create a copy of our current deployment. It is iterative because we will first create a simple, not configurable, and probably not working version, test it, come back and make it more complete and configurable, test it again, and so on.
-
List all the API Objects to get an idea of the different parts that it consists of:
$ oc get dc,deployment,pvc,secret,configmaps,service,route -o name deployment.apps/nginx persistentvolumeclaim/html secret/builder-dockercfg-h4cwh secret/builder-token-6fngh secret/builder-token-n2rdm secret/default-dockercfg-dqfxm secret/default-token-kfjlb secret/default-token-znxls secret/deployer-dockercfg-rpsxj secret/deployer-token-gnvzt secret/deployer-token-pvws5 service/glusterfs-dynamic-ed156002-8a7e-11ed-b60d-fa163e0d8841 service/nginx route.route.openshift.io/nginx
-
From the list above, we are only interested in
deployment.apps
,persistentvolumeclaim/html
,service/nginx
androute.route.openshift.io/nginx
. The rest are auto-generated like thesecret/
tokens or created by other objects like theservice/glusterfs-dynamic-ed156002-8a7e-11ed-b60d-fa163e0d8841
object was created as a result of the creation of thepersistentvolumeclaim/
(PVC) creation.
We will write templates one by one, starting with the Volume. There are two simple approaches to accomplish this task, "get and clean" or "recreate from template". We will first try the "get and clean" method.
Get and clean
The idea of get and clean is simple, we will retrieve a yaml
representation of an object running in the Kubernetes cluster and then delete all unnecessary information, like status and default configuration options.
Persistent Volume Claim
Get the PVC object in YAML format into the file pvc.yaml
:
Most of the information in the YAML
retrieved is status information generated by OpenShift and can be deleted:
@@ -1,18 +1,7 @@
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
- annotations:
- pv.kubernetes.io/bind-completed: "yes"
- pv.kubernetes.io/bound-by-controller: "yes"
- volume.beta.kubernetes.io/storage-provisioner: kubernetes.io/glusterfs
- creationTimestamp: "2023-01-02T09:22:06Z"
- finalizers:
- - kubernetes.io/pvc-protection
name: html
- namespace: test
- resourceVersion: "1771053847"
- selfLink: /api/v1/namespaces/test/persistentvolumeclaims/html
- uid: ed156002-8a7e-11ed-b60d-fa163e0d8841
spec:
accessModes:
- ReadWriteOnce
@@ -20,10 +9,4 @@
requests:
storage: 1Gi
storageClassName: glusterfs-storage
- volumeName: pvc-ed156002-8a7e-11ed-b60d-fa163e0d8841
-status:
- accessModes:
- - ReadWriteOnce
- capacity:
- storage: 1Gi
- phase: Bound
+status: {}
The main fields are metadata > name, spec > resources > requests > storage, and spec > storageClassName, and the result will be:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: html
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 1Gi
storageClassName: standard-csi
status: {}
Deployment
The same process can be repeated for deployment.apps/nginx
:
- First, for the
image
used, we need to replace the@sha256
hash by:latest
. This way we will always get the actual latest version of the image. It is also possible to replace it by a specific version like:1.23.3
. - Then we will delete the status information. An example of "status" entries are
lastTransitionTime
andcreationTimestamp
. This has no place to be in a template as it is information 100% generated by Kubernetes about the current running object, not the one we want to create. - Finally we will delete the auto-generated configuration options. An example of "configuration options" are the
rollingParams
. These configuration options are generated from the defaults of the Kubernetes cluster. It is also possible to keep these default options, so the user of the chart can change them before the creation is started, for example it could be necessary to increase thetimeoutSeconds
because the application takes more than 10 minutes to start.
@@ -1,44 +1,24 @@
apiVersion: apps/v1
kind: Deployment
metadata:
labels:
app: nginx
name: nginx
- namespace: test
- resourceVersion: "1771055913"
- selfLink: /apis/apps.openshift.io/v1/namespaces/test/deployments/nginx
- uid: a828c0db-8a7e-11ed-b60d-fa163e0d8841
spec:
replicas: 1
selector:
matchLabels:
app: nginx
strategy:
- activeDeadlineSeconds: 21600
- resources: {}
- rollingParams:
- intervalSeconds: 1
- maxSurge: 25%
- maxUnavailable: 25%
- timeoutSeconds: 600
- updatePeriodSeconds: 1
type: RollingUpdate
template:
metadata:
- annotations:
- openshift.io/generated-by: OpenShiftWebConsole
- creationTimestamp: null
labels:
app: nginx
spec:
containers:
- - image: bitnami/nginx@sha256:abe48bff022ec9c675612653292b2e685c91ce24bc4374199723c4f69603a127
- imagePullPolicy: Always
+ - image: docker.io/bitnami/nginx:latest
name: nginx
ports:
- containerPort: 8080
@@ -46,21 +26,12 @@
- containerPort: 8443
protocol: TCP
resources: {}
- terminationMessagePath: /dev/termination-log
- terminationMessagePolicy: File
volumeMounts:
- mountPath: /opt/bitnami/nginx/html/
name: html
- dnsPolicy: ClusterFirst
- restartPolicy: Always
- schedulerName: default-scheduler
- securityContext: {}
- terminationGracePeriodSeconds: 30
volumes:
- name: html
persistentVolumeClaim:
claimName: html
- test: false
@@ -71,29 +42,5 @@
- kind: ImageStreamTag
- name: nginx:latest
- namespace: test
- lastTriggeredImage: bitnami/nginx@sha256:abe48bff022ec9c675612653292b2e685c91ce24bc4374199723c4f69603a127
- type: ImageChange
The result being:
apiVersion: apps.openshift.io/v1
kind: Deployment
metadata:
labels:
app: nginx
name: nginx
spec:
replicas: 1
selector:
matchLabels:
app: nginx
strategy:
type: RollingUpdate
template:
metadata:
labels:
app: nginx
spec:
containers:
- image: docker.io/bitnami/nginx:latest
name: nginx
ports:
- containerPort: 8080
protocol: TCP
- containerPort: 8443
protocol: TCP
resources: {}
volumeMounts:
- mountPath: /opt/bitnami/nginx/html/
name: html
volumes:
- name: html
persistentVolumeClaim:
claimName: html
Recreate from template
For the two remaining objects: service
and route
, we will use the "recreate from template" method, where we start from a simple object and fill up the blanks with the configuration we need.
Route
This is a minimal route:
Where XXXX
is the name of the route, YYYY
is the host where the application will be configured to listen, and ZZZZ
is the Service connected to it.
Service
A minimal possible service is:
apiVersion: v1
kind: Service
metadata:
name: XXXX
spec:
selector:
app: YYYY
ports:
- nodePort: 0
port: NNNN
protocol: TCP
targetPort: NNNN
Where XXXX
is the name of the service we filled up in the Route, YYYY
is the name of the deployment, and NNNN
is the port the deployment is listening to.
Test
The tests should be done in a separate namespace, and there are two approaches:
-
Test them individually by:
The command above will create a PVC object in the selected namespace. You should check that it works as expected. You can destroy it by:
where
XXXX
is the name of the volume. -
Or all together in the Helm chart, by copying all the
yaml
files we created into thetemplates
folder.- We can then install it:
Note: With
--dry-run
you can preview what helm will deploy without making any change.- We can see the status of the installed chart by:
$ helm ls NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION test-name test 1 2023-01-03 14:59:04.026623633 +0200 EET deployed example-0.1.0 1.16.0
- After making a change in the chart templates we can upgrade it:
$ helm upgrade test-name example Release "test-name" has been upgraded. Happy Helming! NAME: test-name LAST DEPLOYED: Tue Jan 3 15:54:15 2023 NAMESPACE: test STATUS: deployed REVISION: 2 TEST SUITE: None
- And finally to destroy it:
Configuration
One of the powerful aspects of Helm is the possibility to, instead of using hardwired values, parametrize them, or to use provided built-in values. By removing hardwired values, we provide easy customization that will allow to use the template in more circumstances and for a longer period of time, for example by changing the template image. A user of the template just needs to worry about the values.yaml
file and not how these values fit into the complexities of the Chart. Helm uses Go templates to accomplish this.
The built-in values can be very handy, but we will mention only two of sets of them:
Release
variables offer the basic information of this chart deployment. Information like theNamespace
we are deploying this template to, or theName
of the Chart.Capabilities
is a more advance feature that provides information about what API objects and that version the Kubernetes cluster supports. For example the version of Kubernetes, or ifIngress
orRoute
are supported. These two pieces of information allow us to make more widely compatible templates, as different versions of Kubernetes will need slightly different options.
We can also define our own configuration variables. We will set up a default value and at the same time allow the values.yaml
file to override it. Let's start with the Volume yaml
file from previous steps:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: html
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 1Gi
storageClassName: glusterfs-storage
status: {}
In this template we would like to parametrize at least two of the values: the storage
size and the storage class
. One allows the user to use more or less disk space, and the second allows to change the driver used for the storage. The resulting file would be something like this:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: html
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: {{ .Values.storage.size | default "500Mi" }}
storageClassName: {{ .Values.storage.class | default "glusterfs-storage" }}
status: {}
And then a values.yaml
file like:
As you can see all variables in the values.yaml
file can be found from {{ .Values }}
. We have also set up a default of 500Mi
in the template itself using the default
function, and configured a value of 1Gi
at values.yaml
. In this example the storage of 1Gi
will be used, but if we removed the line size: 1Gi
, the storage will be the default of 500Mi
.
Other values that might be interesting to configure:
- The
host
which the application will be available at. - The
image
to use. - The number of
replicas
for the application.
Conditionals
It is possible to have conditionals in the templates, based on capabilities of the cluster, or in a configuration option. For example to activate tls
or not
apiVersion: route.openshift.io/v1
kind: Route
metadata:
name: nginx
spec:
host: {{ .Values.host }}
{{ if eq .Values.tls "active" }}
tls:
insecureEdgeTerminationPolicy: Redirect
termination: edge
{{ end }}
to:
kind: Service
weight: 100
name: nginx
status:
ingress: []
Which needs to have an option in values.yaml
like:
Final product
The final chart will be:
$ find
.
./Chart.yaml
./templates
./templates/deployment.yaml
./templates/service.yaml
./templates/pvc.yaml
./templates/route.yaml
./values.yaml
./charts
./.helmignore
Chart.yaml
:
apiVersion: v2
name: example
description: A Helm chart for Kubernetes
type: application
version: 0.1.0
appVersion: "1.16.0"
templates/deployment.yaml
:
apiVersion: apps/v1
kind: Deployment
metadata:
labels:
app: nginx
name: nginx
spec:
replicas: 1
selector:
matchLabels:
app: nginx
strategy:
type: Rolling
template:
metadata:
labels:
app: nginx
spec:
containers:
- image: docker.io/bitnami/nginx:latest
name: nginx
ports:
- containerPort: 8080
protocol: TCP
- containerPort: 8443
protocol: TCP
resources: {}
volumeMounts:
- mountPath: /opt/bitnami/nginx/html/
name: html
volumes:
- name: html
persistentVolumeClaim:
claimName: html
templates/service.yaml
:
apiVersion: v1
kind: Service
metadata:
name: nginx
spec:
selector:
app: nginx
ports:
- nodePort: 0
port: 8080
protocol: TCP
targetPort: 8080
templates/pvc.yaml
:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: html
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: {{ .Values.storage.size | default "500Mi" }}
storageClassName: {{ .Values.storage.class | default "glusterfs-storage" }}
status: {}
templates/route.yaml
:
apiVersion: route.openshift.io/v1
kind: Route
metadata:
name: nginx
spec:
host: {{ .Values.host }}
{{ if eq .Values.tls "active" }}
tls:
insecureEdgeTerminationPolicy: Redirect
termination: edge
{{ end }}
to:
kind: Service
weight: 100
name: nginx
status:
ingress: []
./values.yaml
:
image: bitnami/nginx:latest
host: example.2.rahtiapp.fi
tls: active
storage:
size: 1Gi
class: glusterfs-storage
Using online Helm Repo
Adding a repo
It's also possible to install repository
and then use them for deploying applications.
For example JupyterHub, you can install the repo
with this command:
If you type this command:
It also possible to check for updates (and update your repo
) with this command:
You can search and check the packages available:
helm search repo jupyterhub
NAME CHART VERSION APP VERSION DESCRIPTION
jupyterhub/jupyterhub 3.1.0 4.0.2 Multi-user Jupyter installation
Check values
When installing an online repo
, you can check the default values
used when you deploy it. For that, type the following command:
Edit and install from a repo
After exporting the default values
you can edit or create a config
file with your own values.
Then, you can install the Chart
using this command: