Helm - Many bugs and regressions

[Crivaledaz]

Hi,

Describe the bug

I tried to install CodeTogether version 5 on premises, using the official Helm chart, but many changes in the chart YAML are required to get a running CodeTogether server.

Firstly, note the official documentation as a typo : The chart genuitec/codetogether does not exist, so it can not be pulled. The chart name is genuitec/CodeTogether.

Among the issues I encountered , the following are regressions from the initial chart I made on the PR #300 :

• Impossible to configure the imagePullSecrets parameters from the chart's values. The secret to pull the CodeTogether is hardcoded to the name codetogether-hub-login. It is not mentioned in the official install documentation (for Helm method). I think this secret name should not be hardcoded, to let the user choose a name that correspond to his name convention.
• No annotation can be added to the Ingress object. The Nginx ingress uses annotations for ingress/route configuration. In my environment, annotations are required for a production service.
• Image name cannot be specified. For some reasons, user could be using a mirror registry on-premises and should be able to point to his registry, not directly on the CodeTogether registry.

Following problems are bugs or security issues that need to be patched :

• Version 5 brings AV feature. You can disable this by turning to false the parameters av.enabled. However, AV ports (4443 and 10000) are still open and served by the CodeTogether Kubernetes service even if the user has chosen to turn it off. It is not the expected behaviour for me. I think port should be close and every reference to AV should be disabled, to reduce attack surface.
• The CodeTogether server pod is crashing in loop because of Readiness and Liveness probes. In my context, CodeTogether take more than 30 seconds to start and listen on port 1080, so all probes get a connection refused while CodeTogether is starting, and after 3 failed probes, Kubernetes restart the pod. Please adapt probes to include the startup time.

Expected behavior

The default workflow to install a Helm chart is the following :

1. Add the repo : helm repo add genuitec https://genuitec.github.io/CodeTogether/helm/
2. Update the local chart database : helm repo update
3. Get chart default values and adapt them helm show values genuitec/CodeTogether > values.yaml; vim values.yaml
4. Install the chart : helm install codetogether-server -n my-ns genuitec/CodeTogether -f values.yaml

Nothing more, or only pre-install task (create the Docker registry secret for example). Pull the chart is useless, unless you need to adapt it because it is not working ...

Environment

• Helm : 3.5.2
• Kubernetes : 1.18.8 / 1.22.1
• CodeTogether : 4.2 / 5.0

Workaround

For the moment the workaround is to modify chart sources to implement standard Helm use cases. I start from scratch with helm create codetogether and add CodeTogether changes without removing default Helm code block. However, this is time consuming and it's hard to keep updated with this workaround.

Please improve the chart to meet Helm standard or at least patch regressions.

Regards

[toddewilliams]

@Crivaledaz Thank you very much for this clear and detailed bug report. I've marked this bug as a "candidate" and the team will look at addressing each of these issues as quickly as possible.

[TimWebb]

@Crivaledaz Just a note that unfortunately it looks like that page is a bit out of date. You can get more updated instructions here: https://artifacthub.io/packages/helm/codetogether/codetogether?modal=template&template=codetogether-deployment.yaml

That said, we have a 5.0.1 release going out next week and I've assigned this item to ensure the Helm chart is reviewed for the above points. For instance in 5.0.1 we have some other changes for Podman support as well. Thanks again for bringing this to our attention and we'll make sure the documentation page is similarly updated.

[Crivaledaz]

My bad, I missed the link to Artifacthub page in the website documentation. In fact, documentation is more complete on Artifacthub. Maybe, you should redirect the user clicking on Helm install directly on the Artifactoryhub page.

Thank you for the fast answer, I am waiting for the next version to reinstall our CodeTogether preprod server with the official chart.

Regards

[GenuitecKanbot]

Happy to report this issue is resolved! (from Kanbanbot)