In my company we use fluxcd gitops operator to manage our kubernetes clusters. We also use flux official helm-operator, to deploy our helm charts. the helm-operator was originally designed for helm2 charts, but once helm V3 was out, fluxcd team did a great work on the helm-operator, and now it is is compatible with both helm2 and helm3 charts. In this post I will cover the steps I took to complete my charts migration.

What is Flux and how to get started

(skip if you are already a flux ninja )

Flux is a tool that automatically ensures that the state of your Kubernetes cluster matches the configuration you’ve supplied in Git. It uses an operator in the cluster to trigger deployments inside Kubernetes, which means that you don’t need a separate continuous delivery tool.

1. Fork flux repo and rename it. For example, my fork is https://github.com/Efrat19/local-cluster.

2. Follow the official installation steps. Those will help you get started with flux + the flux helm-operator, allowing you to gitopsly manage your workloads & helm charts.

3. test:

   ~ $ kubectl get po -n flux
   NAME                              READY   STATUS    RESTARTS   AGE
   flux-6b578c8cd-p696h              1/1     Running   0          4m32s
   flux-memcached-8647794c5f-gdw2p   1/1     Running   0          4m32s
   helm-operator-66d5477cb7-xwgzh    1/1     Running   0          4m1s
   

Why I couldn’t use helm 2to3 plugin

the helm team created a plugin which is supposed to help you migrate your released from helm2 to helm3. BUT if you are using fluxcd to manage your cluster, this tool is helpless because your changes will conflict with the chart source, and overwritten by the flux operator on the next sync cycle.

Upgrading Steps:

Step 1: switch to multiversioned helm-operator

Figure out in which namespace your flux helm-operator is running, and its version:

~ $ helm ls flux-helm-op --output yaml
Next: ""
Releases:
- AppVersion: 1.0.0-rc7
  Chart: helm-operator-0.5.0
  Name: flux-helm-op
  Namespace: flux
  Revision: 1
  Status: DEPLOYED
  Updated: Tue Jan 21 12:44:31 2020             Tue Jan 21 12:44:31 2020        DEPLOYED        helm-operator-0.5.0     1.0.0-rc7       flux            

as you can see, my helm operator is deployed into flux namespace. I should take care of 2 things:

1. The image version must be 1.0.0-rc5 or higher, otherwise I should bump it up.
2. I will add “HELM_VERSION=v2,v3” environment variable.

~ $ echo "
image:
  tag: "1.0.0-rc9"
extraEnvs: 
- name: HELM_VERSION
  value: "v2,v3"
" >> flux-helm3-values.yaml

 ~ $ helm upgrade helm-operator fluxcd/helm-operator --wait \
--namespace flux -f flux-helm3-values.yaml

As you can see in helm-operator source code, once the change is done, both v2 and v3 helm versions will be respected.

Step 2: Update the repo:

Bump the Chart API Version:

When using fluxcd, your cluster content is version-controller in a dedicated git repository, and you keep your charts under the /charts path. cd into charts/example. In the Chart.yaml file, change apiVersion from v1 to v2.

(Theres are a lot more differences between the versions, which you can view if you run helm create helm2-chart and helm3 create helm3-chart and compare the generated directories. In this post I only cover the minimal changes needed to get this done, for my super busy readers )

Bump the HelmRelease Helm version:

go to your helm release (/releases/example.yaml in the repo). Set the spec.helmVersion to v3. (the default value is v2).

Step 3: Make the swap

commit and push the changes. Meanwhile watch the helm-operator pod logs. you will see something like:

ts=2020-03-12T08:55:24.281650359Z caller=release.go:216
component=release release=example
targetNamespace=example resource=example:helmrelease/example
helmVersion=v3 error="Helm release failed"
revision=ccdf7cc0e4b57c624019d43fb2f066b8afd24ce9
err="failed to upgrade chart for release [example]:
rendered manifests contain a resource
that already exists. Unable to continue with install:
existing resource conflict:
kind: PersistentVolumeClaim, namespace: example, name: example-efs-pvc"

Don’t Panic Yet!

this error message tell us about resources conflicts. This is happening because By default, Flux operator does not delete resources. It is creating & updating resources but deletions must be done manually. So flux is now trying to deploy your helm3 new release resources but bumps into the helm2 old release pieces. So once you see those errors you know you can safely delete the helm2 release:

helm delete --purge example

Congrats!

You have cleared the way for the new helm3 release, and now the logs will show:

ts=2020-03-12T09:02:22.244932236Z caller=helm.go:69
component=helm version=v3 info="creating 7 resource(s)"
targetNamespace=example release=example
ts=2020-03-12T09:02:22.374214468Z caller=release.go:266
component=release release=example targetNamespace=example
resource=example:helmrelease/example helmVersion=v3
info="Helm release sync succeeded" revision=d058e851afb648eddd4d5ec29be0c163d16d0c85

From now on, your release can be viewed using helm3. (In this post I cover helm3 client installation steps):

~ $ helm3 ls
NAME            NAMESPACE       REVISION        UPDATED                                 STATUS          CHART           APP VERSION
example         example         1               2020-03-12 09:02:21.66010842 +0000 UTC  deployed        example-0.1.0   1.0        

Gently roll this update across your cluster. If you are careful and patient like me 🧘🏻‍♀️, You will not experience any downtime (The helm2 release pods will terminate and the helm3 release pods will start on the same time).

Happy Helm3ing!